For agencies reselling lead-to-booking CRM workflows

For $200 and in 72 hours, one lead path from your own brief runs on your accounts

— with a written report of every way I tried to break it, from a fixed list of 20, and what happened each time. Credited in full to the $1,200 build.

Samvel Avagyan, Yerevan

Run this on your job

Send your brief. I build one lead path from it.

Start my $200 test

Upwork escrow via direct contract, or pay after your run.

20cases · 19pass · 0fail · 0blocked · 1by design

Run date
2026-10-24
Pinned versions
n8n 2.43.4 · Postgres 16.15 · Node 20.20.1

Full HTML reportUnbranded report copy

Failing cases stay visible. I mark whether each failure came from the service or was caused for the test.

Every run is logged; every failure alerts with the lead and the step.

A lead got lost
What I do to itI stop CRM; restart a worker.
What must happenBounded retries must check state, keep the lead, alert and add a human task.
A slot was booked twice
What I do to itI race leads for 1 slot.
What must happenOne slot must be held and confirmed; the other must get a new slot in its zone.
A wrong or duplicate CRM record
What I do to itI replay events; test matches.
What must happenOne record must remain; updates must use clear matches or a human decision.
The bot talked over a person
What I do to itI ask for a person and send a human reply.
What must happenThe bot must stop for that lead and cancel queued messages.
The bot went silent
What I do to itI break the AI reply.
What must happenRules must ask the 3 qualifying questions and send an alert.
Someone faked a lead
What I do to itI fake signatures; omit data.
What must happenForged leads must be rejected; missing data must be requested, never invented.
All 20 casesA fixed catalogue of 20; before the clock we agree which apply to your brief; the report marks the rest not applicable.On the reference stack, WhatsApp or a web form, HubSpot and Google Calendar, all 20 apply.
  1. 1The full lead path runs end to endA lead got lostpass
  2. 2Replay same event now and 15 min later: 1 bookingA slot was booked twicepass
  3. 3Existing customer: update the contact, no duplicateA wrong or duplicate CRM recordpass
  4. 4Same name, other phone: human task, no mergeA wrong or duplicate CRM recordpass
  5. 5No phone, email: ask by email or human task; invent nothingA wrong or duplicate CRM recordpass
  6. 6Other language: reply in it or hand to a humanA lead got lostpass
  7. 7No free slot in 7 days: say so, human task, take no slotA lead got lostpass
  8. 82 leads pick 1 slot: book one, offer the other a new slotA slot was booked twicepass
  9. 9Different time zones: show and book in the right zonesA slot was booked twicepass
  10. 10Calendar error: retry cap, alert, human task, 'we will confirm'A lead got lostpass
  11. 11CRM rate limit: queue and retry; lose nothingA lead got lostpass
  12. 12CRM down 5 min, worker restarts: check state, no repeat writeA wrong or duplicate CRM recordpass
  13. 13Timeout after success: no second bookingA slot was booked twicepass
  14. 14AI timeout or bad output: same 3 questions by rules, alertThe bot went silentpass
  15. 15Lead asks for a person: stop bot, cancel queued sendsThe bot talked over a personpass
  16. 16Human replies during queued send: do not send itThe bot talked over a personpass
  17. 17Follow-ups: WhatsApp 24 h only; STOP ends sends, consent loggedA lead got lostpass
  18. 18Window closed: approved template or no send and human taskA lead got lostpass
  19. 19Reschedule: book the new slot before freeing the old oneA slot was booked twicepass
  20. 20Valid, bad signature, replay: only valid webhook changes dataSomeone faked a leadby design

Voice is not offered. SMS and GoHighLevel are not in this sample. n8n does not restart stuck work. A sweeper picks up unfinished claims and alerts a human.

72 hours: WhatsApp or a web form, HubSpot and Google Calendar, in n8n. For other channels or CRMs, I confirm in writing whether 72 hours holds and what the test covers before the clock starts.

Make or Zapier is a build option, quoted separately.

Planned from a public job post: WhatsApp qualification, booking, CRM, hand-off. Fictional home-service business, test numbers. No client relationship implied.

What you send through the form

  • Your brief
  • A Meta test number
  • CRM access type: a HubSpot test account or your CRM's equivalent
  • A test calendar
  • Your 3 qualifying questions and business hours
  • The email address that gets alerts

Before the 72 hours start, I check the account, message template, webhook subscription and access.

The clock starts when I confirm in writing that readiness passed and, on Upwork, that escrow is funded.

If another test is already running, I give you the start date before the clock starts.

What you hold at hour 72

  • Your unbranded report. I add your phone: message the test number; see your test calendar booking and test CRM record.
  • A 90 s recording of your path.
  • Up to 3 things in your brief that will break, each with its fix.
  • Your fixed quote: paths; $1,000 this workflow, $1,200 each extra; 5 working days from full access; agreed cases to pass.

