<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: 2pizza.team</title>
    <description>The latest articles on DEV Community by 2pizza.team (@2pizza).</description>
    <link>https://dev.to/2pizza</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4120075%2Fd7a96585-41f3-45fd-b656-4ee574f1587a.png</url>
      <title>DEV Community: 2pizza.team</title>
      <link>https://dev.to/2pizza</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/2pizza"/>
    <language>en</language>
    <item>
      <title>Why more new customers no longer fixes the problem</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Mon, 16 Nov 2026 12:11:20 +0000</pubDate>
      <link>https://dev.to/2pizza/why-more-new-customers-no-longer-fixes-the-problem-3ahn</link>
      <guid>https://dev.to/2pizza/why-more-new-customers-no-longer-fixes-the-problem-3ahn</guid>
      <description>&lt;p&gt;&lt;a href="https://ivanbolonikhin.com/blog" rel="noopener noreferrer"&gt;Retention from zero&lt;/a&gt; · Part 1 of 12&lt;/p&gt;

&lt;p&gt;Ivan Bolonikhin&lt;/p&gt;

&lt;p&gt;8 November 2026 · 4 min read&lt;/p&gt;

&lt;p&gt;Most owners I talk to can tell me what a new customer costs them. Almost none can tell me what a returning one costs. That gap is where a lot of money hides, and it takes one division to see it.&lt;/p&gt;

&lt;h2&gt;
  
  
  What does a returning customer cost you?
&lt;/h2&gt;

&lt;p&gt;Say you spend $200 in ads to win each new customer. Ten of them cost you $2,000. Three come back within three months for a second visit or a second order. So each regular you ended up with cost you about $667. The other seven paid once and were gone.&lt;/p&gt;

&lt;p&gt;Now say four of the ten come back. Same $2,000, and each regular costs $500.&lt;/p&gt;

&lt;p&gt;Nothing about the ads changed between those two stories. What moved the number was the work done on people who were already on your list: a reminder at the right moment, a call, a reason to book or reorder before they forgot you. That work costs something too, so count it. In the businesses I have worked with, it is small next to the ad bill.&lt;/p&gt;

&lt;p&gt;I use the word regular for anyone who came back at least once. It is the group that pays for your rent, and most owners never count it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why does buying more new customers stop working?
&lt;/h2&gt;

&lt;p&gt;Because you are filling a bucket that has holes, and you are paying full price for every cup of water.&lt;/p&gt;

&lt;p&gt;Stay with the example. Say you spend $4,000 a month and get 20 new customers at $200 each. If three in ten come back, you end up with 6 regulars a month. If four in ten come back, you end up with 8. Two extra regulars from the same budget.&lt;/p&gt;

&lt;p&gt;To get those same 8 regulars with the old return rate, you would need about 27 new customers. That is roughly $5,300 a month in ads instead of $4,000, so about $1,300 more every month to make up for people who left without anyone asking them to stay. These are example numbers. Yours will differ, but the shape holds: a small change in how many come back is worth a large change in ad spend.&lt;/p&gt;

&lt;p&gt;There is a second reason. When ads get more expensive, and many owners tell me theirs have over the years, the bucket gets more costly to fill. You can see this in your own books by dividing your ad spend by the new customers it brought, year by year. If that number is climbing, every customer who leaves quietly costs you more to replace than the one before.&lt;/p&gt;

&lt;h2&gt;
  
  
  What is the work on people who already know you?
&lt;/h2&gt;

&lt;p&gt;It is less glamorous than ads, and it is the part of the business where small changes pay off fastest.&lt;/p&gt;

&lt;p&gt;It comes down to a short list of actions:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  Knowing when each customer should come back.&lt;/li&gt;
&lt;li&gt;  Sending a short, personal note when that date arrives and they have not booked.&lt;/li&gt;
&lt;li&gt;  Calling the ones who matter most if a week passes in silence.&lt;/li&gt;
&lt;li&gt;  Checking later whether the people you contacted returned more often than similar people you did not.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Nothing here needs special software. It needs a date for every customer, a person who owns the list, and the habit of looking at it. The rest of this series takes those pieces one at a time, starting with &lt;a href="https://ivanbolonikhin.com/blog/what-customer-retention-is" rel="noopener noreferrer"&gt;what retention actually is&lt;/a&gt;.&lt;/p&gt;

&lt;p&gt;If your customers pay on a subscription and you use Stripe, there is a more direct way to see the money. A revenue leak report puts failed payments, cancellations and lapsed trials in dollars per month. I explain what it contains in &lt;a href="https://retivo.digital/blog/what-is-a-revenue-leak-report" rel="noopener noreferrer"&gt;what a revenue leak report shows&lt;/a&gt;. I run Retivo, which sells one, so weigh that accordingly. The idea of counting the leak first works with or without my product.&lt;/p&gt;

&lt;h2&gt;
  
  
  Does this apply if I sell once, like furniture or a wedding?
&lt;/h2&gt;

&lt;p&gt;Partly. If a customer buys once every ten years, you will not build a visit schedule around them. But even then there is a second purchase: a related product, a service visit, a referral to a friend. The question is the same. What would a customer need next, and when? If the honest answer is nothing, then your growth really does depend on new customers, and the cost of winning one is the number to watch.&lt;/p&gt;

&lt;p&gt;For most businesses with services, subscriptions, food, pets, beauty, health or anything consumed, the answer is yes. People return, or they do not, and what you do in the weeks after a purchase has a large say in which one it is.&lt;/p&gt;

&lt;h2&gt;
  
  
  What can you do today?
&lt;/h2&gt;

&lt;p&gt;Work out your own number. It takes about ten minutes.&lt;/p&gt;

&lt;ol&gt;
&lt;li&gt; Take last quarter's ad spend and any other cost you attribute to winning customers.&lt;/li&gt;
&lt;li&gt; Count how many new customers it brought.&lt;/li&gt;
&lt;li&gt; Count how many of those came back at least once, however you define it: a second visit, a second order, a renewal.&lt;/li&gt;
&lt;li&gt; Divide the spend by the number who came back.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;That last figure is what a regular costs you. Compare it with what a regular is worth to you over a year. If it surprises you, you have found the first place to look.&lt;/p&gt;

&lt;p&gt;If you have never tracked who came back, that is the first thing to fix. A spreadsheet with a customer name, a first purchase date and a second purchase date is enough to start. Next week it will already tell you more than the ad dashboard does.&lt;/p&gt;

&lt;p&gt;Published 8 November 2026. &lt;a href="https://ivanbolonikhin.com/blog" rel="noopener noreferrer"&gt;All parts&lt;/a&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://ivanbolonikhin.com/blog/why-new-customers-are-no-longer-enough" rel="noopener noreferrer"&gt;ivanbolonikhin.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Ivan Bolonikhin writes about customer retention and building products. &lt;a href="https://ivanbolonikhin.com" rel="noopener noreferrer"&gt;ivanbolonikhin.com&lt;/a&gt;&lt;/p&gt;

</description>
      <category>retention</category>
      <category>startup</category>
      <category>business</category>
      <category>marketing</category>
    </item>
    <item>
      <title>Pflegekräfte aus Indonesien: Was Einrichtungen wissen müssen</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Mon, 16 Nov 2026 12:06:09 +0000</pubDate>
      <link>https://dev.to/2pizza/pflegekrafte-aus-indonesien-was-einrichtungen-wissen-mussen-5hlc</link>
      <guid>https://dev.to/2pizza/pflegekrafte-aus-indonesien-was-einrichtungen-wissen-mussen-5hlc</guid>
      <description>&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqifu86gh6ucmk3me9rvr.jpg" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fqifu86gh6ucmk3me9rvr.jpg" alt="Auf einer Station in einem Krankenhaus in Bandung misst eine indonesische Pflegefachkraft einer älteren Patientin den Blutdruck, die Tochter sitzt daneben." width="800" height="450"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Symbolbild&lt;/p&gt;

&lt;p&gt;Indonesien hat viele diplomierte Pflegekräfte und kaum private Vermittlung nach Deutschland. Vom Vorstellungsgespräch bis zur ersten Schicht vergehen üblicherweise &lt;strong&gt;6 bis 9 Monate&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://pandupflege.de/author/ivan-bolonikhin" rel="noopener noreferrer"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Farticles%2Fy0iev8owcntmkgq6drus.jpg" alt="Ivan Bolonikhin" width="400" height="400"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Stand: 6. Oktober 2026&lt;/p&gt;

&lt;h2&gt;
  
  
  Wer kommt
&lt;/h2&gt;

&lt;p&gt;Zwei Gruppen. Pflegefachkräfte mit Diplom und Berufserfahrung für die direkte Anstellung. Daneben Auszubildende, Schulabsolventinnen mit Deutsch &lt;strong&gt;B1&lt;/strong&gt; für eine dreijährige Ausbildung in der Pflege bei Ihnen. Nach 3 Jahren sind das eigene Pflegefachkräfte, die Ihr Haus bereits kennen.&lt;/p&gt;

&lt;h2&gt;
  
  
  Welches Deutsch
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;B1&lt;/strong&gt; beim Vorstellungsgespräch, &lt;strong&gt;B2&lt;/strong&gt; bei der Ankunft. Sie prüfen das Sprachniveau selbst im Vorstellungsgespräch, bevor Sie sich entscheiden.&lt;/p&gt;

&lt;h2&gt;
  
  
  Wie lange es dauert
&lt;/h2&gt;

&lt;p&gt;Anonymisierte Profile erhalten Sie, sobald passende Kandidatinnen verfügbar sind. Vom Vorstellungsgespräch bis zur ersten Schicht dauert es üblicherweise &lt;strong&gt;6 bis 9 Monate&lt;/strong&gt;. In dieser Zeit laufen Deutschkurs, Anerkennungsverfahren, Visum, Flug und Ankunft.&lt;/p&gt;

&lt;h2&gt;
  
  
  Anerkennung des Diploms
&lt;/h2&gt;

&lt;p&gt;Welche Behörde entscheidet, hängt vom Bundesland ab. Bis zur vollen Anerkennung arbeitet die Kandidatin bereits bei Ihnen als &lt;strong&gt;Pflegehilfskraft&lt;/strong&gt;. Details und die zuständige Stelle für 5 Bundesländer finden Sie im Anerkennungs-Ratgeber.&lt;/p&gt;

&lt;h2&gt;
  
  
  Faire Anwerbung
&lt;/h2&gt;

&lt;p&gt;Die Kandidatin zahlt keine Vermittlungsgebühren und nichts für Flug oder Dokumente. Deutsch bis &lt;strong&gt;B1&lt;/strong&gt; lernt sie selbst, den Kurs von &lt;strong&gt;B1&lt;/strong&gt; auf &lt;strong&gt;B2&lt;/strong&gt; übernehmen wir. Dass Kandidatinnen keine Vermittlungsgebühren zahlen, gehört zu den Kriterien des staatlichen Gütesiegels Faire Anwerbung Pflege Deutschland.&lt;/p&gt;

&lt;h2&gt;
  
  
  Was es für Sie kostet
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;20%&lt;/strong&gt; des Jahresbruttogehalts, als Richtwert rund &lt;strong&gt;8.500 Euro&lt;/strong&gt; je &lt;strong&gt;Pflegefachkraft&lt;/strong&gt; (bei &lt;strong&gt;42.500 Euro&lt;/strong&gt; Jahresbrutto). Gezahlt wird in drei Schritten: &lt;strong&gt;40%&lt;/strong&gt; Grundhonorar bei Unterschrift des Arbeitsvertrags, &lt;strong&gt;40%&lt;/strong&gt; Visumhonorar bei Visumerteilung, &lt;strong&gt;20%&lt;/strong&gt; Erfolgshonorar bei der ersten Schicht. Der Flug kommt dazu: Wir buchen ihn und stellen ihn Ihnen zum Selbstkostenpreis ohne Aufschlag in Rechnung (Richtwert 500 bis &lt;strong&gt;1.000 Euro&lt;/strong&gt;). Verlässt die Pflegekraft Sie in den ersten &lt;strong&gt;6 Monaten&lt;/strong&gt;, finden wir &lt;strong&gt;kostenlos&lt;/strong&gt; einen Ersatz. Details und Vergleich mit anderen Anbietern im Artikel zu den Kosten.&lt;/p&gt;

&lt;h2&gt;
  
  
  Fragen und Antworten
&lt;/h2&gt;

&lt;p&gt;Wie lange dauert es bis zur ersten Schicht?&lt;/p&gt;

&lt;p&gt;Anonymisierte Profile erhalten Sie, sobald passende Kandidatinnen verfügbar sind. Vom Vorstellungsgespräch bis zur ersten Schicht dauert es üblicherweise &lt;strong&gt;6 bis 9 Monate&lt;/strong&gt;.&lt;/p&gt;

&lt;p&gt;Wie gut ist das Deutsch der Kandidatinnen?&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;B1&lt;/strong&gt; beim Vorstellungsgespräch, &lt;strong&gt;B2&lt;/strong&gt; bei der Ankunft. Sie prüfen es selbst im Vorstellungsgespräch.&lt;/p&gt;

&lt;p&gt;Wann kann sie als Pflegefachkraft arbeiten?&lt;/p&gt;

&lt;p&gt;Nach der Anerkennung des Diploms. Bis dahin ist sie bereits als &lt;strong&gt;Pflegehilfskraft&lt;/strong&gt; im Dienst.&lt;/p&gt;

&lt;p&gt;Arbeiten Sie auch mit Österreich?&lt;/p&gt;

&lt;p&gt;Ja. Gleiches Deutsch, gleiche Kandidatinnen.&lt;/p&gt;

&lt;p&gt;Kandidatinnen aus Indonesien kennenlernen&lt;/p&gt;

&lt;p&gt;Sie erhalten anonymisierte Profile, sobald passende Kandidatinnen verfügbar sind.&lt;/p&gt;

&lt;h2&gt;
  
