Form Fill Enrichment End to End

Outcome: the whole pipeline assembled from intake to enriched record, with a defined path for every failure mode and a test suite of deliberately awkward submissions.

Surface: Sync GTM app
Level: Advanced
Uses: Everything from lessons 02–04
Credits: ~3 per lead
Prerequisite: Lessons 02–04

The full chain


Form submitted
  ↓
Webhook → row created            < 5s
  ↓
Validate: email present, not disposable, not a bot
  ↓ fail → quarantine, alert if volume spikes
Normalize: email, name, company, domain
  ↓
Free-mail?  → yes → company-name resolution → personal-email path
  ↓ no
enrich_organization on domain    < 10s
  ↓ miss → mark partial, continue
Fit gate (free)
  ↓ below → skip person enrichment, still route and respond
Person resolution + enrich       < 15s
  ↓ miss → mark partial, continue
verify_email                     < 3s
  ↓
Score (free)
  ↓
Route (free)
  ↓
Follow up                        < 60s total

Every arrow with a “miss” branch is a path you have to define. Undefined branches are where leads disappear.

Every failure path

Validation fails

Quarantine, not delete. Review daily. A legitimate lead caught by a bot filter is a real cost.

Company enrichment misses

Continue on partial data. The domain alone still routes by territory, and the follow-up does not depend on headcount.

Person resolution misses

Continue. Use the form’s own name field. A follow-up addressed to the name they typed is fine.

Verification fails

Route to a human. A lead whose email will not verify may have typed it wrong, and someone should check rather than sending into a bounce.

Any step times out

Proceed with what you have. Time is the constraint this whole pipeline optimizes.

Everything fails

Human queue, immediately, with the raw submission attached.

Notice what never happens in that list: the lead is never dropped, and the response is never blocked. Every failure degrades the richness of the record, not the fact that someone gets back to them.

Timeouts are a design decision

Set them explicitly rather than inheriting whatever the default is.

Step	Timeout	On timeout
Company enrichment	10s	Continue partial
Person resolution	10s	Continue partial
Verification	5s	Route to human
Whole chain	60s	Respond on whatever exists

The whole-chain timeout is the one that matters. Without it, one slow provider call turns a sixty-second promise into a four-minute one.

Test with awkward inputs

A pipeline tested only with clean submissions will meet its first real edge case in production.

Test submission	Should produce
Perfect corporate email, full details	Complete enrichment
Free-mail address, company name given	Company resolved, person path attempted
Free-mail, nothing else	Unenriched, human queue
Typo’d domain (`@gmial.com`)	Verification failure, human queue
Company name in the name field	Normalization handles it, or human queue
All fields filled with the same string	Quarantined as a bot
Existing customer’s email	Routed to their account owner, not the new-lead path
Unsubscribed address	Suppressed, not sequenced
Two identical submissions, seconds apart	One row, one follow-up

Run all nine before you trust the pipeline. Each one represents something that will happen in the first week.

Do this now

Draw the chain with every branch

Including every miss path.

Define the action for each failure

Six paths from the list above.

Set explicit timeouts

Per step and for the whole chain.

Build the nine test submissions

Run them all and record what happened

Fix the branches that produced nothing

Any input that vanishes is a defect, not an edge case.

Measure end-to-end latency on the clean case

Check your work

Every step has a defined miss path
No failure path drops the lead or blocks the response
Timeouts are explicit, including a whole-chain one
All nine test submissions produced a defined outcome
End-to-end latency is measured

Where this breaks

An undefined failure branch silently swallows leads. A row that fails enrichment and has no defined path sits in the table with a blank status, is not scored, is not routed, and nobody is told — and because inbound volume varies naturally, the missing leads are invisible. Every branch must end somewhere, and “human queue” is a perfectly good ending.

Further automation

Once the pipeline handles the awkward cases, the same chain works retrospectively over historical form fills nobody enriched. Most companies have thousands of bare email addresses from old campaigns sitting unqualified — running them through this pipeline is often the cheapest lead source available.

Next lesson

06 — Score inbound leads, computing fit and intent tolerantly.

Reference for this lesson: Actions, Webhooks, How enrichment works, Tables.