The path, its export and the runbook move to you with the build.

Payment for the $200 path

Found me on Upwork? Then the test and the build are paid through Upwork. A test paid there means the build stays there.

The test runs in my n8n, connected to your test accounts. Nothing is installed in yours until the build.

A prepaid invoice is available only if you ask for one.

  • Upwork escrow via a direct contract, funded before work and released on acceptance. Without Upwork, you pay after you see it run.
  • Before the clock starts, I agree with you in writing which of the 20 cases apply to the brief. Missing access on your side pauses the clock.
  • Acceptance: the path runs end to end and every agreed case is run and reported. A case blocked on my side counts as not run.
  • Payment after the run is due within 3 days of the hour-72 report when acceptance is met.
  • If unpaid, the 14-day deletion applies.
  • I delete the test path, its data and connections and revoke access 14 days after the hour-72 report, unless the build starts.

Test, build and optional care

  • Test

    $200

    I build one lead path from your brief.

    • Case report
    • 90 s recording of your path
    • Up to 3 things in your brief that will break, each with its fix
    • Message the test number from your phone
    • Miss hour 72: escrow returned, no invoice or a full refund of a prepaid invoice

    Fixes for failed agreed cases are included in the $1,000 build after credit.

  • Build

    $1,200 per workflow

    With the test credited: $1,000, paid $500 at the start and $500 at acceptance.

    Every agreed case must pass on your accounts, shown in the report.

    • 5 working days from full access
    • The agreed cases, scripted, run and logged
    • Runbook: setup, alerts, pause and human hand-off
    • Path and workflow export
    • 30 days of fixes

    If acceptance has not come within 5 working days of full access, you may cancel and I refund the first $500.

    For the credited build, Upwork uses a fixed-price contract with 2 milestones of $500 each; escrow is funded per milestone.

    Elsewhere, I invoice the same $500 stages, with the same cancellation and refund terms.

    Small Postgres in your account: your n8n database if accessible; else I set it up in the build. You pay hosting; the test store is mine.

    Build inputs: live WhatsApp Business partner access and production CRM tokens.

  • Care (optional)

    $250/month/workflow

    I provide alerts to your channel and up to 2 hours of fixes a month.

    • 20 cases re-run monthly, and after any n8n update, on the n8n I can access
    • A report with each re-run
    • Care covers n8n only, not Make or Zapier
    • No automatic enrolment
    • Cancel by one email

One workflow is one lead path: 1 or 2 entry channels, 1 calendar, 1 CRM pipeline, follow-ups and hand-off. I price extra locations or pipelines from your brief.

Send test and build credentials by a one-time secret link, never by plain email.

Not included in my fee

  • Messaging charges on your accounts
  • Build AI usage charges on your accounts
  • CRM plan charges on your accounts
  • Hosting the workflow for you after hand-over
  • Postgres hosting in your account

In the test, AI runs on my key; in the build, on yours.

I build white-label in your agency's or your client's account. You get an unbranded report to forward. I contact your client only if you ask. An NDA is available on request.

The finished workflow runs in your business's account. You own the workflow, export, runbook and report.

Code tests and other work

I trust a code test only after it fails on a deliberately broken copy. The fault counts are for code, not an AI feature. This offer has no client result.

  • On 2 tasks in October 2026, new tests caught 16 of 17 and 11 of 12 planted faults; the tests they replaced caught 5 and 1.
  • I publish DirectiveForge's failures next to its gains. Its public table keeps the F1 drop from 0.9091 to 0.8889.

Other work:

“…We threw him our OpenAI integration mess, and somehow he made it play nice with our 3D stuff.”

Arshak Minasyan, Sabrina AI (client), July 2025

Before you send it

Do I pay if a case fails?

Yes, if the path runs end to end by hour 72 and all agreed cases run and are reported. Failed agreed cases are fixed within the $1,000 build price.

What if hour 72 passes and it is not ready?

If I miss hour 72, I return escrow, refund a prepaid invoice in full, or issue no invoice. Missing access on your side pauses the clock.

Does my client's live traffic touch your n8n?

No. Your client's live traffic stays out of the test. I do not send production messages or change production records.

My brief is SMS or GoHighLevel, what then?

Before the clock starts, I confirm in writing whether 72 hours holds and what the test covers. No SMS or GoHighLevel sample is shown here.

What access do you need, and do you ever need a password?

Never a password. I need a Meta test number, a HubSpot test account or your CRM's equivalent, and a test calendar. Live WhatsApp Business partner access and production CRM tokens are build inputs.

When does the prepaid invoice with a refund apply?

Only if you ask for one, for example because your company pays only against an invoice.

Send a real lead path

No call needed. Send your brief and access types.

Send the path to build

Send your brief.

Start my $200 test