  
  Weitere Artikel
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/indonesische-pflegekraefte-kosten" rel="noopener noreferrer"&gt;Indonesische Pflegekräfte: Was die Vermittlung kostet&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/anerkennung-indonesische-pflegeausbildung" rel="noopener noreferrer"&gt;Anerkennung der indonesischen Pflegeausbildung&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/faire-anwerbung-indonesien" rel="noopener noreferrer"&gt;Faire Anwerbung aus Indonesien&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/was-ist-pflegekraefte-anwerbung-indonesien" rel="noopener noreferrer"&gt;Was ist die Anwerbung von Pflegekräften aus Indonesien?&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/pflegefachkraft-aus-indonesien-einstellen-schritt-fuer-schritt" rel="noopener noreferrer"&gt;Pflegefachkraft aus Indonesien einstellen: Schritt für Schritt&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/kosten-pflegefachkraft-indonesien-alle-posten" rel="noopener noreferrer"&gt;Was kostet eine Pflegefachkraft aus Indonesien wirklich?&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/anerkennungspartnerschaft-16d-aufenthg-arbeitgeber" rel="noopener noreferrer"&gt;Anerkennungspartnerschaft nach § 16d AufenthG: Was Arbeitgeber wissen müssen&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/defizitbescheid-was-bedeutet-das-fuer-ihre-einrichtung" rel="noopener noreferrer"&gt;Defizitbescheid: Was er bedeutet und was Ihre Einrichtung jetzt plant&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/kenntnispruefung-oder-anpassungslehrgang-unterschied" rel="noopener noreferrer"&gt;Kenntnisprüfung oder Anpassungslehrgang: Was unterscheidet sie?&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/pflegehilfskraft-bis-zur-anerkennung-einsatz-planen" rel="noopener noreferrer"&gt;Pflegehilfskraft bis zur Anerkennung: So planen Sie den Einsatz&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/deutsch-b1-b2-pflegekraefte-indonesien" rel="noopener noreferrer"&gt;Deutsch B1 und B2: Welches Sprachniveau Pflegekräfte aus Indonesien brauchen&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/d3-s1-ners-welche-kandidatin-passt" rel="noopener noreferrer"&gt;D3, S1 oder Ners: Welche Kandidatin passt zu Ihrem Haus?&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/zustaendige-behoerde-anerkennung-pflege-bundeslaender" rel="noopener noreferrer"&gt;Welche Behörde erkennt das Pflegediplom an? Fünf Bundesländer im Überblick&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/oesterreich-nostrifikation-pflegekraefte-indonesien" rel="noopener noreferrer"&gt;Nostrifikation in Österreich: Wie Pflegekräfte aus Indonesien anerkannt werden&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/rot-weiss-rot-karte-pflegekraefte-oesterreich" rel="noopener noreferrer"&gt;Rot-Weiß-Rot-Karte für Pflegekräfte: Punkte, Gebühr, Fristen&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/staatliches-programm-oder-vollservice-anbieter-vergleichen" rel="noopener noreferrer"&gt;Staatliches Programm, Vollservice oder junger Anbieter: Wie vergleiche ich?&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/vertrag-pruefen-checkliste-anwerbung-pflegekraefte" rel="noopener noreferrer"&gt;Vertrag prüfen: Checkliste für die Anwerbung von Pflegekräften&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/seriose-anbieter-erkennen-who-liste-p3mi-guetesiegel" rel="noopener noreferrer"&gt;Seriöse Anbieter erkennen: WHO-Liste, P3MI-Lizenz und Gütesiegel prüfen&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/fehler-bei-der-anwerbung-auslaendischer-pflegekraefte" rel="noopener noreferrer"&gt;Zehn Fehler bei der Anwerbung von Pflegekräften aus dem Ausland&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  &lt;a href="https://pandupflege.de/ratgeber/pflegekraefte-ausland-anwerben-agenturen-wege-vergleich" rel="noopener noreferrer"&gt;Pflegekräfte aus dem Ausland anwerben: Agenturen und Wege im Vergleich&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://pandupflege.de/ratgeber/pflegekraefte-aus-indonesien" rel="noopener noreferrer"&gt;pandupflege.de&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Pandu vermittelt qualifizierte Pflegekräfte aus Indonesien an deutsche Einrichtungen. &lt;a href="https://pandupflege.de" rel="noopener noreferrer"&gt;pandupflege.de&lt;/a&gt;&lt;/p&gt;

</description>
      <category>pflege</category>
      <category>healthcare</category>
      <category>germany</category>
      <category>recruiting</category>
    </item>
    <item>
      <title>Allowance: How Much to Give and What It Should Pay For</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Mon, 16 Nov 2026 12:00:58 +0000</pubDate>
      <link>https://dev.to/2pizza/allowance-how-much-to-give-and-what-it-should-pay-for-2njf</link>
      <guid>https://dev.to/2pizza/allowance-how-much-to-give-and-what-it-should-pay-for-2njf</guid>
      <description>&lt;p&gt;&lt;a href="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fhihuman.space%2F_next%2Fimage%3Furl%3D%252Flanding%252Fivan-bolonikhin.jpg%26w%3D96%26q%3D75" class="article-body-image-wrapper"&gt;&lt;img src="https://media2.dev.to/dynamic/image/width=800%2Cheight=%2Cfit=scale-down%2Cgravity=auto%2Cformat=auto/https%3A%2F%2Fhihuman.space%2F_next%2Fimage%3Furl%3D%252Flanding%252Fivan-bolonikhin.jpg%26w%3D96%26q%3D75" alt="Ivan Bolonikhin" width="96" height="96"&gt;&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;&lt;a href="https://hihuman.space/en/author/ivan-bolonikhin" rel="noopener noreferrer"&gt;Ivan Bolonikhin&lt;/a&gt;&lt;/p&gt;

&lt;p&gt;Founder, HiHuman · November 8, 2026&lt;/p&gt;

&lt;p&gt;There is no universal amount, and I will not invent one. There is an order that works: first decide what the money is for, then pick how often to give it, then separate unpaid family duties from paid extras. Every number below is just an example.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why give kids an allowance at all?
&lt;/h2&gt;

&lt;p&gt;An allowance is teaching material. The child gets a limited sum and decides for themselves what to do with it. A mistake with a small amount is much cheaper than a mistake at twenty with a big one.&lt;/p&gt;

&lt;p&gt;If the money is only a pressure tool, it loses its point. It exists so a child can feel a limit: it did not cover everything, so I have to choose.&lt;/p&gt;

&lt;h2&gt;
  
  
  How much allowance should I give?
&lt;/h2&gt;

&lt;p&gt;I do not tie the amount to someone else's average, because families have different incomes and different expenses. I tie it to two questions: what does the child buy on their own, and how much can I trust them with without worrying?&lt;/p&gt;

&lt;p&gt;For example, if a child buys small treats and stickers, $2-3 a week might be enough. A teenager who pays for their own rides and lunches may need noticeably more. These are examples of scale, not a recommendation.&lt;/p&gt;

&lt;p&gt;The rule is simple: the amount should be enough to have something to choose from, and small enough that a mistake is not a tragedy.&lt;/p&gt;

&lt;h2&gt;
  
  
  What should I pay for, and what should be unpaid?
&lt;/h2&gt;

&lt;p&gt;I split chores into two groups. First, the duties of anyone who lives in the house: clean up after yourself, pack your backpack, do not leave your stuff everywhere. No money for those, because they are not a job, they are being part of a family. Second, extras above the baseline, like washing the car or helping with a big cleanup. Those can be paid.&lt;/p&gt;

&lt;p&gt;If you pay for everything, the child starts asking how much will you give me for every request. That is a dangerous habit, and it sets fast.&lt;/p&gt;

&lt;h2&gt;
  
  
  Is a fixed allowance better than paying for chores?
&lt;/h2&gt;

&lt;p&gt;I use a hybrid. A small base amount comes regularly, so the child learns to plan. Extra money can be earned with chores above the baseline. That gives the child both predictability and a reason to try.&lt;/p&gt;

&lt;p&gt;In our system, points and money are separate. Points go toward privileges inside the house, money goes toward purchases. You can exchange one for the other at a rate set in advance if you want to link them. That is each family's call.&lt;/p&gt;

&lt;h2&gt;
  
  
  Should I take away allowance as a punishment?
&lt;/h2&gt;

&lt;p&gt;I avoid it. If money is a teaching tool, taking it away means pausing the teaching. Also, a child quickly figures out that the money depends on a parent's mood, not on rules.&lt;/p&gt;

&lt;p&gt;Instead, tie the consequence to the thing that happened. Broke something through carelessness? They help pay for the replacement from money they earned. That is fairer than an abstract fine.&lt;/p&gt;

&lt;h2&gt;
  
  
  What age should a child start getting allowance?
&lt;/h2&gt;

&lt;p&gt;When the child understands that money runs out and can be traded for a thing. For many kids that happens in elementary school, but it differs from child to child. I started with very small amounts and very simple rules: your money, your choice, no top-ups.&lt;/p&gt;

&lt;p&gt;For teenagers it makes sense to move from weekly to monthly. It is harder, because you have to stretch the money over a longer time. But it is closer to adult life, where a paycheck comes once a month.&lt;/p&gt;

&lt;h2&gt;
  
  
  What do I do when my child asks for more?
&lt;/h2&gt;

&lt;p&gt;Ask what for. If it is a specific goal, suggest earning extra with additional chores, or waiting. If it is just I want it, that is a reason to talk about how a budget works. We look together: what was bought, what is left, what could be saved.&lt;/p&gt;

&lt;p&gt;Do not hand out extra in a moment of weakness. Give in once and the child remembers that pressure works. Better to announce the rule ahead of time: extra money comes only for extra work.&lt;/p&gt;

&lt;p&gt;Gifts from relatives work the same way: part gets saved and part is spent as the child likes, for example half and half, as an example only.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do I link money to chores without killing motivation?
&lt;/h2&gt;

&lt;p&gt;My principle: we do not pay for the baseline, we pay for extras, and the child funds big goals partly on their own. If they want an expensive item, they cover part from what they earned, and I add the rest. A split like that, say half and half as an example, shows them the price of a wish and does not leave them alone with it.&lt;/p&gt;

&lt;p&gt;Watch that money does not crowd out interest. A child who walks the dog only for pay will stop when the pay stops being exciting. So I try to say out loud, in parallel, why it matters: the dog needs the walk, and the dog needs you.&lt;/p&gt;

&lt;h2&gt;
  
  
  What if my child loses the money or spends it all at once?
&lt;/h2&gt;

&lt;p&gt;The first time it is nearly unavoidable, and I count it as part of learning. Do not scold and do not compensate. Ask calmly: what did you want to buy, and how did it turn out? Often the child says what you wanted to say and hears it better from themselves.&lt;/p&gt;

&lt;p&gt;If the money was lost rather than spent, you can look for it together and retrace where it might have been left. But replacing the loss from your own pocket is not worth it. Then the lesson about the price of money never happens, and the child learns that a parent always bails them out.&lt;/p&gt;

&lt;p&gt;I make an exception when a child was cheated or someone took their money. That is no longer about a lesson, it is about safety, and the conversation should be different: detailed and warm.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do I explain needs versus wants?
&lt;/h2&gt;

&lt;p&gt;Better through practice than through a lecture. Sometimes we play a simple game: before a store trip, the child decides what they will buy with their own money and what they will ask us for. Candy and toys come from their pocket. School supplies and necessary clothes come from the family budget.&lt;/p&gt;

&lt;p&gt;That makes the line visible. The child learns by doing that the family buys one kind of thing because it is needed, and they buy the other kind because they want it.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do I teach a child to save?
&lt;/h2&gt;

&lt;p&gt;Give them a goal. Not an abstract save up, but a specific item a few weeks away. Watch how the child decides whether to spend now or wait.&lt;/p&gt;

&lt;p&gt;Do not rescue them from disappointment. If they spent everything on day one and want something else, they wait for the next payout.&lt;/p&gt;

&lt;p&gt;If you notice that money causes your child strong anxiety or fixation, it is worth talking to a specialist. For most families, a regular allowance is a calm and useful tool.&lt;/p&gt;

&lt;p&gt;Want to run this as a game instead of in your head?&lt;/p&gt;

&lt;p&gt;HiHuman has tasks priced in points, a weekly target, rewards, rules and a ranking. The parent stays in charge, and the child sees clear rules.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://hihuman.space/en/blog/allowance-how-much-and-for-what" rel="noopener noreferrer"&gt;hihuman.space&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;HiHuman is a family app that teaches kids responsibility through chores, goals and a points system. &lt;a href="https://hihuman.space" rel="noopener noreferrer"&gt;hihuman.space&lt;/a&gt;&lt;/p&gt;

</description>
      <category>parenting</category>
      <category>family</category>
      <category>education</category>
      <category>kids</category>
    </item>
    <item>
      <title>What is patient win-back for an aesthetic clinic?</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Mon, 16 Nov 2026 11:55:48 +0000</pubDate>
      <link>https://dev.to/2pizza/what-is-patient-win-back-for-an-aesthetic-clinic-1ii0</link>
      <guid>https://dev.to/2pizza/what-is-patient-win-back-for-an-aesthetic-clinic-1ii0</guid>
      <description>&lt;p&gt;Patient win-back is a routine in which your clinic contacts people who were already your patients when a review is due, instead of waiting for them to remember you. It is one of the three things Conduit does.&lt;/p&gt;

&lt;p&gt;I built Conduit for a cosmetic surgery group. This article is the plain version of how win-back works, what a message can and cannot say in Australia, and where its limits are.&lt;/p&gt;

&lt;h2&gt;
  
  
  What is patient win-back in an aesthetic clinic?
&lt;/h2&gt;

&lt;p&gt;Patient win-back is a standing rule that says: after this treatment, this long, the patient hears from the clinic once. It is not a campaign with a start and an end date, and it is not a discount.&lt;/p&gt;

&lt;p&gt;Most clinics already hold the raw material. Every person who has booked sits in a booking system such as Timely, Zenoti, Fresha, Kitomba or Cliniko, with a last visit date next to their name. Win-back turns that date into a prompt. When the date is far enough back, the patient gets one message from the clinic, in the clinic's name, with a way to book and a way to opt out.&lt;/p&gt;

&lt;p&gt;You will also hear it called patient return, reactivation or recall. On our pricing page the reminders are called patient return reminders. The idea is the same in every case: the patient has already chosen you once, and a short, honest message is the simplest way to make it easy to choose you again.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who counts as a lapsed patient?
&lt;/h2&gt;

&lt;p&gt;A lapsed patient is someone who has not been in for longer than your clinic would expect for their treatment. There is no universal cut-off, so you set it by treatment, with your practitioners.&lt;/p&gt;

&lt;p&gt;In the clinic check on our home page, the starting assumption is a patient with no visit in 6 or more months. The same calculator assumes that 35% of a patient database falls in that group and that 10% of them return after reminders. Those are starting assumptions, not measured results. The page says so, shows the formula, and lets you change every one of them.&lt;/p&gt;

&lt;p&gt;If your own numbers look different, use yours. If you do not know them yet, pull a last-visit report from your booking system before you do anything else. That report is more useful than any assumption of mine.&lt;/p&gt;

&lt;h2&gt;
  
  
  How is win-back different from a promotion?
&lt;/h2&gt;

&lt;p&gt;A win-back message says that a review is due. A promotion says that something is cheaper or will not be available for long. In Australian cosmetic advertising the second kind is the one that causes trouble.&lt;/p&gt;

&lt;p&gt;The check on this site treats discounts, add-ons at no charge and time-limited offers as inducements to have a procedure, which the cosmetic guidelines do not allow. That is why a Conduit message contains no offer and no deadline. The reason to contact the patient is your clinic's own schedule, not a deal.&lt;/p&gt;

&lt;p&gt;It also changes what you can say honestly. A clinic that writes 'you are due for your next review' is stating a fact about the patient's file. A clinic that writes 'book this week' is applying pressure. I would rather send the first kind and accept that it is quieter.&lt;/p&gt;

&lt;h2&gt;
  
  
  What can a win-back message say under Ahpra and TGA rules?
&lt;/h2&gt;

&lt;p&gt;It can say who you are, that a review is due, how to book and how to stop receiving messages. It cannot name prescription medicines, make an offer, quote a patient review or promise a result.&lt;/p&gt;

&lt;p&gt;Those are the four things the site's check looks for in every message before it sends:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  No prescription medicine names, direct or indirect&lt;/li&gt;
&lt;li&gt;  No offers or deadlines&lt;/li&gt;
&lt;li&gt;  No patient reviews or testimonials&lt;/li&gt;
&lt;li&gt;  No ranking claims or promises about results&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Every message also carries a STOP option. A message that passes looks like this:&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;Hi [first name], you are due for your next review at [clinic name]. You can book here: [link]. Reply STOP to opt out&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Notice what is missing: no treatment name, no price, no reason to rush, no story about someone else's result. That can feel plain. It is meant to.&lt;/p&gt;

