From spreadsheets and Zapier to centralized leads: a two-week plan
Nobody migrates the whole operation at once. This is the plan that works: two weeks, one source a day, each step with proof that it worked — and the spreadsheet staying alive until the day nobody opens it anymore.
The biggest objection to centralizing leads is not technical; it is the fear of stopping the operation. The spreadsheet works. Zapier works — until the third tool. Replacing everything at once is the way to break what worked and not have the new thing ready. The plan below avoids that: nothing is turned off before its replacement is proven.
Each day has a task and a PROOF — the number that shows it worked. Without the proof, do not move on.
Week 1: the sources come in
| Day | Task | Proof |
|---|---|---|
| 1 | Install the pixel on the site, with form capture on | A test lead shows up on the profile with name and phone, not only e-mail |
| 2 | Connect the payment platform (Hotmart) with history | The month’s total approved sales match the Hotmart dashboard |
| 3 | Connect Meta Ads and Google Ads | Yesterday’s spend matches the managers, to the cent |
| 4 | Wire the CRM through the webhook, with stage events and a data block | A real lead’s stage change appears on their timeline within seconds |
| 5 | Measure identity: how many buyers exist as leads without the purchase? | Under 5%. Above that, some bridge (phone, document) is missing |
Day 5 is the most important and the most skipped. It is where you find out whether the form is capturing phones, whether the webhook sends country codes, whether the session marker is reaching checkout whole. Fixing that in week 1 is cheap; discovering it in week 4, looking at a wrong audience, is not.
Week 2: the data starts coming back
| Day | Task | Proof |
|---|---|---|
| 6 | Send the Lead to Meta via CAPI, with identity | The Lead’s match quality above 7 in Events Manager |
| 7 | Send the Hotmart Purchase to Meta, deduplicated | Purchase shows up with “website” origin and Meta does not count the sale twice |
| 8 | First living audience: opened pre-checkout 30 min ago and has not bought | The count matches a manual query; buyers leave within the next hour |
| 9 | Connect the AI through MCP and ask the first ten questions | The first three answers match the Hotmart dashboard and the ad managers |
| 10 | Publish the living report for Monday’s meeting | The team opens the link before the meeting and nobody asks for the spreadsheet |
The spreadsheet is not turned off on day 10. It is turned off on the day nobody has opened it for a week — and that usually happens on its own.
What remains of Zapier
Not everything. “When X, send Y” automation stays useful for what is not lead data — post to Slack, create a task. What leaves Zapier is what it did badly: being the place the lead passes through. Each “zap” that copied a contact from one tool to another becomes a connector or a direct webhook into the base, and the base starts feeding the tools — not the other way round. The automation that remains fires from a base event, with the person already resolved.
Mistakes that delay the plan
- Starting with the audience or the report. Without the sources and without measured identity, both come out wrong and trust dies on the first number.
- Wiring everything on the same day. When something does not match, there is no way to know which source is wrong.
- Skipping the proof. “Connected” is not proof; “the total matches” is.
- Migrating history by hand. The payment platform delivers history through the API; the CRM can resend events with the real date. The old spreadsheet becomes a queryable dataset, not a source.
Two weeks later, the question “how many leads from campaign X replied and bought” has a one-sentence answer, Meta receives events with identity, the abandonment audience maintains itself and Monday’s meeting has a link instead of a spreadsheet. What came after that, in 30 days of AI in management, starts here.
In CrazyLeads every step of the plan is a screen or an MCP connector tool: sources with ready connectors, a pixel with capture, the generic webhook, the identity resolver, Meta deliveries, living audiences and published reports — with each day’s proof readable in the product itself.