Data sourcing & enrichment waterfalls

Build lists that are accurate, affordable and repeatable, not just big.

Roadmap: Data, enrichment & signals, AI agents & measurement

Why it matters

This is the most visible part of the job. Tools like Clay made it possible to chain dozens of data providers in one table, which is why the GTM engineer title took off. But data is expensive, coverage differs wildly by region and segment, and bad data quietly wrecks everything downstream: bounces hurt your domains, wrong titles waste reps' time, and stale records poison the CRM. The skill is designing a pipeline that gets the right fields at a known cost and accuracy.

What it's made of

Choosing sources: databases, scrapers, public data, first-party data
Designing a waterfall: provider order, stop conditions, fallbacks
Email and phone verification, including catch-all domains
Tracking cost per enriched record and per valid email
Spot-checking accuracy with a hand-verified sample
Using AI web research for fields no provider sells
Refresh cadence and handling job changes
Writing enriched data back without overwriting good values

How a waterfall works

A waterfall asks provider A for a field. If A returns nothing (or a low-confidence value), it asks provider B, then C, and stops at the first good answer. Because each provider covers different segments, three providers in sequence often find emails for 70–85% of a list where any single one might find 40–60%. Exact numbers vary a lot by region and seniority, so measure your own.

Order matters for cost. Put the cheapest provider with a decent hit rate first and the expensive, high-coverage one last. Always end with verification: a found email isn't a deliverable email.

Measure before you scale

Run 100–200 rows that look like your real target, then compute per provider:

Hit rate: what share of rows got a value.
Accuracy: on 20–30 rows you checked by hand, how many were right.
Unique contribution: rows only this provider found. A provider with a 50% hit rate that adds 2% on top of the others may not be worth it.
Cost per valid result: credits spent ÷ verified, correct values.

Catch-alls, bounces and verification

Some domains accept mail for any address (catch-all), so verifiers can't confirm a mailbox exists. Treat "catch-all" as a separate status: send to them in smaller volumes, or only when a second source agrees on the address. Keep hard bounces below about 2% per campaign. Above that, mailbox providers start treating you as a sender with a bad list, and the damage outlasts the campaign.

AI research fills the gaps

Some of the most useful fields aren't sold anywhere: "do they sell to enterprises?", "which competitor do they mention?", "do they have a free trial?". AI web-research steps can answer these from websites, job posts and news. Give the step a strict output format, an explicit "unknown" answer, and check 30 answers by hand before you trust it on 3,000 rows. Run it only on rows that passed cheaper filters, because it's usually the most expensive column in the table.

Mid-level looks like

Builds a working table that finds companies, people and emails using a few providers, with a verification step before sending.

Senior looks like

Orders providers by measured hit rate and cost for the specific segment, sets accuracy thresholds, budgets credits per account tier, and designs refresh and write-back rules with RevOps.

Common pitfalls

Running every row through every provider, then discovering the table cost ten times the budget.
Trusting a provider's "valid" flag without testing it on your own segment.
Enriching the whole TAM before checking whether the first 100 rows are even correct.
Overwriting a rep-entered phone number or title with a guess from an enrichment provider.

Practice drills

Drill 1
Build a table from 200 target companies to verified emails for two personas. Report hit rate, cost per verified email and accuracy on a 25-row hand check, per provider.
A strong answer: A filter step before expensive enrichment, measured per-provider numbers, catch-alls handled separately, and a recommendation on which providers to drop.
Drill 2
Your budget is fixed at a number of credits that covers about 1.5 lookups per contact on average. Design a waterfall and tiering plan that stays within it for 5,000 contacts.
A strong answer: Cheaper sources first, expensive steps only for top-tier accounts, a stop condition per field, and arithmetic that shows the plan fits the budget.
Drill 3
Write an AI research step that decides whether a company sells to enterprises, then evaluate it on 40 companies you labelled yourself.
A strong answer: A clear definition, a structured output with an unknown option, cited evidence, measured agreement with your labels, and a revised prompt based on the errors.

Watch path

Play all · 4 videos · 212 min

Best resources for this skill

Repos, courses, articles and podcasts, ranked by how much they teach.

Repo

kkrlstrm/gtm-pipeline

Reference implementation of a list-building path for GTM engineers: a plain-English campaign brief becomes a deduped, qualified, enriched, sequencer-ready contact list in seven stages, with swappable data providers.

kkrlstrmSep 202616 ★ stars

Repo

forma-norden/clay-claude-code-skill-pack

Claude Code skill files for building Clay enrichment tables, AI columns, and outbound exports - for GTM engineers and revenue teams.

forma-nordenMar 202612 ★ stars

Course

Clay University

Free courses, cohorts and certifications on Clay covering enrichment waterfalls, signals, CRM enrichment and automated outbound, the core toolkit of most GTM engineering roles.

Clay

Podcast

AI Is Changing The Economics Of GTM ft. Lennart Kornblum

In today's episode, I chat with Lennart, GTM Engineering Lead at Moss, about running fast-moving SMB outbound across Germany, the Netherlands, and the UK. Lennart supports both BDRs, who focus purely on outbound, and SDRs, who handle marketing handoffs. His strongest SDR play targets funnel drop-offs: when someone enters an email but doesn't finish, his team reverse-enriches the account and conta…

The GTM Engineer PodcastSep 202613 min