&lt;p&gt;The rules behind those four lines are the Ahpra guidelines for advertising higher-risk non-surgical cosmetic procedures (2 September 2025), the Ahpra and Medical Board guidelines for advertising cosmetic surgery (2023), the TGA advertising rules and section 133 of the Health Practitioner Regulation National Law. We hold messages to the same rules as your website. I am not your lawyer, and the check is a wording check. Your clinic approves every template, so final sign-off stays with you.&lt;/p&gt;

&lt;h2&gt;
  
  
  How many messages does a patient get?
&lt;/h2&gt;

&lt;p&gt;One message when a review is due, not a stream. This is a design choice. A run of reminders starts to read as pressure, and in a regulated field I would rather send one message and let the patient decide.&lt;/p&gt;

&lt;p&gt;You set how long after each treatment a review is due. Different treatments have different rhythms, and your practitioners know them better than I do. Conduit does not decide clinical timing. It holds the intervals you choose and applies them to the list.&lt;/p&gt;

&lt;p&gt;Every message goes out in your clinic's name, from a template you approved, with a STOP option. Conduit does not write to your patients under its own name.&lt;/p&gt;

&lt;h2&gt;
  
  
  How does win-back work with Timely, Zenoti, Fresha, Kitomba or Cliniko?
&lt;/h2&gt;

&lt;p&gt;It runs from an export of your patient list, so you do not switch booking system and you do not retype anyone. You export from your system, or send a CSV, and we map it to your treatments with you. Updates run from a fresh export on the schedule you choose.&lt;/p&gt;

&lt;p&gt;A direct Timely connection is in progress, but it is not live today. The export is how it works now. If I told you it was a live integration, I would be wrong, and you would find out in the first week.&lt;/p&gt;

&lt;p&gt;Your bookings stay where they are. Nothing about win-back needs your front desk to learn a second booking system.&lt;/p&gt;

&lt;h2&gt;
  
  
  What does win-back not do?
&lt;/h2&gt;

&lt;p&gt;It does not do five things, and it is better to hear them from me now than to discover them later.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  It does not promise a number of patients, bookings or sales. We say that on the pricing page.&lt;/li&gt;
&lt;li&gt;  It does not give medical advice. Clinical questions go to your team.&lt;/li&gt;
&lt;li&gt;  It does not write to patients under its own name. Messages go out in your clinic's name.&lt;/li&gt;
&lt;li&gt;  It does not replace the people who talk to your patients. It starts the conversation and they carry it on.&lt;/li&gt;
&lt;li&gt;  It does not turn a wording check into legal advice. The check can miss things and can flag things that are fine.&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  How do you know if win-back is working?
&lt;/h2&gt;

&lt;p&gt;Compare patients who were messaged with patients who were not. A raw count of returns is not enough, because some patients would have come back anyway.&lt;/p&gt;

&lt;p&gt;The report in Conduit shows which enquiries became consults and what returning patients brought back, compared with patients we did not message. You can run the same test without any tool: hold back a part of your lapsed list for one cycle, message the rest, and compare the two groups afterwards.&lt;/p&gt;

&lt;p&gt;I would be suspicious of any provider who shows you a return rate without saying what it was compared against. Ask for the comparison group first.&lt;/p&gt;

&lt;h2&gt;
  
  
  What does patient win-back cost?
&lt;/h2&gt;

&lt;p&gt;In Conduit, patient return reminders are the Start plan: from A$99 a month, plus a one-off setup of about A$400. There is no minimum term and you can cancel any time.&lt;/p&gt;

&lt;p&gt;The other plans add things around it. Grow, priced on a call after we see your clinic, adds a reply to every enquiry in seconds. Pro, also priced on a call, adds an AI agent on your clinic number and a patient portal. The &lt;a href="https://conduitclinics.com/pricing" rel="noopener noreferrer"&gt;pricing page&lt;/a&gt; has the full list.&lt;/p&gt;

&lt;p&gt;Whether Start is worth it depends on your last-visit data and your average visit value, which I do not have. The &lt;a href="https://conduitclinics.com/ahpra-check" rel="noopener noreferrer"&gt;clinic check&lt;/a&gt; and the calculator on the home page work it out from your own numbers and show you the formula, so you can argue with it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where should you start?
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt; 1Pull a last-visit report from your booking system.&lt;/li&gt;
&lt;li&gt; 2Decide with your practitioners how long after each treatment a review is due.&lt;/li&gt;
&lt;li&gt; 3Write the message in plain words and test it against the four rules above.&lt;/li&gt;
&lt;li&gt; 4Run your website through the &lt;a href="https://conduitclinics.com/ahpra-check" rel="noopener noreferrer"&gt;Ahpra check&lt;/a&gt;, because a message is only as safe as the site it links to.&lt;/li&gt;
&lt;li&gt; 5Send to a small group first and read what comes back.&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;If you want to see it on your own clinic, &lt;a href="https://conduitclinics.com/book" rel="noopener noreferrer"&gt;book a 20-minute demo&lt;/a&gt;. We run a test enquiry through the system so you can see each step, and you decide whether it fits.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://conduitclinics.com/blog/what-is-patient-win-back-for-an-aesthetic-clinic" rel="noopener noreferrer"&gt;conduitclinics.com&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Conduit helps aesthetic clinics win back patients and capture every enquiry. &lt;a href="https://conduitclinics.com" rel="noopener noreferrer"&gt;conduitclinics.com&lt;/a&gt;&lt;/p&gt;

</description>
      <category>healthcare</category>
      <category>marketing</category>
      <category>ai</category>
      <category>business</category>
    </item>
    <item>
      <title>What is a retention CRM for online casinos?</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Mon, 16 Nov 2026 11:50:37 +0000</pubDate>
      <link>https://dev.to/2pizza/what-is-a-retention-crm-for-online-casinos-474m</link>
      <guid>https://dev.to/2pizza/what-is-a-retention-crm-for-online-casinos-474m</guid>
      <description>&lt;p&gt;A retention CRM for an online casino is a system that holds the history of every player, predicts what each one is likely to do next, and turns that into a short list of actions for your team: who to contact today, with which offer, through which channel. I run Sellrise, so I am biased, but I will try to keep this definition useful even if you never buy from us.&lt;/p&gt;

&lt;h2&gt;
  
  
  What does a retention CRM actually do?
&lt;/h2&gt;

&lt;p&gt;It does three jobs, and the order matters. First it keeps one record per player: segments, actions taken, contact history. Second it scores each player with models: churn risk, VIP potential, the chance of the next deposit, and the value the player is likely to bring over 90 days. Third it converts those scores into an action for a person: contact this player today, by WhatsApp, with a reload from your catalogue.&lt;/p&gt;

&lt;p&gt;The third job is what separates a retention CRM from a reporting tool. A report tells you that churn went up last month. A retention CRM tells a manager that player 48392 looks like a VIP in the first week and should get a call today.&lt;/p&gt;

&lt;p&gt;In Sellrise the screen your team opens every morning is exactly that list. For each player it shows the reason, the priority, the channel, the time to contact, the offer and the next step. The point is to move from "10,000 players" to "the 86 who need attention today". Those numbers are from our demo data, not from a client.&lt;/p&gt;

&lt;h2&gt;
  
  
  How is it different from a generic CRM or a marketing automation tool?
&lt;/h2&gt;

&lt;p&gt;A generic CRM stores contacts and deals. A marketing automation tool runs journeys that your team designs: if a player does X, send Y. Both are useful, and both depend on a human deciding the rules and the segments in advance.&lt;/p&gt;

&lt;p&gt;A retention CRM adds a prediction layer underneath. Instead of a segment such as "no deposit in 14 days", you get a per-player score and a reason. Instead of one offer per cohort, you get an offer matched to the player's stage and value, and sometimes the answer "do not give this player a bonus now, they are already winning".&lt;/p&gt;

&lt;p&gt;That last point is where a lot of bonus budget goes. The same bonus for the whole base is expensive, because part of it goes to players who were going to deposit anyway. A retention CRM should be able to show you what bonuses actually did: deposits after the bonus, and how much of that the bonus itself brought.&lt;/p&gt;

&lt;p&gt;None of this makes a generic tool bad. If your team already runs strong event-driven journeys, you may only need the prediction layer next to them. I wrote up the honest trade-offs against established platforms in our &lt;a href="https://sellrise.ai/vs" rel="noopener noreferrer"&gt;comparison pages&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What data does a retention CRM need?
&lt;/h2&gt;

&lt;p&gt;Less than most teams expect. For the forecasts in Sellrise we need four tables keyed by Player ID:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  Players&lt;/li&gt;
&lt;li&gt;  Deposits&lt;/li&gt;
&lt;li&gt;  Bets&lt;/li&gt;
&lt;li&gt;  Sessions&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Name, phone and email are optional and only needed if messages and calls go through the system. A Player ID is enough for the models. Every client gets a separate database, and integration is at our cost.&lt;/p&gt;

&lt;p&gt;In general terms, the work behind any scoring system is mostly in the data layer, not in the model. Features should be computed from raw transactions at the moment of scoring, never from pre-aggregated platform statistics that may already contain information from the future. I wrote about this in more detail in two engineering notes on &lt;a href="https://2pizza.team/blog/per-player-scoring-architecture-casino" rel="noopener noreferrer"&gt;per-player scoring architecture&lt;/a&gt; and &lt;a href="https://2pizza.team/blog/igaming-retention-playbook-gradient-boosting" rel="noopener noreferrer"&gt;gradient boosting for iGaming retention&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  What forecasts should it produce?
&lt;/h2&gt;

&lt;p&gt;Four are worth asking for, and each should be shown honestly:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;  &lt;strong&gt;Churn risk&lt;/strong&gt;: who is about to leave, flagged before they are gone rather than after they show up in a report.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;VIP potential&lt;/strong&gt;: future VIPs spotted from the first week after the first deposit, so a manager can pick them up early.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Chance of the next deposit&lt;/strong&gt;: especially the step from a one-time depositor to a returning player, when a well-timed offer matters most.&lt;/li&gt;
&lt;li&gt;  &lt;strong&gt;Value over 90 days&lt;/strong&gt;: always as a range. A single number on a 90-day horizon is fiction, and a system that gives you one is hiding its uncertainty.&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If a vendor shows you only a risk score with no reason, ask for the reason. Your team will not act on a number nobody can explain.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who decides and who sends?
&lt;/h2&gt;

&lt;p&gt;In our model, your team. Sellrise suggests, a person decides and sends, through your own channels or through ours (SMS, Viber, WhatsApp, email, Telegram and calls; push and in-app go through your platform). Nothing goes out on its own.&lt;/p&gt;

&lt;p&gt;There is a scale of automation and you choose where to stand on it. As a CRM, all players, segments and contacts live in Sellrise and your team works inside it. With suggestions, Sellrise proposes the action and a person decides. On autopilot, scenarios run within your rules. We switch autopilot on only after the pilot, once the result is proven on your own players.&lt;/p&gt;

&lt;p&gt;I prefer that order because a model that has not yet been checked against your data should not be sending messages to your customers.&lt;/p&gt;

&lt;h2&gt;
  
  
  Does a retention CRM replace the CRM we already have?
&lt;/h2&gt;

&lt;p&gt;Not necessarily. Sellrise works alongside your platform and CRM: your scenarios stay where they are, and the system gives your team lists and suggestions as an export or over an API. If you do not have a CRM, it can be your main one.&lt;/p&gt;

&lt;p&gt;Whether to run it alongside or as the main system is a practical question about your team. If you have a CRM team that already builds journeys, keep them and add the forecasts. If you have a small team and a thin ARPU, one system is simpler.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do you know a retention CRM works?
&lt;/h2&gt;

&lt;p&gt;You measure it on your own players. The test I recommend, and the one we run in a pilot, has three groups: a control group that nobody contacts, your team working as it does today, and your team following the system's suggestions. After one month you compare average turnover per player in each group. The difference is the effect.&lt;/p&gt;

&lt;p&gt;We do not quote percentages before the test, because any percentage we could quote would come from someone else's players. We agree the group sizes and the metric before we start, so nobody can move the goalposts afterwards.&lt;/p&gt;

&lt;p&gt;This is also the question to ask any vendor: how will you prove it on my data, and what happens if you cannot?&lt;/p&gt;

&lt;h2&gt;
  
  
  Who is a retention CRM not for?
&lt;/h2&gt;

&lt;p&gt;Be honest about fit. Sellrise has no gamification: no loyalty mechanics, missions or tournaments inside your product. If gamification is your first lever, look at platforms built for it. If you want a full enterprise marketing hub at the centre of your stack with a dedicated team building campaigns, that is a different category, and I would not pretend otherwise.&lt;/p&gt;

&lt;p&gt;It fits when you want a forecast for every player and a short list for today, when your ARPU is thin enough that per-seat pricing hurts, and when you want to keep the stack you have. If that sounds like you, the &lt;a href="https://app.sellrise.ai/demo" rel="noopener noreferrer"&gt;demo&lt;/a&gt; runs on synthetic data and shows the morning list, or you can &lt;a href="https://sellrise.ai/pilot" rel="noopener noreferrer"&gt;start a pilot&lt;/a&gt; on your own players.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://sellrise.ai/blog/what-is-a-casino-retention-crm" rel="noopener noreferrer"&gt;sellrise.ai&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Sellrise is an AI retention CRM for online casinos and iGaming operators. &lt;a href="https://sellrise.ai" rel="noopener noreferrer"&gt;sellrise.ai&lt;/a&gt;&lt;/p&gt;

</description>
      <category>crm</category>
      <category>ai</category>
      <category>analytics</category>
      <category>saas</category>
    </item>
    <item>
      <title>Best tools to recover failed Stripe payments in 2026</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Mon, 16 Nov 2026 11:50:12 +0000</pubDate>
      <link>https://dev.to/2pizza/best-tools-to-recover-failed-stripe-payments-in-2026-ac1</link>
      <guid>https://dev.to/2pizza/best-tools-to-recover-failed-stripe-payments-in-2026-ac1</guid>
      <description>&lt;p&gt;If you are on Stripe and want to recover failed payments, start with Stripe's own retries and emails, which cost nothing extra. Add a dedicated service when you can show a leak large enough to pay for it. Below are seven options with who each fits and where each is limited. I founded Retivo, which is one of the seven, so I have not put it first and I have said what it does not do.&lt;/p&gt;

&lt;h2&gt;
  
  
  What are the options?
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt; Stripe's built-in retries and customer emails&lt;/li&gt;
&lt;li&gt; Churn Buster&lt;/li&gt;
&lt;li&gt; Baremetrics Recover&lt;/li&gt;
&lt;li&gt; Churnkey&lt;/li&gt;
&lt;li&gt; Slicker&lt;/li&gt;
&lt;li&gt; A do-it-yourself setup&lt;/li&gt;
&lt;li&gt; Retivo&lt;/li&gt;
&lt;/ol&gt;

