DEV Community

Cover image for Your leads aren't dying in a channel. They die between channels. Here's a 4-checkpoint lead trace that fixed it for us
Tommy Shelby
Tommy Shelby

Posted on

Your leads aren't dying in a channel. They die between channels. Here's a 4-checkpoint lead trace that fixed it for us

Over three months, two export businesses of ours (industrial equipment and fresh fruit) collected 6,800+ overseas leads. 55 buyers wrote in on their own. 21 got a quote. 4 paid online. (Both businesses combined, as of Nov 11, 2026.)

Every door loses people. The ones that hurt aren't the buyers who said no. They're the ones who fell through the gaps between your social posts, your website and your WhatsApp.

What the gap looks like

A buyer sees your post, opens your site, reads three pages, ends up with three questions, and taps the WhatsApp button. Your salesperson sees one line: "price?" They don't know what the buyer read, so they start from zero. The buyer gets tired and leaves.

We hit this ourselves, in four different ways:

  • One site had almost no real visitors reaching the contact page: in 14 days, 59 search clicks, and 0 to 1 real person got to the contact page.
  • A site inquiry form was broken for about 6 days. Every real submission failed and nobody noticed. Our on-call AI only watched inquiries that arrived, and the test script always passed. Later the logs showed a real buyer had tried the form 3 times on a product page, then gave up.
  • Our AI auto-replied to a customer with a price of $140 each. The most expensive tier on the website was $70. The error came from a hand-filled price sheet nobody maintained.
  • Of 14 customers we had quoted, the median waited 16 days without paying. The ones who could place an order on the website themselves paid within 0 to 2 days.

What this is not

It's not mass DMs, and it's not letting AI close deals. We once sent 1,448 cold emails for a factory and got 1 reply. We stopped that day.

It's one thing: give every lead an ID, so you can trace it from first touch to payment and see exactly which step lost it.

I call it one line, four checkpoints: Arrive, Enter, Answer, Chase. One sheet, one row per lead, first column is the lead ID.

Checkpoint 1: Arrive (where did this lead come from?)

Common mistake: counting likes and views.

Fields: lead ID, source platform, source URL, company, contact role, market, matched product, why this is a lead (one sentence), first touch date.

Leads without a source URL or a real company don't go in the sheet.

AI does: collects public info, attaches the source link, scores the fit.
A human does: decides whether to contact. Every comment, DM, connection request and email is sent by a person.

Checkpoint 2: Enter (does the door actually open?)

Common mistake: assuming the form and the WhatsApp button work because you installed them. "Inquiries received" only counts the people who succeeded.

Do this: every day, a real browser goes from the landing page to submit and sends a real test inquiry. If it fails, push an alert.

Fields: entry page, test time, result, failure reason, retries by the same visitor (3 or more: contact them the same day).

AI does: runs the test, logs it, reports failure. A human: fixes the page.

Checkpoint 3: Answer (can the first reply carry the context?)

Common mistake: the salesperson doesn't know what the buyer saw, and asks everything again. Worse: letting AI quote prices.

Do this: when an inquiry arrives, first fill a confirmation card: country, use case, model, quantity, customization, documents needed, reply deadline, owner, what's missing.

AI does: pre-fills from what you already have and lists the gaps.
A human confirms: MOQ, price, lead time, certifications, payment terms. The AI makes no commercial promises.

Checkpoint 4: Chase (who follows up after the quote?)

Common mistake: send the quote, wait.

Fields: quote date, amount, valid-until, next follow-up date, one-click order link, blocker.

Every quote we send now carries a one-click order link.

AI does: reminders, today's chase list, follow-up drafts. Invoices and packing lists too, which a human checks before sending.
A human does: negotiate, send, confirm payment.

A 7-day minimum test

Start with one product and one market.

  1. Day 1: pick one product, one market, build the sheet, decide how lead IDs work.
  2. Day 2: backfill your last 20 real inquiries. Fill "Arrive" completely and count how many have no traceable source.
  3. Day 3: play the buyer. Go from social to site to form, note every snag.
  4. Day 4: set up the daily real-browser test, alerts to yourself.
  5. Day 5: build the confirmation card, try it on 3 old inquiries.
  6. Day 6: contact the high-intent leads by hand and log each one.
  7. Day 7: review three numbers: share of leads with a complete source; days the entry test passed; quotes with no movement for more than 3 days.

Don't judge it by new followers or posts published.

A prompt you can paste into Codex or Claude Code

I have one export product line and one target market. Build me a "lead trace" sheet, one row per lead, with four column groups: Arrive (lead id, source platform, source url, company, role, market, matched product, why lead, first touch date), Enter (entry page, test time, result, failure reason, retries), Answer (country, use case, model, quantity, customization, documents needed, reply deadline, owner, missing), Chase (quote date, amount, valid until, next follow-up date, one-click order link, blocker).
Pre-fill Arrive and Answer from the public links and existing inquiries I give you.
Hard rules: never invent customers, emails, purchase intent, certifications, MOQ, prices, lead times or sales records; write "TBD" when you can't find it. Never send any DM, email or WhatsApp message automatically; all outbound is drafts only. Quotes, samples, contracts, payment terms, MOQ, lead time and certifications are always "needs human confirmation". Cite a source for every cell.
Enter fullscreen mode Exit fullscreen mode

Last

The point isn't one more channel. It's making sure someone catches the lead between channels. Run one product and one market through all four checkpoints and you'll know exactly where yours falls out.

More notes from running this: https://tommyapi.com/tommy/en/ · Code: https://github.com/Tommyapi

Numbers are from our own two export businesses, as of 2026-11-11.

Top comments (1)

Some comments may only be visible to logged-in visitors. Sign in to view all comments.