People rarely write to us with the word "automation". They write: every Monday I copy the orders into a spreadsheet. Or: I answer the same five questions on WhatsApp all day. Or: I chase invoices by hand and I hate it.
That's the right way in. An automation isn't a technology you buy. It's a task you stop doing, because something else now does it for you, the same way every time, including at three in the morning.
This article opens the bonnet: what happens inside, a few examples, and the tools that do the work, from Zapier to a few lines of .NET.
The drawing on the cover
The cover image shows a man drawing boxes on a sheet of glass. WhatsApp, an AI agent, a CRM, a calendar, a payment, joined by arrows. On the original article it's a video, and you can watch it being drawn.
That drawing is an automation. Every one we build starts as a sketch like it, usually on a call, sometimes on the back of whatever is lying on the desk.
Read it as a sentence: when this happens, do that. When a message arrives on WhatsApp, have it read, file the contact, book the visit, send the payment link. Nothing more mysterious than that.
Three parts, always the same
Every automation has a trigger, some steps and some actions.
The trigger is the event that starts everything: a new message, a form filled in, a payment, a file dropped in a folder, or simply a time of day. Most of the time it arrives as a webhook, the instant notice one app sends another the moment something happens.
The steps are what the automation does with what it received. It checks the data first, so nothing incomplete goes any further. It reshapes it, because your CRM and your accounting software never describe a customer the same way. And sometimes it lets an AI read it and decide: what the message is about, which figures matter on an invoice, who should deal with it.
The actions are where the result lands. A record in the CRM, an email or a WhatsApp reply to the customer, a line in the accounts, a payment link.
Most automations we build go through the same seven steps. The animation below shows them one at a time, each with its dictionary entry. On our site it's interactive: click a step to jump to it.
The seven entries, if you'd rather read than watch:
| Step | On the tile | What it means |
|---|---|---|
| Webhook /ˈweb.hʊk/ noun | Trigger | An instant message one app sends another the moment something happens. |
| Validate /ˈvæl.ə.deɪt/ verb | Rules | To check every field is complete and correct before it moves on. |
| Transform /trænsˈfɔːrm/ verb | JavaScript | To reshape data with code so every tool speaks the same language. |
| Reason /ˈriː.zən/ verb | Claude | To let AI read, classify and decide what should happen next. |
| Store /stɔːr/ verb | Airtable | To save a clean record where your whole team can find it. |
| Notify /ˈnoʊ.tə.faɪ/ verb | Twilio | To reach your customer by SMS or WhatsApp, automatically. |
| Charge /tʃɑːrdʒ/ verb | Stripe | To take payment securely, without chasing a single invoice. |
One message, followed from start to finish
Take the drawing and send a real message through it.
A buyer writes at ten in the evening: is the flat on Calle Mayor still available? WhatsApp passes the message on. The automation checks there's a number, a message and a listing it knows. The AI agent reads the question, sees it's about availability and a visit, and answers with the real price and the free slots. The contact is saved in the CRM with the conversation. The buyer picks Thursday at five, and the visit lands in the agent's calendar. If the agency takes a deposit, a payment link goes out.
The agent wakes up to a booked visit and a lead already filed. He did nothing. That's the whole point.
This one is an illustration, the same example our home page uses. The next three are real.
Three real ones
An accounting firm in Sydney. Each quarter, the accountant emails a client's spreadsheet to an address. Behind it, an n8n workflow reads every line, checks each figure against the accounting records, and sends the review back with every issue flagged. A full working day of review became a couple of minutes, and the firm measured over 70% less review time. The full story.
An online course with a paid community. Someone buys the course at two in the morning. Thirty seconds later they're in the Discord community with the right role, and nobody had to be awake for it. Two n8n workflows connect Stripe to Discord. The full story.
Three holiday apartments. A booking comes in. The door code is created on the lock for the right dates and emailed to the guest, and the monthly linen order writes itself from the real arrivals. The full story.
Different trades, same shape: something happens, a chain of small steps runs, and a person gets their evening back.
The tools that do it
There are two families. Platforms where you connect blocks on a screen, and code written for the job. Neither is better in general. Each is better for something.
Zapier is the easiest way in. It connects thousands of apps, and a simple chain is set up in an afternoon. You pay per task, though, so an automation that runs all day gets expensive quickly.
Make draws your automation as a scenario on a canvas, with branches and loops, and costs less than Zapier once the volume climbs. Large scenarios become hard to read if nobody keeps them tidy.
n8n is the one for serious workflows. It's open source, it can run on your own server, and when a block isn't enough you can drop a few lines of JavaScript or Python into any step. It runs the accounting and Discord projects above. The price of hosting it yourself is that someone has to keep it updated.
Gumloop is newer and built around AI from the start: reading documents, summarising, sorting, pulling data out of web pages. Worth a look when the AI step is the heart of the job. Check first that it connects to the tools you already use.
Node.js, .NET and Java come in when a platform stops being the right shape: high volume, sensitive data, rules too tangled for boxes and arrows, or an automation that has to live inside your own software. You own everything, including the maintenance. .NET is part of our core stack, for the systems that have to hold up under real load.
Our rule of thumb: start with the lightest tool that will still be right in a year, and move a step to code only when that step has earned it. Often the answer is a mix.
How we choose
The question is never which tool is best. It's which tool is right for this business, a year from now.
How often will it run? Ten times a day and ten thousand times a day don't produce the same bill. How sensitive is the data? Customer records and money sometimes have to stay on your own server. Who will look after it? If someone on your team will adjust it, a platform they can read beats code they can't. And what do you already use? The best automation fits around your tools instead of replacing them.
The part nobody draws
The drawing on the glass shows the day when everything goes right. The work is in all the other days.
An automation that crashes is fine. You see it, you fix it. The one that hurts keeps running and gets it slightly wrong, for months, without a word.
So ours follow a few dull rules. When the data is wrong, it stops and says so: it would rather do nothing than do it wrong. An odd case gets flagged instead of dropped, because an odd line gets noticed and a missing one doesn't. Anything that needs a human, an urgent message, a refund, an unhappy customer, goes to a person straight away. And every run leaves a trace, so when something unusual happens you can see exactly where and why.
Nobody draws those rules on the glass. They're the difference between an automation you forget about and one you check every morning.
If this sounds familiar
If there's something you do every week with your fingers crossed, copying, pasting, chasing, answering the same questions, tell us about it. We'll tell you honestly whether it's worth automating, and with what.
Originally published on wolfdevlabs.com, where the seven steps are interactive.
Top comments (0)