&lt;p&gt;Competitor facts come from each company's own public page, read on 9 November 2026. I do not quote prices and I do not rate anyone's quality. Check each site for current details.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Stripe's built-in retries and emails
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Fits:&lt;/strong&gt; every Stripe Billing account, as the first step.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Strengths:&lt;/strong&gt; no extra vendor. Smart Retries chooses retry times with signals Stripe describes in its docs, and for new accounts it is on by default at up to 8 retries within 2 weeks. You can also switch on failed-payment, expiring-card, trial-ending and renewal emails that link to a card update page.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Limits:&lt;/strong&gt; per its docs, hard declines such as a lost or stolen card are not retried without a new payment method, and you configure and measure everything yourself. See &lt;a href="https://retivo.digital/blog/retivo-vs-stripe-smart-retries-and-recovery-emails" rel="noopener noreferrer"&gt;Retivo vs Stripe Smart Retries and recovery emails&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  2. Churn Buster
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Fits:&lt;/strong&gt; subscription businesses that want failed-payment recovery and cancel flows run on an ongoing basis, on Stripe or several other platforms.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Strengths:&lt;/strong&gt; its page lists tuned retries, recovery emails from a dedicated pool on your domain, then SMS and a team alert, and a no-login card update page with 3D Secure on Stripe and Braintree. It also lists reason-based cancel flows. Billing is a flat monthly fee by MRR tier, month-to-month.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Limits:&lt;/strong&gt; its page does not state a minimum size or list limitations. Details in &lt;a href="https://retivo.digital/blog/retivo-vs-churn-buster" rel="noopener noreferrer"&gt;Retivo vs Churn Buster&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Baremetrics Recover
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Fits:&lt;/strong&gt; teams already on Baremetrics who want dunning in the same product.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Strengths:&lt;/strong&gt; timed emails, optional SMS, an in-app banner that can turn into a paywall, and a hosted card update form on a Baremetrics subdomain or your own domain.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Limits:&lt;/strong&gt; its page says it has no retry schedule of its own, and write-back of an updated card is limited to Stripe, Braintree and Recurly. See &lt;a href="https://retivo.digital/blog/retivo-vs-baremetrics-recover" rel="noopener noreferrer"&gt;Retivo vs Baremetrics Recover&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  4. Churnkey
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Fits:&lt;/strong&gt; Stripe businesses that want cancel flows and payment recovery in one product, especially without engineering time to spare.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Strengths:&lt;/strong&gt; its page says it is built specifically for Stripe, with setup in as little as 15 minutes. Payment Recovery lists retries layered on Stripe's retry logic, email, SMS and in-app dunning, and a payment wall that blocks features until the card is updated. It also lists cancel flows with exit surveys and offers.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Limits:&lt;/strong&gt; the page says its savings figures are results of top-performing customers, not guaranteed outcomes. I have not compared it with the others in depth.&lt;/p&gt;

&lt;h2&gt;
  
  
  5. Slicker
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Fits:&lt;/strong&gt; companies with recurring billing across several billing systems or payment providers, including Stripe Billing.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Strengths:&lt;/strong&gt; per its page, models decide whether, when and for how much to retry using gateway codes and customer behaviour; customers are emailed only when action is needed; and it runs crossover tests against your current setup. It prices recovery on results: you pay only if it beats your current setup with statistical significance.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Limits:&lt;/strong&gt; its page says SOC 2 certification is in progress, and some of its retry logic uses industry heuristics alongside machine learning.&lt;/p&gt;

&lt;h2&gt;
  
  
  6. A do-it-yourself setup
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Fits:&lt;/strong&gt; businesses below roughly 2,000 paying customers, and anyone who wants to learn their own numbers first.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Strengths:&lt;/strong&gt; costs only your time, and you own every message. Use Stripe's settings, write a short sequence, and keep a holdout.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Limits:&lt;/strong&gt; the counting rules are the hard part and nobody checks your work. I describe the comparison in &lt;a href="https://retivo.digital/blog/diy-spreadsheet-report-or-retention-software" rel="noopener noreferrer"&gt;DIY spreadsheet, one-time report or retention software&lt;/a&gt; and the steps in &lt;a href="https://retivo.digital/blog/recover-failed-stripe-payments" rel="noopener noreferrer"&gt;how to recover failed payments on Stripe subscriptions&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  7. Retivo
&lt;/h2&gt;

&lt;p&gt;&lt;strong&gt;Fits:&lt;/strong&gt; Stripe subscription businesses with roughly 2,000 or more paying customers that want the leak sized before buying a tool, and want any fix proven.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Strengths:&lt;/strong&gt; a $490 Revenue Leak Report from a restricted read-only Stripe key, delivered within 48 hours of my having the key: failed payments, unconverted trials, cancellations and downgrades in dollars per month, plus the top three fixes. An optional 90-day pilot costs $1,500 setup plus 20% of measured net uplift against a 20% holdout.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Limits:&lt;/strong&gt; it is Stripe only. The report is a read of history and proves nothing about a fix. Email is the live channel; WhatsApp, SMS, Viber and Telegram are on request. It is not an ongoing tool you log into. See a &lt;a href="https://retivo.digital/sample-report" rel="noopener noreferrer"&gt;sample report with synthetic data&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do I choose among them?
&lt;/h2&gt;

&lt;ol&gt;
&lt;li&gt; Have you read what your Stripe retries and emails are set to? If not, do that first.&lt;/li&gt;
&lt;li&gt; Do you know how many dollars a month you lose to failed payments? If not, count it by hand or order a report. A tool cannot be judged without that number.&lt;/li&gt;
&lt;li&gt; Is the leak large enough to pay for a recurring service? If so, ask each provider how it proves the result, what you pay if nothing improves, whose brand and domain the messages use, and whether you can undo every change.&lt;/li&gt;
&lt;li&gt; Do you also lose money to cancellations and trials? Then check whether the product covers those, not just payments.&lt;/li&gt;
&lt;/ol&gt;

&lt;h2&gt;
  
  
  Which option recovers the most?
&lt;/h2&gt;

&lt;p&gt;I cannot say, and I would doubt anyone who gives you a ranking without your data. Recovery depends on your decline mix, your customers and your price. The only comparison that holds up is a test on your own customers with a group left alone. I explain it in &lt;a href="https://retivo.digital/blog/what-is-a-holdout-group-in-retention" rel="noopener noreferrer"&gt;what a holdout group is&lt;/a&gt;.&lt;/p&gt;

&lt;h2&gt;
  
  
  Where does this information come from?
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;  Stripe: &lt;a href="https://docs.stripe.com/billing/revenue-recovery/smart-retries" rel="noopener noreferrer"&gt;automate payment retries&lt;/a&gt;, &lt;a href="https://docs.stripe.com/billing/revenue-recovery/customer-emails" rel="noopener noreferrer"&gt;automate customer emails&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  Churn Buster: &lt;a href="https://churnbuster.io/" rel="noopener noreferrer"&gt;churnbuster.io&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  Baremetrics: &lt;a href="https://baremetrics.com/recover" rel="noopener noreferrer"&gt;baremetrics.com/recover&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  Churnkey: &lt;a href="https://churnkey.co/" rel="noopener noreferrer"&gt;churnkey.co&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  Slicker: &lt;a href="https://www.slickerhq.com/" rel="noopener noreferrer"&gt;slickerhq.com&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;  Retivo: &lt;a href="https://retivo.digital/" rel="noopener noreferrer"&gt;retivo.digital&lt;/a&gt; and this blog&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Want your own numbers?
&lt;/h2&gt;

&lt;p&gt;The Revenue Leak Report puts failed payments, unconverted trials, cancellations and downgrades in dollars per month, read from a restricted read-only Stripe key. $490, delivered within 48 hours of receiving the key. Nothing in your account is changed.&lt;/p&gt;

&lt;p&gt;&lt;a href="https://retivo.digital/sample-report" rel="noopener noreferrer"&gt;See a sample (synthetic data)&lt;/a&gt;&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://retivo.digital/blog/best-tools-to-recover-failed-stripe-payments" rel="noopener noreferrer"&gt;retivo.digital&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

&lt;p&gt;Retivo recovers failed Stripe payments and reduces involuntary churn for subscription businesses. &lt;a href="https://retivo.digital" rel="noopener noreferrer"&gt;retivo.digital&lt;/a&gt;&lt;/p&gt;

</description>
      <category>saas</category>
      <category>stripe</category>
      <category>payments</category>
      <category>startup</category>
    </item>
    <item>
      <title>Best AI Automation Agencies for Small Businesses: How to Choose</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Wed, 11 Nov 2026 13:00:09 +0000</pubDate>
      <link>https://dev.to/2pizza/best-ai-automation-agencies-for-small-businesses-how-to-choose-11oi</link>
      <guid>https://dev.to/2pizza/best-ai-automation-agencies-for-small-businesses-how-to-choose-11oi</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;TL;DR: There is no single answer to who to hire, because agencies differ on region, industry, size of client and what you end up owning. Pick by those four things first. Aivy fits Australian mid-market firms in six named sectors. The Crunch fits Malaysian, Singapore and Hong Kong businesses that want a WhatsApp agent. AY Automate fits larger teams that want an embedded engineer. NextAutomation is built for commercial real estate. We fit founders who want a fixed-scope build they own. I founded 2pizza.team, so weigh my entry accordingly. Everything about the others comes from their own websites, read on 9 November 2026.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Founders ask me this every week: which AI automation agency should I use? I run one, so I am a biased source, and I would rather say that in the first paragraph than have you find out in the last. What I can offer is a way to sort the options and a straight description of each, including ours, with the limits I know about.&lt;/p&gt;

&lt;p&gt;One rule for how I wrote this. For every agency below I used only what its own site says, and where the site is silent I wrote that it does not say. I do not quote their prices, and I do not rate their quality, because I have not worked with them and a ranking from me would be a guess. If you spot something that has since changed on their site, their site wins.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do I choose an AI automation agency for a small business?
&lt;/h2&gt;

&lt;p&gt;Start with four filters, in this order: where the agency works, which industries it names, how it scopes and prices a first project, and what you own when it finishes. Skills matter less than you expect at this stage, because most agencies in this space build on the same handful of tools. The filters decide whether the engagement can work at all.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The four filters&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Region and law. If you hold customer data under a specific privacy regime, check whether the agency says it builds for it. Some state this on the site, some do not.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Industry. A team that names your sector has probably seen your documents and your messy inputs before. A generalist can still do the work, it just spends the first weeks learning your vocabulary.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;First project shape. Look for a fixed scope with a named deliverable and a rough timeline. Open-ended hourly work is harder to judge from the outside.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Ownership and handover. Ask who holds the accounts, the workflows and the documentation when the project ends, and how changes get made afterwards.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What are the main options?
&lt;/h2&gt;

&lt;p&gt;Seven, by my count: four named agencies, our own studio, and two routes with no agency at all. I have put ours in the middle of the list on purpose, since the order is not a ranking. Each entry says who it fits, what its own site shows, and where it may not suit you.&lt;/p&gt;

&lt;h3&gt;
  
  
  1. Aivy (Melbourne, Australia)
&lt;/h3&gt;

&lt;p&gt;Aivy describes itself as an Australian AI automation agency based in Kensington, Victoria, with a Sydney offering and national delivery. Its site names six sectors: accounting and advisory, e-commerce and retail, healthcare, NDIS providers, professional services and real estate. It lists AI agents across web, email, WhatsApp and voice, finance automation, document processing and consulting.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fits: mid-market Australian businesses in those sectors, especially where Australian data handling matters. The site says it is designed around the Australian Privacy Act 1988 and keeps data in Australia.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;How it works, per its site: a free first conversation, an architecture study with a fixed-scope proposal, one pilot workflow live in about four to six weeks, then handover with documentation, training and SLA-backed support.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Integrations it names: Xero, MYOB, HubSpot, Salesforce, Shopify, WooCommerce, Stripe, WhatsApp Business, plus Anthropic, OpenAI, Azure and AWS.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Limits: the site targets mid-market firms, so a one-person business should check whether a pilot fits its budget. Ownership terms are not stated on the page I read.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  2. The Crunch (Malaysia)
&lt;/h3&gt;

&lt;p&gt;The Crunch says it is an AI chatbot and automation agency in Malaysia, operating since 2019, serving Malaysia, Singapore and Hong Kong. Its offer is a set of agents: customer service, sales and lead qualification, voice for phone calls, CRM automation and marketing content.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fits: small and mid-sized businesses in the region that want WhatsApp, Instagram, Facebook or website chat answered by an agent. The site says it uses the official WhatsApp Business API and that the client keeps their number.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;How it works, per its site: a free 30-minute demo call, tool selection from platforms it has tested, training the agent on your products, prices and FAQs with your approval of replies, then go-live. Mid-sized scope is stated as about 30 days.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Languages stated: English, Bahasa Malaysia, Mandarin and Cantonese. It says it builds to Malaysia's PDPA.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Limits: the focus is customer-facing agents, so back-office workflow automation is not what the site leads with.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  3. 2pizza.team (ours)
&lt;/h3&gt;

&lt;p&gt;I founded this studio. We are a five-person team building on Claude, and our site says we have 28 AI systems in production across e-commerce, finance, logistics, healthcare and SaaS. We build workflows in Make, n8n or code, AI agents, document processors, voice bots and RAG systems, and larger products where AI is the core. We also build white label under an agency's brand.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fits: founders and small ops teams with one painful, repetitive process, and growing teams with several. Pricing on our site: fixed-price scopes from $1,000 and monthly retainers from $1,500, with no hourly billing.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;How it works: you pick a tier, we ship a defined deliverable with error alerts and a handover session, and the larger tiers include a support window. Hosting is set up by us and owned by you.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Limits: we are small, so we take a limited number of projects at once. We have no industry specialism the way Aivy or NextAutomation do. We are not an on-site embedded team. If you need someone in your office five days a week, look at the next entry.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  4. AY Automate
&lt;/h3&gt;

&lt;p&gt;AY Automate describes itself as an AI transformation and automation agency. Its site says it has shipped work for more than 60 companies and lists three tracks: automate (agents, workflows, retrieval pipelines, maintenance), build (SaaS MVPs, security audits of Claude Code projects) and grow (engineer placement, strategy, workshops). It also offers free tools such as a readiness assessment and an ROI calculator.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fits: larger teams, in particular US and European enterprises as its site frames it, that want an embedded engineer who maps how the work happens and builds around it.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;How it works, per its site: a free 30-minute strategy call and audit, strategy in week two, implementation in roughly weeks three to eight, then ongoing optimisation. It states that month one is refundable if it does not deliver.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Integrations it names: n8n, Make, Zapier, Claude Code, MCP, HubSpot, Salesforce, Slack, Stripe, Shopify and others.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Limits: its location is not stated on the page I read, and the enterprise framing means a very small business should ask whether the engagement is sized for it.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  5. NextAutomation (commercial real estate)
&lt;/h3&gt;

&lt;p&gt;NextAutomation presents itself as an agentic AI sourcing platform for commercial real estate, configured around each firm. It covers off-market deal sourcing, owner resolution, underwriting and deal documents for investors, brokers, lenders and developers. Its site says tools, data access, AI providers and deployment are agreed before implementation, and offers managed, cloud or on-premises deployment.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fits: commercial real estate firms. If that is you, a vertical specialist knows your documents and workflow already.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Limits: it is a poor match outside that sector. Pricing, location and who owns what is built are not stated on the page I read.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h3&gt;
  
  
  6. A freelancer or a marketplace builder
&lt;/h3&gt;

&lt;p&gt;One person who builds in Make, Zapier or n8n is the cheapest route to a single working automation, and for a simple, well-defined task it is often the right one. The risks are practical. A single person is a single point of failure when they move on, and documentation is whatever they chose to leave. Ask for the workflow export, a written description and access under your own accounts before you pay the final instalment.&lt;/p&gt;

&lt;h3&gt;
  
  
  7. Build it yourself
&lt;/h3&gt;

&lt;p&gt;If you or someone on your team enjoys tinkering, the no-code tools are approachable and you will understand your own process better afterwards. This works well for one or two linear workflows. It gets expensive in attention when the process has exceptions, when an AI step needs checks on its output, or when the person who built it leaves. I wrote more about the platform choice in our Make vs Zapier vs n8n comparison and about the cost side in AI automation vs hiring.&lt;/p&gt;

&lt;h2&gt;
  
  
  Which one fits which kind of small business?
&lt;/h2&gt;

&lt;p&gt;Match on region and sector first, and size second. If you are in Australia in one of Aivy's named sectors, start there. If you are in Malaysia, Singapore or Hong Kong and your pain is customer messages, start with The Crunch. If you are in commercial real estate, NextAutomation is the specialist. If you want an embedded engineer and have a larger team, look at AY Automate.&lt;/p&gt;

&lt;p&gt;If none of those describes you, you are probably the person we are built for: a founder or ops lead with a specific, repetitive process, outside a regulated niche, who wants a fixed price and the keys at the end. If you are outside that, I will tell you on the first call and point you to a better fit, which is a thing I have done before.&lt;/p&gt;

&lt;h2&gt;
  
  
  What should I ask before signing with any agency?
&lt;/h2&gt;

&lt;p&gt;Five questions, and the answers matter more than the pitch. They apply to us as well, and you should ask them of us.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;What exactly is delivered, by when, for a fixed amount? Ask for it in writing.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Whose accounts do the workflows, API keys and data live in?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;What happens when a step fails at 2 a.m.? Who is alerted, and who fixes it?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;If an AI model is in the loop, how are its outputs checked, and what caps the spend on model calls?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;What does it cost to change something a month after launch?&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  When should I not hire an agency at all?
&lt;/h2&gt;

&lt;p&gt;When the work is not repetitive. Automation pays on consistent inputs and rules you can write down. If every case needs judgment, or the process changes weekly, a person is the better answer, and an agency will take your money to prove it. If you have not yet done the task by hand enough times to describe it, do that first.&lt;/p&gt;

&lt;p&gt;A second case: you cannot name an owner. Every automation needs someone inside your business who reads the alerts and decides when the rules change. Without that person, even a well-built system drifts. Agencies, ours included, can maintain what we built, but we cannot decide what your business should do.&lt;/p&gt;

&lt;h2&gt;
  
  
  Frequently asked questions
&lt;/h2&gt;

&lt;h3&gt;
  
  
  How long does a first AI automation project take?
&lt;/h3&gt;

&lt;p&gt;It depends on scope. The agencies above state their own figures: Aivy says a pilot goes live in about four to six weeks, The Crunch says about 30 days for mid-sized scope, and AY Automate describes implementation in roughly weeks three to eight. Treat these as their claims and ask for a date in the proposal.&lt;/p&gt;

&lt;h3&gt;
  
  
  Do I need an agency in my own country?
&lt;/h3&gt;

&lt;p&gt;Not always, but check data rules. If you hold personal data under a regime like Australia's Privacy Act or Malaysia's PDPA, prefer an agency that states it builds for it, or ask in writing how it handles that data. For internal back-office processes with little personal data, location matters less.&lt;/p&gt;

&lt;h3&gt;
  
  
  Is it cheaper to use a freelancer than an agency?
&lt;/h3&gt;

&lt;p&gt;Often at the start, yes. The cost shows up later if the freelancer disappears or the workflow was never documented. Compare the total over a year, including who maintains it, not just the build price.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://2pizza.team/blog/best-ai-automation-agencies-small-business" rel="noopener noreferrer"&gt;2pizza.team&lt;/a&gt;. We build AI and automation systems for small teams - fixed price, two to six weeks. &lt;a href="https://2pizza.team/work" rel="noopener noreferrer"&gt;See the work&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>automation</category>
      <category>ai</category>
      <category>startup</category>
    </item>
    <item>
      <title>What Is a Company Brain? A Definition From Someone Who Runs One</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Tue, 10 Nov 2026 13:00:09 +0000</pubDate>
      <link>https://dev.to/2pizza/what-is-a-company-brain-a-definition-from-someone-who-runs-one-cb6</link>
      <guid>https://dev.to/2pizza/what-is-a-company-brain-a-definition-from-someone-who-runs-one-cb6</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;TL;DR: a company brain is persistent memory of the calls, chats, documents and tasks your company already produces. It answers questions with a source you can open, logs decisions with a date and an owner, checks task follow-ups against what actually happened, and sends the owner one brief a week. It is useful for owners whose decisions happen on calls and whose team is remote. It is a waste for a team of five in one room.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Most of what is written about a company brain is either a vendor page or a prediction. I have run one every day since early October 2026, inside one of the companies we operate, so this is a definition from the operator side: what it is, what it costs to keep alive, and where it does not help.&lt;/p&gt;

&lt;p&gt;I will also say what it gets wrong. Answers can be wrong. That is the reason every answer shows its source, and it is the reason I trust the system at all.&lt;/p&gt;

&lt;h2&gt;
  
  
  What is a company brain?
&lt;/h2&gt;

&lt;p&gt;A company brain is a persistent memory built from what your company already says and writes: call recordings and transcripts, team chats, documents and the task board. On top of that memory sit three jobs. It answers questions and shows where the answer came from. It keeps a record of what was decided. It keeps people on track by comparing what was promised with what happened.&lt;/p&gt;

&lt;p&gt;The word persistent matters. A chat with a general AI tool forgets everything when you close the tab. A brain remembers that on a call three weeks ago someone agreed to send a revised quote by Friday, and it can tell you whether that happened.&lt;/p&gt;

&lt;p&gt;The word brain is a bit grand. In practice it is a database, a set of connectors, a retrieval layer, an AI model on top, and a set of rules about who can see what. The value comes from the connectors and the rules, not from the model.&lt;/p&gt;

&lt;h2&gt;
  
  
  What goes into a company brain?
&lt;/h2&gt;

&lt;p&gt;Four kinds of input, all things that exist already. Nobody has to start writing anything new.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Call recordings and transcripts: this is where most decisions in a remote company are actually made&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Team chats: the running commentary, the small agreements, the questions that never reached a document&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Documents: plans, specs, contracts, notes, whatever the team keeps&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;The task board: what is assigned, to whom, and what state it is in&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;These four are what we connected in the company where we run it. Calls matter most. If you only connect one source and you run a distributed team, connect the calls.&lt;/p&gt;

&lt;h2&gt;
  
  
  What comes out of it?
&lt;/h2&gt;

&lt;p&gt;Four outputs. Each one replaces a chore that someone currently does badly or not at all.&lt;/p&gt;

&lt;h3&gt;
  
  
  Answers with sources
&lt;/h3&gt;

&lt;p&gt;You ask what was agreed with a supplier, or why a price changed, and you get an answer with a link to the call, message or document it came from. I treat the source as the real answer and the sentence as a summary of it. When the sentence looks off, I open the source, and sometimes the sentence is the part that is wrong.&lt;/p&gt;

&lt;h3&gt;
  
  
  A decisions log with a date and an owner
&lt;/h3&gt;

&lt;p&gt;Decisions get pulled out of calls and written down with the date and the person responsible. This sounds minor until the third time someone asks why you chose a vendor and nobody remembers. It also ends the quiet argument about whether something was ever decided.&lt;/p&gt;

&lt;h3&gt;
  
  
  Task follow-ups checked against facts
&lt;/h3&gt;

&lt;p&gt;A task marked done is a claim. A brain can set that claim next to the calls and chats around it and flag the gap. A person saying the work is done and the transcript showing the client still waiting are two different facts, and the owner usually only hears the first one.&lt;/p&gt;

&lt;h3&gt;
  
  
  One weekly owner brief
&lt;/h3&gt;

&lt;p&gt;A single summary for the owner: what was decided, what is overdue, what changed. One page, once a week, built from the same sources. It replaces the status round where five people each tell you their version.&lt;/p&gt;

&lt;h2&gt;
  
  
  How is a company brain different from a wiki?
&lt;/h2&gt;

&lt;p&gt;A wiki depends on people writing into it. A brain reads what the company already produces. That one difference decides which of the two is still accurate six months in.&lt;/p&gt;

&lt;p&gt;Every team I know has started a wiki with good intentions. It is current for about a month, then it drifts, because writing documentation is nobody's job and the real decisions keep happening on calls. A wiki records what someone had time to write down. A brain records what was said.&lt;/p&gt;

&lt;p&gt;This does not make a wiki worthless. A clean, written policy is better than a transcript for anything that must be exact. The brain can read your wiki too, as one more document source.&lt;/p&gt;

&lt;h2&gt;
  
  
  How is it different from enterprise search?
&lt;/h2&gt;

&lt;p&gt;Search finds documents. A brain also tracks what was decided and whether it happened. Those are different jobs.&lt;/p&gt;

&lt;p&gt;Enterprise search is good at the question: where is the file about this. If you type a topic, you get a list of places to read. You still do the reading, the reconciling and the remembering. A brain is built around the question: what is the current state of this, who owns it, and did it get done. That needs a record of decisions and a link between a decision and a task, which a search index does not keep.&lt;/p&gt;

&lt;p&gt;If your only problem is that people cannot find files, you do not need a brain. Fix search first.&lt;/p&gt;

&lt;h2&gt;
  
  
  How is it different from a personal second brain?
&lt;/h2&gt;

&lt;p&gt;A personal second brain is one person's notes. It has no permissions, because there is only one reader, and no conflicting sources, because the author is the only source. A company has neither luxury.&lt;/p&gt;

&lt;p&gt;In a company, two people describe the same decision differently. A call from March says one thing and a chat from May says another. Finance data should be visible to the owner and not to everyone. A company brain has to handle both: it has to show where sources disagree instead of quietly picking one, and it has to know who is allowed to ask what. Those two requirements are most of the engineering.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who needs a company brain?
&lt;/h2&gt;

&lt;p&gt;Three kinds of owner, in my experience. They share one trait: the decisions happen somewhere the owner cannot easily re-read.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;The owner of a remote or distributed team, where decisions happen on calls and the only record is whatever someone remembered to write&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;The investor-owner who is not in the daily work and needs to know what is going on without sitting in every meeting&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;The owner running several businesses, who cannot hold the context of each one in their head and keeps asking the same questions in different companies&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;What these have in common is a gap between what happens and what the owner knows. The brain narrows that gap. It does not remove it, and anyone who promises otherwise is selling you something.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who should not buy one?
&lt;/h2&gt;

&lt;p&gt;Two clear cases, and I would rather tell you now than after the setup.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;A team of five in one room. You already share the context by being there. A brain would summarise what everyone heard an hour ago.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;A company that does not record calls or keep a task board. The brain reads what exists. If the calls are not recorded and the work is not tracked, there is nothing to read, and the first project is getting those habits in place, not buying software.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;There is a third case I would add: an owner who does not want to look at the output. The weekly brief only works if someone reads it and acts on it.&lt;/p&gt;

&lt;h2&gt;
  
  
  What do you need to start?
&lt;/h2&gt;

&lt;p&gt;Four things. The software is the easy part of the list.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Access to the sources: calls, chats, documents and the task board, with someone who can grant it&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;One person who knows how the company really works, who can say that this channel is noise and this recurring call is where pricing gets decided&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;A decision on who sees what, made in advance, with finance kept for the owner&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Consent from the people on recorded calls, asked for plainly and recorded as given&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The third item is where setups stall. Permissions are an owner decision, and they are easier to set at the start than to tighten after people have seen things they should not have.&lt;/p&gt;

&lt;p&gt;The fourth item is not optional. Putting a person's words into a system that anyone can query is a real change for them. Tell them, and make it a yes they can give freely.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do people actually use it?
&lt;/h2&gt;

&lt;p&gt;Through the tools they already use, with their own login and access that matches their role. If people have to open a new app to ask a question, they stop asking within a week.&lt;/p&gt;

&lt;p&gt;In the company where we run it, people reach it from Telegram, from WhatsApp, from a web chat, and through MCP from their own AI tools. MCP is the standard that lets an AI assistant you already use connect to an outside source, so the brain shows up inside the assistant instead of as a separate destination. Each person sees what their role allows and nothing more.&lt;/p&gt;

&lt;h2&gt;
  
  
  What does it cost to run, and what goes wrong?
&lt;/h2&gt;

&lt;p&gt;Two things surprised me. One is a cost I measured, the other is a cost I did not expect.&lt;/p&gt;

&lt;p&gt;The measured one is model spend. Over 20 days in one company, chat on a top model cost about $0.08 per answer. We then tested a smaller model on 30 real questions from the team. It answered at about $0.006 per answer, against $0.12 on the larger model in the same test. So routine chat moved to the smaller model, and anything that touches decisions stayed on the stronger one. Thirty questions is a small sample, and I would run it again on your questions before trusting my result for your company.&lt;/p&gt;

&lt;p&gt;The unexpected one is approval. Decisions extracted from calls go in as proposals, and a human approves them before they count. That is the right design, because a model reading a transcript will sometimes call a hypothetical a decision. But within two weeks the unapproved proposals had piled up into the hundreds. Approval has to be somebody's job, with time set aside for it, or the decisions log becomes a backlog nobody trusts.&lt;/p&gt;

&lt;p&gt;Beyond that, the failure I watch for is the confident wrong answer. The defence is the source. If a team stops opening sources, the brain gets less reliable without anyone noticing, so I treat the habit of checking as part of the system.&lt;/p&gt;

&lt;h2&gt;
  
  
  Should you build it yourself or have it set up?
&lt;/h2&gt;

&lt;p&gt;You can build one. The pieces exist and a capable developer can connect them. The work that takes time is not the first demo, it is the permissions, the approval routine, the model choices and the weeks of correcting what it gets wrong.&lt;/p&gt;

&lt;p&gt;We offer it as a service: we set it up with your team, then we run it together with you. We pick the sources with you, set the access rules, train the approval routine, and keep tuning it as the questions change. If you have an engineer with spare capacity and a clear owner for the project, building in-house is a reasonable choice, and I would say so on a call.&lt;/p&gt;

&lt;p&gt;If you are not sure whether your company is the kind that benefits, the quickest test is a question: when did you last wish you could re-listen to a call from last month? If the answer is this week, you probably have the problem. If the answer is never, you probably do not.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://2pizza.team/blog/what-is-a-company-brain" rel="noopener noreferrer"&gt;2pizza.team&lt;/a&gt;. We build AI and automation systems for small teams - fixed price, two to six weeks. &lt;a href="https://2pizza.team/work" rel="noopener noreferrer"&gt;See the work&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>automation</category>
      <category>ai</category>
    </item>
    <item>
      <title>What Breaks When You Run a Company Brain on Real Calls</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Mon, 09 Nov 2026 13:00:09 +0000</pubDate>
      <link>https://dev.to/2pizza/what-breaks-when-you-run-a-company-brain-on-real-calls-2dno</link>
      <guid>https://dev.to/2pizza/what-breaks-when-you-run-a-company-brain-on-real-calls-2dno</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;TL;DR: AI meeting notes fail in boring, predictable ways: the same call imported twice, early proposals logged as decisions, the wrong person credited, old decisions never retired, a bot that goes silent with no alert. Every one of these is fixable, and none of them shows up in a demo. Below are the nine we hit in a month on real calls, the fix for each, and what to demand from any vendor before you trust it with your decisions.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;A company brain is a system that listens to your calls and chats, pulls out decisions and tasks, and answers questions like "what did we agree about the supplier" months later. In a demo it looks finished. On real calls, with real audio, three note-taker bots in one meeting and a team that talks over each other, it breaks in specific places.&lt;/p&gt;

&lt;p&gt;I run an automation studio and we sell this, so discount accordingly. I am writing the failure list anyway because it is the most useful thing I can give a buyer. Everything below happened in one of the companies we run it in, over about a month. I am not naming the company or the people.&lt;/p&gt;

&lt;p&gt;If you are searching for AI meeting notes accuracy, an AI decision log from meetings, or the usual AI knowledge base problems, this is the list I wish someone had handed me before we started.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why did the same meeting show up twice?
&lt;/h2&gt;

&lt;p&gt;Because three note-taker bots joined one call and each produced its own transcript. Our import tried to detect duplicates by matching speaker names. One of the three copies had only "Speaker 1" and "Speaker 2", so the names never matched and both copies went in.&lt;/p&gt;

&lt;p&gt;The result is a knowledge base that counts one decision twice, and later answers that sound extra confident because two sources agree. They are the same source.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fix and vendor question&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fix: same time window means same meeting. Match on when the call happened, not on who the transcript says was there. Keep the best copy and link the others to it.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Ask the vendor: what happens when two tools record the same call? Ask them to show you, not describe it.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Why do early proposals end up logged as decisions?
&lt;/h2&gt;

&lt;p&gt;Because long transcripts get cut into pieces before analysis. Ours were split at about 30,000 characters and each piece was analysed on its own. Someone floated a proposal early in the call and the group settled on something different near the end. The first piece logged the proposal as a decision, the last piece logged the real one, and the system stored both side by side.&lt;/p&gt;

&lt;p&gt;Each piece was read correctly. The mistake was never reading the whole call as one conversation. Calls are not a list of independent paragraphs. The last word usually overrides the first.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fix and vendor question&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fix: a reconciliation pass over the whole call after the pieces are done. It looks at every extracted decision together and drops or marks the ones that a later part of the call replaced.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Ask the vendor: how do you handle a call longer than the model can read in one go? If the answer is "we chunk it", ask what happens after the chunks.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Can transcription change what was said?
&lt;/h2&gt;

&lt;p&gt;Yes, and it did. The transcription service translated English speech into another language and made errors in the process. A weekday changed. A product term turned into a word with an unrelated meaning. Nobody noticed until someone asked when a meeting was and got the wrong day.&lt;/p&gt;

&lt;p&gt;Logistics are where this hurts. Days, dates, names and amounts are exactly what people ask a brain for, and exactly what a translation step can quietly corrupt.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fix and vendor question&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fix: keep the original language end to end, and forbid the model from guessing logistics. If a day or time is not clearly in the transcript, the answer says it is not clear.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Ask the vendor: is the transcript stored in the language spoken? Does the system ever translate before extracting? What does it do when a date is ambiguous?&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Why did the system credit decisions to the wrong person?
&lt;/h2&gt;

&lt;p&gt;In our case the owner of the company was credited with decisions on calls he was not on. Another participant talks like a decision maker, confident and in the imperative, and the model attributed the authority to the person it knew had the authority. A passing remark also turned into a task that nobody had agreed to.&lt;/p&gt;

&lt;p&gt;This is the failure with the highest cost, because a wrong attribution reads as fact. A status report that says "the owner decided X" ends arguments. It should start one.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fix and vendor question&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fix, part one: aliases per person, so the same human is recognised under every name, nickname and transcript label, and only people who were on the call can be credited.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Fix, part two: status reports for the owner are built from the transcript, not from the extracted task list. The task list is a guess. The transcript is the record.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Ask the vendor: how do you know who said something? Can I see the exact line in the transcript behind any decision or task? If every item does not link back to the words, walk away.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What happens to proposals nobody has approved?
&lt;/h2&gt;

&lt;p&gt;They pile up. At one point about 326 proposals were waiting for a human to approve or reject them. The search was ranking them alongside approved facts, so unapproved guesses leaked into answers and sounded just as authoritative as confirmed decisions.&lt;/p&gt;

&lt;p&gt;Human approval is a good control. A human approval queue nobody empties is not a control, it is a pile of unverified text sitting inside your search index.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fix and vendor question&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fix: proposals expire after 7 days. In the first pass 135 expired. Approved items rank higher than anything pending, so a guess can never outrank a confirmed fact.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Ask the vendor: is unreviewed content searchable? Is it labelled as unreviewed in the answer? What happens to items nobody reviews?&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  How does the brain know a decision was replaced?
&lt;/h2&gt;

&lt;p&gt;In our first version it did not. Decisions were stored and never marked as replaced when a newer one contradicted them. Ask about a price, a deadline or an owner and the answer blended the old position with the new one, because both were equally present and equally well phrased.&lt;/p&gt;

&lt;p&gt;Companies change their minds constantly. A knowledge base that only adds is a record of everything you ever believed, not of what you believe now.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fix and vendor question&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fix: superseded marking. When a new decision replaces an old one, the old one is flagged as replaced, kept for history, and left out of current answers.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Ask the vendor: when two decisions conflict, which one wins and how does the system know? Can I see the history of a decision, old and new?&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  What if the bot stops answering and nobody tells you?
&lt;/h2&gt;

&lt;p&gt;This happened. The model API key hit its spending limit and the bot went quiet for everyone for almost two days. There was no alert. People assumed it was slow, then assumed it was broken, then stopped asking. Trust in a tool like this is expensive to build and cheap to lose.&lt;/p&gt;

&lt;p&gt;The cost side matters too. Using the strongest model for every routine chat message is how you hit a limit like that. We measured both on 30 real questions: about $0.006 per answer with a cheaper model for routine chat, against about $0.12 per answer on the strong model.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fix and vendor question&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fix: an alert when the key approaches its limit, so a human hears about it before the users do, and a cheaper model for routine chat with the strong model kept for the hard questions.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Ask the vendor: what happens when the model provider fails or the budget runs out? Who is told, how fast, and what does the user see in the meantime?&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  Do scheduled checks run when nobody is at the computer?
&lt;/h2&gt;

&lt;p&gt;Not if they live on a laptop. We had "daily" health checks scheduled on a laptop. The app was closed for a week and the checks did not run. From the outside they looked alive, because the schedule was still there. They were dead.&lt;/p&gt;

&lt;p&gt;The dangerous version of a failure is the one that looks like success. A check that never runs reports nothing, and nothing looks the same as all clear.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Fix and vendor question&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Fix: run the checks on a server, and check the last-run time, not just whether a schedule exists. A check that has not run on schedule is itself a failure.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Ask the vendor: where do your monitoring jobs run? What alerts you if the monitor itself stops?&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  How do you test an AI knowledge base before you trust it?
&lt;/h2&gt;

&lt;p&gt;With a fixed set of real questions and a list of phrases each answer must not contain. We use 19 real questions taken from actual use. For each one we write down the facts the answer should include and, more important, the phrases it must never contain: the stale price, the wrong person, the guessed date.&lt;/p&gt;

&lt;p&gt;The strong model passes 19 of 19 on this set, at about $4.69 a run. That price is the point. A test this cheap can run after every change, and we do run it after every change. Most of the failures above would have been caught on day one by a question set like this, if we had had one on day one.&lt;/p&gt;

&lt;p&gt;Be careful with how you read your own score. Nineteen questions is a small set, and passing it means the known failures are covered. It does not mean the system is correct on everything. Every new failure becomes a new question, so the set grows with the bugs.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Vendor question&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Ask the vendor: what is your regression test? How many questions, how often does it run, and can I add my own?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Ask for the negative checks specifically. A test that only checks the right answer appears does not catch a wrong answer sitting next to it.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;h2&gt;
  
  
  When should you not buy a company brain?
&lt;/h2&gt;

&lt;p&gt;If your team has few calls, your decisions live in two documents and the owner makes them all, a shared folder and a weekly summary will serve you better. A brain earns its keep when decisions are scattered across many calls and many people, and when someone has to answer "what did we decide" regularly.&lt;/p&gt;

&lt;p&gt;Also hold off if nobody will own it. Someone has to approve proposals, notice when an answer is wrong and report it. Left alone, even a good system drifts, and the fixes above are what stop it drifting fast.&lt;/p&gt;

&lt;h2&gt;
  
  
  What should I ask any vendor before I buy?
&lt;/h2&gt;

&lt;p&gt;Print this and put it on the table. A vendor who has run the product on real calls will have answers. A vendor who has only run demos will start talking about features.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;Vendor checklist&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Duplicates: what happens when two or three tools record the same call?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Long calls: how do you handle a transcript longer than the model reads at once, and what happens after you split it?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Language: is the original language kept? Does anything translate before extraction? What happens with an ambiguous date?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Attribution: how do you know who said it? Can I see the transcript line behind every decision and task?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Aliases: how does one person map to several names and labels?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Unreviewed content: is it searchable, and is it labelled? What happens to items nobody reviews?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Replaced decisions: when two decisions conflict, which wins, and can I see the history?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Failure alerts: what happens when the model budget runs out or the provider fails, and who hears first?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Monitoring: where do scheduled checks run, and what tells you if the monitor itself stops?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Testing: what is your regression set, how often does it run, and can I add my own questions with phrases that must not appear?&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Cost: what does a routine answer cost, and what does a hard one cost?&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;None of these questions needs a technical background. They need a vendor who has watched the system fail and fixed it. If you want to talk it through, or want a second opinion on a tool you are already considering, I will tell you plainly whether it covers these. Ivan / 2pizza.team&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://2pizza.team/blog/company-brain-what-breaks" rel="noopener noreferrer"&gt;2pizza.team&lt;/a&gt;. We build AI and automation systems for small teams - fixed price, two to six weeks. &lt;a href="https://2pizza.team/work" rel="noopener noreferrer"&gt;See the work&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>automation</category>
      <category>ai</category>
    </item>
    <item>
      <title>Automated Cold Email: The Mistakes That Only Happen Once</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Sun, 08 Nov 2026 13:00:09 +0000</pubDate>
      <link>https://dev.to/2pizza/automated-cold-email-the-mistakes-that-only-happen-once-1j18</link>
      <guid>https://dev.to/2pizza/automated-cold-email-the-mistakes-that-only-happen-once-1j18</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;TL;DR: Automated cold email fails in ways you cannot undo, because a sent email stays sent. Our own campaigns taught us seven guardrails: read back what the sending tool stored, test the full sequence on your own address, check personalization as a batch, send only to verified addresses, cap volume on new domains, run a watchdog that pauses on bounces, and have someone other than the writer approve the copy. We also learned that for a cheap product, cold email lost to a warm channel.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;Most writing about cold email automation is about getting more replies. This post is about the other half: what goes wrong when you automate it, and what an owner should insist on before a system is allowed to send in their name.&lt;/p&gt;

&lt;p&gt;The reason this deserves its own post is simple. A bad landing page can be fixed in an hour and nobody remembers. A bad email batch lands in real inboxes, with your real name and your real domain on it, and no setting brings it back. Every mistake below happened to us, in campaigns we run for a client and for ourselves. I am not naming the client. The numbers are real.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why is automated cold email riskier than a normal marketing tool?
&lt;/h2&gt;

&lt;p&gt;Because it is irreversible and it is aimed at strangers. A newsletter goes to people who signed up. Cold email goes to people who did not, so each mistake costs you in three places at once: the person's opinion of you, your domain's reputation with mailbox providers, and your own pipeline for the next few months.&lt;/p&gt;

&lt;p&gt;The second thing that makes it risky is that the failures are quiet. A broken website throws an error. A broken email sequence just sends, on schedule, exactly as misconfigured. Nothing turns red. You find out when someone replies with a confused or angry message, or when you happen to look.&lt;/p&gt;

&lt;p&gt;That is the design problem. A safe system cannot rely on someone noticing. It has to check itself before it sends and keep checking while it runs.&lt;/p&gt;

&lt;h2&gt;
  
  
  What happens when a follow-up fires ten minutes after the first email?
&lt;/h2&gt;

&lt;p&gt;49 people got two emails in a row. The follow-up was supposed to go three days after the first message. It went ten minutes after.&lt;/p&gt;

&lt;p&gt;The cause was a misunderstanding about one field. In the sending tool we used, the delay on step N means the wait after step N, not the wait before it. We set the three-day gap on the follow-up step, as you would naturally read it. The tool read it as "wait three days after this follow-up", and the first step had a delay of almost nothing.&lt;/p&gt;

&lt;p&gt;Nobody caught it in the moment. We found it two days later, through a random check, not through any alarm. By then the 49 people had already received the same pitch twice within minutes, which reads as either a bug or a spam script. Both are bad for a first impression.&lt;/p&gt;

&lt;p&gt;Three fixes came out of it, and I would treat all three as the minimum for any system you let send:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;A preflight script that reads back what the tool actually stored, not what we typed in. It checks delays, the sending window, daily limits, links, the signature, empty fields and duplicates, and it fails the launch on any single error.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;A live test of the full sequence, sent to the owner's own address, so a human sees every step arrive in the real order and at the real timing before a stranger does.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;A guard that pauses a campaign automatically if one person receives two emails within two days, or if bounces go above 3%.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The point of the first one is worth stating directly. Reading back stored values matters because the bug lived in the gap between what we meant and what the tool understood. Checking your own settings screen would have shown the three days we intended. Checking the tool's stored state would have shown the problem.&lt;/p&gt;

&lt;h2&gt;
  
  
  Can AI personalized cold email still sound like a template?
&lt;/h2&gt;

&lt;p&gt;Yes, and it is the most common failure of AI personalized cold email. Each email looked fine on its own. Checked as a batch, 27% of the "personal observations" were the same template, and 20% were compliments.&lt;/p&gt;

&lt;p&gt;The cause was an example phrase in the prompt. We had included one sample observation to show the model the tone we wanted. The model treated it as the form to copy. Reading one email, you would never see it. Reading fifty in a row, the same sentence skeleton kept returning with a different company name dropped in.&lt;/p&gt;

&lt;p&gt;Recipients notice this faster than you would expect, because they get a lot of cold email and the pattern of a fake personal line is familiar. A generic compliment is, if anything, worse than no personalization. It tells the reader that someone automated the flattery.&lt;/p&gt;

&lt;p&gt;What we changed:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Check the whole batch, not single emails. Count repeated observations and repeated openings across the batch, and fail it if the same shape keeps appearing.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Require a word from the person's own profile in the observation. If the line could have been written about anyone, it does not pass.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;If the observation fails, drop the personalization line entirely. Do not insert a filler. A shorter email with no fake detail beats a longer one with a hollow detail.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The broader lesson for anyone buying or building AI personalization: be suspicious of examples in prompts. They look like guidance and they behave like templates.&lt;/p&gt;

&lt;h2&gt;
  
  
  Should you email every address on your list?
&lt;/h2&gt;

&lt;p&gt;No. Send only to addresses that verified as valid. In one client base of 5,764 addresses, 2,562 were valid, 2,494 were catch-all, 560 were invalid and 148 were unknown. Only the valid ones were sent.&lt;/p&gt;

&lt;p&gt;That is a lot of list to leave unsent, and it is the right call. Invalid addresses bounce, and bounces damage the sending domain. Catch-all domains accept any address, so you cannot tell whether a real person is behind it. Sending to them is a gamble on your reputation, and the downside lands on the mailboxes you need for the valid ones.&lt;/p&gt;

&lt;p&gt;Owners sometimes read this step as wasting a purchased list. I read it the other way: a list is not an asset until it has been checked, and a smaller verified list outperforms a larger dirty one on the only thing that matters, which is whether your emails keep reaching inboxes at all.&lt;/p&gt;

&lt;h2&gt;
  
  
  How many emails can a new domain send safely?
&lt;/h2&gt;

&lt;p&gt;Fewer than you want to. We kept the daily limit at 15 per campaign while the domains were about five weeks old.&lt;/p&gt;

&lt;p&gt;Mailbox providers watch new sending domains closely, and a young domain that suddenly sends in volume looks like exactly what it would be if it were abusive. The temptation is to ramp quickly because the list is ready and the calendar is empty. Slow is cheaper than a burned domain, because replacing a domain means replacing its history too, and the history is the thing that takes weeks.&lt;/p&gt;

&lt;p&gt;I will not give you a universal number, because the right limit depends on the domain, the mailboxes and the provider. The principle holds either way: set the cap low, write it down, and have the preflight check that the stored limit matches the intended one.&lt;/p&gt;

&lt;h2&gt;
  
  
  What should monitor a cold email campaign while it runs?
&lt;/h2&gt;

&lt;p&gt;A watchdog that does not depend on anyone remembering to look. Ours runs on a server, checks every mailbox and every campaign every 30 minutes, and pauses a campaign on 3 or more bounces, or on a bounce rate over 3% once 30 or more emails have been sent. It alerts to Telegram when it pauses something.&lt;/p&gt;

&lt;p&gt;Both thresholds exist for a reason. Three bounces catches a small campaign early, where percentages are meaningless. The percentage rule catches a larger campaign where three bounces would be normal noise. The 30-email minimum stops the percentage from firing on tiny samples.&lt;/p&gt;

&lt;p&gt;The watchdog would have caught the double-send bug within hours, not days. That alone justified building it.&lt;/p&gt;

&lt;p&gt;The second half of this is memory. We keep a lessons log, currently 26 incidents, and it is read before any new launch. Every one of the mistakes in this post is in it. Incidents that live only in someone's head get repeated by the next person, or by the same person six months later.&lt;/p&gt;

&lt;h2&gt;
  
  
  Who should approve the emails before they go out?
&lt;/h2&gt;

&lt;p&gt;Someone other than the person who wrote them. We run an independent QA step: the person who writes the emails does not approve them, and any FAIL blocks the campaign.&lt;/p&gt;

&lt;p&gt;This is the oldest idea in quality control and cold email teams skip it constantly, usually because a single marketer or founder owns the whole flow. The writer reads their own copy and sees what they meant. A second reader sees what is actually on the page, including the broken link, the wrong first name and the line that sounds like every other cold email.&lt;/p&gt;

&lt;p&gt;The rule that makes it work is that a FAIL is a hard stop, not a comment. If QA can be overruled by the writer under deadline pressure, it is decoration. If your outbound system keeps a human in the loop, make sure that human is not the author.&lt;/p&gt;

&lt;h2&gt;
  
  
  Does cold email work for every product?
&lt;/h2&gt;

&lt;p&gt;No, and our own numbers say so. For one low-priced product, cold email to 521 contacts produced zero interested replies in the first wave. A warm channel, a webinar invite, produced 53 registrations and a call request within 48 hours.&lt;/p&gt;

&lt;p&gt;I include this because every guardrail above assumes cold email is the right channel, and sometimes it is not. A cold email asks a stranger to spend attention on someone they have no reason to trust. That ask is easier to justify when the deal is large enough that a short conversation is worth it. For a cheap product, the buyer does the math the other way, and a warm invitation to something they can attend wins.&lt;/p&gt;

&lt;p&gt;My current view: cold email fits higher-priced B2B offers. For cheap products, warm channels won in our case. One result is not a law, and I would not build a strategy on a single campaign. But if someone sells you automated cold email for a low-priced product without asking what you charge, that is worth noticing.&lt;/p&gt;

&lt;h2&gt;
  
  
  What should an owner check before an outbound system sends anything?
&lt;/h2&gt;

&lt;p&gt;Ask whoever built it to show you these seven things. If any answer is "we will keep an eye on it", that is the gap.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;A preflight that reads back stored settings from the sending tool and fails the launch on any error.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;A live test of the whole sequence to your own address, with the real timing.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;A batch-level check for repeated openings and repeated observations, with a rule for what happens when personalization fails.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Address verification, with a stated policy on catch-all and unknown addresses.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;A low daily cap on new domains, written down, and checked by the preflight.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;An automatic pause on duplicates and on bounces, running on a schedule, with alerts that reach a person.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Independent approval, where the author cannot approve their own work and a FAIL stops the send.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;None of this is advanced. It is the boring layer between "the copy is written" and "the campaign is live", and it is the layer people skip because the first campaign looks fine without it. Ours looked fine too, right up to the point where 49 people got two emails in ten minutes.&lt;/p&gt;

&lt;h2&gt;
  
  
  What does it cost to build this properly?
&lt;/h2&gt;

&lt;p&gt;For a full outbound system with enrichment, sequencing, qualification and these guardrails, a build is typically $5,000-12,000 fixed price, depending on how many signal sources and sequences are in scope. After that, the running cost is the tools themselves, usually $150-500/month.&lt;/p&gt;

&lt;p&gt;I would not buy it if your product is low-priced and your buyers sign up on their own, or if you have no list worth emailing yet. In those cases the money is better spent on a warm channel first, and I will tell you that on the scoping call. If you do sell something where a single conversation is worth real money and you want outbound that will not embarrass you, then the guardrails are the part worth paying for, because the writing is the easy part.&lt;/p&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://2pizza.team/blog/automated-cold-email-mistakes" rel="noopener noreferrer"&gt;2pizza.team&lt;/a&gt;. We build AI and automation systems for small teams - fixed price, two to six weeks. &lt;a href="https://2pizza.team/work" rel="noopener noreferrer"&gt;See the work&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>automation</category>
      <category>ai</category>
    </item>
    <item>
      <title>Niche Landing Pages at Scale Without Getting Flagged as Doorways</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Sat, 07 Nov 2026 13:00:10 +0000</pubDate>
      <link>https://dev.to/2pizza/niche-landing-pages-at-scale-without-getting-flagged-as-doorways-5e</link>
      <guid>https://dev.to/2pizza/niche-landing-pages-at-scale-without-getting-flagged-as-doorways-5e</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;TL;DR: a doorway page is one that exists to catch a query and says the same thing as its neighbours. You avoid the problem by giving every page a different direct answer, real niche detail, and a script that proves no sentence repeats. For one client we shipped 20 query pages plus 3 comparison pages this way. The script found 712 distinct sentences across the 20 and zero repeats. Pages still need weeks to be indexed and ranked, and nobody can promise you positions.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;I get asked about programmatic SEO for small business roughly once a week, and the question underneath is always the same: can I make forty pages from one template without Google deciding I am spamming it. The honest answer is yes, if the pages are different where it matters, and no if only the city name or the keyword changes.&lt;/p&gt;

&lt;p&gt;This post is the format and the checks we use when we ship many query pages. It comes from a real job, a client with an AI video product (I will not name them), and from the checks we run on our own site. I am Ivan, I run 2pizza.team, and I will tell you where this approach is thin as well as where it works.&lt;/p&gt;

&lt;h2&gt;
  
  
  What counts as a doorway page?
&lt;/h2&gt;

&lt;p&gt;A doorway page is built to rank for a specific query and then funnel the visitor to the same destination, with content that adds nothing the neighbouring pages do not already say. The usual tell is a template where a noun is swapped. Fifty pages titled with fifty cities and the same four paragraphs underneath are doorways, however clean the design.&lt;/p&gt;

&lt;p&gt;So the useful test is not how many pages you publish. It is whether each page answers its own question better than the page next to it. Everything below is a way of making that test mechanical, so you are not relying on your own judgment after the fortieth page.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do you choose the queries for niche landing pages?
&lt;/h2&gt;

&lt;p&gt;Pick queries real buyers type, not queries you wish they typed. For the client job we took 20 buyer queries from search autosuggest, one page per query. Autosuggest is crude, but it shows phrasing that people actually use, which beats a keyword tool export you cannot sanity-check.&lt;/p&gt;

&lt;p&gt;We also shipped 3 comparison pages alongside the 20. Comparison queries have a different intent from how-to queries, so they got their own pages instead of being bolted onto the others. I come back to those below, because they are where our checks caught a real mistake.&lt;/p&gt;

&lt;h2&gt;
  
  
  What should every niche page contain?
&lt;/h2&gt;

&lt;p&gt;The same section order on every page, with different content inside it. A fixed order is good for the reader, who learns where to look, and good for you, because a checklist is easier to audit than a free-form page. Here is the order we used:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Hero. The first sentence answers the query directly. If the query is a question, the first sentence is the answer, not a welcome.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Before and after proof from the real product. An actual result, not a stock image or a mockup.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Three steps. How the buyer gets from where they are to the result.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Price block. The real price, stated plainly.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Five niche questions. Questions that only someone in that niche would ask.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Final call to action. One clear next step.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;We built every page from existing design components and added no new ones. That is a rule about risk. A new component on a page set is a new thing that can break on twenty pages at once, and it is the kind of thing that gets invented on page seven because page seven felt different. Constraining the design forces the differences into the words, which is where they belong.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do you make pages different enough to avoid doorway pages?
&lt;/h2&gt;

&lt;p&gt;Write the anti-doorway rule down before you write a single page, and make it checkable. Ours had three parts: a different direct answer on every page, at least 3 niche-specific practical FAQ answers on every page, and no sentence copied between pages except price facts and button labels.&lt;/p&gt;

&lt;p&gt;The first part is the hardest to fake. If you cannot write a different first sentence for the hero, the two queries are probably one query and you should merge the pages. A different direct answer means a different claim about what to do, not the same advice with a new noun.&lt;/p&gt;

&lt;p&gt;The second part is where most template sites fail. The five questions on each page are the longest stretch of unique text, and they are where niche knowledge shows. A practical answer names the real constraint of that buyer: the format they need, the size of file they work with, the thing they worry about. At least 3 of the 5 had to be of that kind. The other two could be general product questions, because some questions are genuinely shared.&lt;/p&gt;

&lt;p&gt;The third part is the one a machine can enforce, which is why it is the one people skip. Price facts and button labels are the only allowed repeats, because they must be identical. A price that is worded three different ways across three pages is a bug waiting to happen.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do you check that no sentence repeats across pages?
&lt;/h2&gt;

&lt;p&gt;With a script, run on every build, that splits each page into sentences and fails if any sentence appears on more than one page outside the allowed list. For the client job it reported 20 pages, 712 distinct sentences, no repeats.&lt;/p&gt;

&lt;p&gt;I want to be careful about what that number proves. It proves the pages do not share wording. It does not prove the pages are good, and it does not prove a search engine will agree they are different. A page can be word-unique and still say nothing. The script is a floor, not a verdict.&lt;/p&gt;

&lt;p&gt;What the script does well is catch drift. When you write twenty pages over a few days, you reuse a good line without noticing. A human reviewer reading page nine does not remember page two. The script remembers all twenty, every time, and it is cheap enough that nobody has an excuse to skip it.&lt;/p&gt;

&lt;p&gt;The weakness is that exact-sentence matching misses paraphrase. If page three says a thing and page eleven says the same thing with two words changed, the script passes both. We cover that gap with the next check, the angle assignment for near-duplicate pairs, which is a human decision made before writing, not a test run after.&lt;/p&gt;

&lt;h2&gt;
  
  
  What do you do with near-duplicate queries?
&lt;/h2&gt;

&lt;p&gt;Decide the different angle for each one in advance and write it down. Some pairs of queries sit very close together, and if you let two writers or two prompts run on them, you get the same page twice. We give each page in a close pair an explicit, different angle.&lt;/p&gt;

&lt;p&gt;Our example from the job: one page covered video for a single listing, the one-off case where someone has one property or one product to show. The neighbouring page covered ongoing marketing for an agent, where the same person needs video again and again. Same product, same buyer type, different job. The first page talks about a single result and a short turnaround. The second talks about volume and consistency over time.&lt;/p&gt;

&lt;p&gt;If you cannot state the angle difference in one sentence, merge the pages. A near-duplicate pair without a distinct angle is the closest thing to a doorway you can build by accident.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do you stop wrong numbers and risky claims on many pages?
&lt;/h2&gt;

&lt;p&gt;Keep one facts file, and make the build fail when a page contains a number that is not in it. Every number on a page must exist in that file. A copy linter checks this on every build and refuses to ship otherwise.&lt;/p&gt;

&lt;p&gt;This paid for itself on the comparison pages. Before launch the linter flagged an outdated price on them. The price had changed in the facts file and the older figure had survived in the comparison copy, where nobody was looking. On a single page you would catch that by eye. Across 23 pages you will not, and a wrong price on a comparison page is the sort of error that makes a buyer stop trusting the rest.&lt;/p&gt;

&lt;p&gt;The same linter holds a list of banned claims. For this client the list was:&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;No-cost claims. Nothing on the page says the product costs nothing.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Superlatives and rank-one claims, of any wording.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Platform-approval claims, such as saying a marketplace approved the product.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Features the product does not have.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;The last item sounds silly until you are writing the twentieth niche page and the niche really wants a feature. A page about one buyer type invites you to promise what that buyer wants. The linter is there to stop the copy from writing a cheque the product cannot cash. Approval claims get the same treatment because they are the easiest to get wrong and the hardest to take back once published.&lt;/p&gt;

&lt;h2&gt;
  
  
  What do you check after the pages go live?
&lt;/h2&gt;

&lt;p&gt;Check that they are reachable, linked, distinct and discoverable, then watch them over weeks. The same automation we use for client sites runs on our own, so here are the real numbers from 2pizza.team.&lt;/p&gt;

&lt;p&gt;At the 8 September 2026 baseline, our sitemap held 178 pages, all returning 200, with zero orphan pages and zero duplicate titles or descriptions. Weekly automated checks cover ranks, backlink mentions across 43 sites, an on-page crawl of every sitemap page, and an IndexNow ping to tell search engines something changed.&lt;/p&gt;

&lt;h2&gt;
  
  
  Can you trust the tracker that tells you if it worked?
&lt;/h2&gt;

&lt;p&gt;Only if it can tell the difference between zero and could not check. Our first baseline said zero positions. That was wrong. The free search endpoint we used was blocking our server on all 37 queries, so every check returned nothing and the tracker recorded that as zero.&lt;/p&gt;

&lt;p&gt;I am telling you because it is the most expensive kind of mistake in this work: a measurement that looks like a result. If we had believed it, we would have drawn conclusions from a number that meant our own server was refused. The tracker now marks a run as blocked instead of recording zero. A blocked run is a hole in the data, and a hole is honest.&lt;/p&gt;

&lt;p&gt;The backlink scan has a similar starting point, and I will state it plainly: it found 0 mentions at the start. That is the honest baseline, and it is where most new sites begin. A page set does not fix that by itself. Pages give search engines something to rank. Links and mentions are a separate, slower job.&lt;/p&gt;

&lt;h2&gt;
  
  
  How long until niche pages rank, and what can you promise?
&lt;/h2&gt;

&lt;p&gt;Weeks to be indexed and ranked, and you cannot promise a position to anyone, including yourself. Twenty good pages published today are not twenty positions next week. Indexing takes time, ranking takes longer, and a page set on a site with no mentions starts behind.&lt;/p&gt;

&lt;p&gt;That is also why the checks matter more than the launch. The launch is a day. The weeks after are where you find out whether pages got indexed, whether titles collided, whether a ping went out, and whether your tracker is lying to you. If a vendor tells you a date for page one, ask them what happens when it is missed.&lt;/p&gt;

&lt;h2&gt;
  
  
  When should you not build niche landing pages?
&lt;/h2&gt;

&lt;p&gt;When you do not have a real answer for each niche. If your product does the same thing for every buyer and you have no niche knowledge, the honest page set is one page, not twenty. Making twenty from that is how you produce doorways on purpose.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;You have no real proof for a niche. A before and after has to come from the product. If you have to invent it, skip the page.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;You cannot name the different angle for two similar queries. Merge them.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Your facts keep changing and nobody owns the facts file. Wrong numbers on many pages cost more than the traffic is worth.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;You need results this month. This channel pays back over weeks, and sometimes longer.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Your site has a basic technical problem. Fix crawling and indexing first, then add pages.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;If most of that list does not describe you, the format above will work as a starting point. Pick the queries from real phrasing, hold the section order, write the rule down, run the script, and keep the facts in one file. None of it is clever. It just means every page earns its place, and you can show that it does.&lt;/p&gt;

&lt;blockquote&gt;
&lt;p&gt;If you want to know whether your product has enough niches to justify a page set, book a call and bring your ten likeliest buyer queries. I will tell you which deserve their own page and which should be merged, and if the answer is that you need fewer pages than you hoped, I will say that. Ivan / 2pizza.team. SEO build-out is from $3,500 fixed price, over 12 weeks.&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://2pizza.team/blog/niche-landing-pages-without-doorway-penalty" rel="noopener noreferrer"&gt;2pizza.team&lt;/a&gt;. We build AI and automation systems for small teams - fixed price, two to six weeks. &lt;a href="https://2pizza.team/work" rel="noopener noreferrer"&gt;See the work&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>automation</category>
      <category>ai</category>
    </item>
    <item>
      <title>Blocked GPTBot? It Does Not Hide You From ChatGPT Search</title>
      <dc:creator>2pizza.team</dc:creator>
      <pubDate>Fri, 06 Nov 2026 13:00:10 +0000</pubDate>
      <link>https://dev.to/2pizza/blocked-gptbot-it-does-not-hide-you-from-chatgpt-search-2ig5</link>
      <guid>https://dev.to/2pizza/blocked-gptbot-it-does-not-hide-you-from-chatgpt-search-2ig5</guid>
      <description>&lt;blockquote&gt;
&lt;p&gt;TL;DR: AI crawlers come in two kinds. Training crawlers (GPTBot, ClaudeBot) collect pages to train models. Search and citation crawlers (OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User) fetch your page to answer a user right now and name you as the source. Blocking a training crawler does not remove you from AI search answers. Blocking a citation crawler does. We published the opposite conclusion in one of our own audits, retracted it in writing and rebuilt our scoring. The rest of this post is how to test your own site per bot and how to decide what to block.&lt;/p&gt;
&lt;/blockquote&gt;

&lt;p&gt;I get one question more than any other from founders who have read a scary article about AI scraping: should I block GPTBot? The honest answer depends on what you want, and the part that trips people up is that blocking it does less than they think in one direction and nothing at all in another. I will start with a mistake we made, because it is the cleanest way to show the problem.&lt;/p&gt;

&lt;h2&gt;
  
  
  What is the difference between GPTBot and OAI-SearchBot?
&lt;/h2&gt;

&lt;p&gt;GPTBot collects pages that OpenAI may use to train its models. OAI-SearchBot is the crawler behind search results in ChatGPT: it finds and indexes pages so they can be shown and cited when someone asks a question. They are separate user agents with separate purposes, and OpenAI documents them separately so that site owners can treat them differently.&lt;/p&gt;

&lt;p&gt;ChatGPT-User is a third one. It fetches a page when a user, or an assistant acting for a user, asks for that specific page in the moment. Perplexity splits the same way: PerplexityBot for its search index and Perplexity-User for fetches on behalf of a person. Anthropic runs ClaudeBot for training data collection.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;The two kinds, as we group them&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Training crawlers - GPTBot (OpenAI), ClaudeBot (Anthropic). They collect pages to train models. Nobody is being answered at that moment and nothing is cited.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Search and citation crawlers - OAI-SearchBot and ChatGPT-User (OpenAI), PerplexityBot and Perplexity-User (Perplexity). They fetch pages to answer a user right now and cite them.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Vendors add and rename agents over time, so treat this list as a snapshot and check each vendor's own crawler documentation before you write rules. The split into two purposes is the part that stays stable.&lt;/p&gt;

&lt;h2&gt;
  
  
  If I block GPTBot, will I disappear from ChatGPT answers?
&lt;/h2&gt;

&lt;p&gt;No. Blocking a training crawler does not remove you from AI search answers, because the answer is built from pages fetched by the search and citation crawlers. Block OAI-SearchBot and you can drop out of ChatGPT Search results. Block GPTBot and, as far as the documented behaviour goes, you have only said something about training use.&lt;/p&gt;

&lt;p&gt;The reverse holds too. If you allow GPTBot and block OAI-SearchBot, you have given away your pages for training and also hidden yourself from the place where a buyer would have been shown your name. That is the worst of both and it is more common than you would expect, usually because someone pasted a rule list from a blog post without reading what each line does.&lt;/p&gt;

&lt;h2&gt;
  
  
  What did we get wrong on a client audit?
&lt;/h2&gt;

&lt;p&gt;On one client audit we reported that GPTBot received a 403, so the site was invisible to AI search. That conclusion was wrong. We had tested one agent and generalised from it. When we ran a per-bot test, every citation crawler received a 200 and could read the pages. The site was visible to the crawlers that matter for answers. Only the training crawler was refused, which may well have been exactly what the owner wanted.&lt;/p&gt;

&lt;p&gt;We retracted the finding in writing to the client. Then we fixed the cause rather than the report: our audit tool now scores only citation crawlers for visibility, and it shows the training crawlers in a separate block with a note explaining that a refusal there is a policy choice, not a visibility problem. I would rather tell you that than have you find an old version of the claim and wonder how many of our other conclusions are built the same way.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do we weight each crawler in our access score?
&lt;/h2&gt;

&lt;p&gt;In the access part of our audit score, the weights are OAI-SearchBot 30, PerplexityBot 30, ChatGPT-User 20 and Perplexity-User 20. Training bots count for 0. That adds up to 100 across the four crawlers that decide whether you can be retrieved and cited, and it means a site that blocks every training crawler can still score full marks on access.&lt;/p&gt;

&lt;p&gt;The weights are our judgement, not a standard. The two search index crawlers carry the most because they build the pool of pages an engine can draw from. The two user-triggered fetchers carry less but still matter, since they are what runs when a person asks about a specific page or when an assistant goes to check a source. If you disagree with the split, the useful part is the principle: score what affects being cited, and report the rest separately.&lt;/p&gt;

&lt;h2&gt;
  
  
  Does Cloudflare's setting replace my robots.txt?
&lt;/h2&gt;

&lt;p&gt;No. We said it did, and we corrected that too. Cloudflare's managed setting for AI crawlers adds its own block of rules in front of your own file. Your robots.txt is still there and still read, but the managed rules come first, so the file you see in your repository is no longer the whole picture of what a crawler is told.&lt;/p&gt;

&lt;p&gt;That matters for two reasons. First, if you are debugging why a crawler is refused, the file you edit may not be the thing refusing it. Open the live robots.txt in a browser or with a plain request and read what is actually served, not what you think you deployed. Second, if you want to allow citation crawlers while keeping training crawlers out, check that the managed rules and your own rules agree with each other, because a managed block placed first can say something different from the lines below it.&lt;/p&gt;

&lt;h2&gt;
  
  
  Can a firewall block a bot that robots.txt allows?
&lt;/h2&gt;

&lt;p&gt;Yes, and this is the failure that robots.txt reading will never reveal. Robots.txt is a polite request. A firewall, a bot-protection layer or a CDN rule is enforcement. A crawler can be told it is welcome in robots.txt and then receive a 403 or a challenge page at the network edge before it ever reaches your server.&lt;/p&gt;

&lt;p&gt;So there are two separate places to check and they can disagree: what robots.txt says, and what the firewall does when the request arrives. If you only read the first, you will tell yourself you are open while the second quietly turns crawlers away. Check both, per bot, every time.&lt;/p&gt;

&lt;h2&gt;
  
  
  How do I test each AI crawler on my own site?
&lt;/h2&gt;

&lt;p&gt;Request your important pages once per crawler, sending that crawler's user agent string, and record two things for each: the status code, and whether the returned HTML actually contains your content. Do it for each agent in your list, not for one agent standing in for all of them. That is the entire method, and it is the one our first report skipped.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;What to record for every page and every bot&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Status code. 200 is what you want from citation crawlers. 403, 429, a redirect to a challenge page or a timeout means that bot is not getting in.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Body check. Look in the HTML for a heading, a product name or a sentence you know is on the page. A 200 with an empty body is not access.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Which rule did it. If a bot is refused, find out whether robots.txt, the CDN's managed rules or your own firewall rule is responsible before you change anything.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Repeat after every infrastructure change. A new CDN setting, a plugin update or a bot-protection toggle can change the result without anyone touching robots.txt.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;User agent strings are public in each vendor's documentation, and you can send them with an ordinary command-line request or a short script. One caution: some firewalls verify a crawler by its published network addresses and not by the user agent text alone, so a test from your own machine with a borrowed user agent can be refused or allowed differently from the real crawler. Treat your test as strong evidence, not proof, and confirm against your server or CDN logs where the real bots show up.&lt;/p&gt;

&lt;h2&gt;
  
  
  Why can a page return 200 and still be empty to a crawler?
&lt;/h2&gt;

&lt;p&gt;Because the status code only says the server answered. If your site renders its content in the browser with JavaScript, the crawler receives the shell and nothing else. We audited a shop whose homepage was 2,591 bytes to crawlers while looking normal to every human visitor. That case has its own write-up in the post on a single-page app that was invisible to AI crawlers, so I will not repeat it here.&lt;/p&gt;

&lt;p&gt;The point for this post is that the body check in the test above is not optional. A per-bot status test that stops at 200 would have called that shop healthy. Status first, content second, always both.&lt;/p&gt;

&lt;h2&gt;
  
  
  Should I block training crawlers, citation crawlers, both or neither?
&lt;/h2&gt;

&lt;p&gt;Decide by what you want, not by what feels safe. If you object to your pages being used for model training, block the training crawlers. If you want to be named in AI answers, keep the citation crawlers open. These two choices are independent and you can make them differently. Blocking everything is a coherent choice only if you are happy to be absent from AI answers.&lt;/p&gt;

&lt;p&gt;&lt;strong&gt;A decision guide&lt;/strong&gt;&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;You want to be recommended in ChatGPT and Perplexity answers: allow OAI-SearchBot, ChatGPT-User, PerplexityBot and Perplexity-User. Whether you allow training crawlers is a separate call.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;You object to training use of your content but still want citations: disallow GPTBot and ClaudeBot, allow the four citation crawlers, then run the per-bot test to confirm the allow rules really let them through.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;You do not want any AI use of your pages: block both kinds, and accept that you will not be cited. Write that down as a decision so nobody later reports it as a bug.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;You publish paid or members-only content: blocking is reasonable, but put the gate in the firewall, since robots.txt only asks.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;I will not tell you which of these is right for you. A publisher worried about its archive and a B2B service company that lives on inbound enquiries have different interests, and both are legitimate. What I will say is that most small companies we see want to be named in answers, and a surprising number of them are blocking the crawlers that would do it because a rule list told them to block anything with AI in the name.&lt;/p&gt;

&lt;h2&gt;
  
  
  What does blocking training crawlers actually cost me?
&lt;/h2&gt;

&lt;p&gt;As far as the vendors document, it costs you your pages being used to train their models and nothing to do with the live answer. I cannot give you a number for what that is worth, and I would distrust anyone who does. It is a values and licensing question more than a traffic question, which is why I keep it separate from visibility in our scoring.&lt;/p&gt;

&lt;p&gt;What does cost you is a blocked citation crawler. In that case the engine cannot fetch your page when a buyer asks, so it answers from other sources and your competitors get named. That is an access problem, it is cheap to fix, and it is the first thing we look at before recommending any other work.&lt;/p&gt;

&lt;h2&gt;
  
  
  When should you not pay anyone for this?
&lt;/h2&gt;

&lt;p&gt;If your whole problem turns out to be one firewall rule or one line in a managed setting, you do not need an agency. You need an afternoon and the per-bot test described above. You can run it yourself with nothing more than a request tool and the user agent strings from the vendors' documentation.&lt;/p&gt;

&lt;p&gt;Where an outside check earns its fee is when the picture is mixed: robots.txt says one thing, a managed CDN block says another, the firewall differs by country or request rate, and the pages are partly rendered in the browser. Untangling that across every crawler is tedious, and tedium is a fair thing to pay for. Our audit starts from $500, and if the finding is just a blocked crawler and one afternoon of work, we will say so and stop there.&lt;/p&gt;

&lt;h2&gt;
  
  
  What should you do this week?
&lt;/h2&gt;

&lt;p&gt;Four steps, in this order, and none of them need a budget.&lt;/p&gt;

&lt;ul&gt;
&lt;li&gt;&lt;p&gt;Read your live robots.txt exactly as a crawler receives it, including any block your CDN adds in front of your own rules.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Run the per-bot test on your homepage and your two or three most important pages, for each of the four citation crawlers, checking status and content.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Check your firewall and bot-protection rules separately for anything that refuses these agents or challenges them.&lt;/p&gt;&lt;/li&gt;
&lt;li&gt;&lt;p&gt;Write down your training policy as a separate decision, then set the training rules to match it without touching the citation rules.&lt;/p&gt;&lt;/li&gt;
&lt;/ul&gt;

&lt;blockquote&gt;
&lt;p&gt;I am telling you about the GPTBot mistake because the category is young and a lot of confident claims in it rest on one test of one bot. If you want a second pair of eyes, the audit at /geo tests each crawler individually and reports training bots separately, so a refusal you chose on purpose is not mistaken for a visibility problem. Ivan / 2pizza.team&lt;/p&gt;
&lt;/blockquote&gt;




&lt;p&gt;&lt;em&gt;Originally published at &lt;a href="https://2pizza.team/blog/blocking-gptbot-chatgpt-search" rel="noopener noreferrer"&gt;2pizza.team&lt;/a&gt;. We build AI and automation systems for small teams - fixed price, two to six weeks. &lt;a href="https://2pizza.team/work" rel="noopener noreferrer"&gt;See the work&lt;/a&gt;.&lt;/em&gt;&lt;/p&gt;

</description>
      <category>automation</category>
      <category>ai</category>
    </item>
  </channel>
</rss>
