MCP

Enrichment and data waterfalls

Last verified: 2026-08-05

TL;DR

  • No single data provider covers a list well. A 2026 vendor benchmark of work-email providers found none that cleared both 95% quality and 90% coverage [1].
  • A waterfall queries providers in sequence and keeps the first usable answer — published examples take coverage from 40% and 63% individually to about 80% combined, and to 94% with a longer chain [1].
  • Order the chain by cost, not by quality, and stop enriching fields you will not use.

Why one provider is never enough

Every contact database is built from its own collection methods, so each has systematic blind spots rather than random ones — a provider strong on large US technology companies may be weak on European mid-market manufacturers. Stacking providers works because the gaps do not overlap perfectly.

The published comparison is instructive precisely because it is not flattering to anyone: one provider returned "the most trustworthy emails in the set" while reaching barely half the contacts, another reached about nine in ten with lower precision, and none managed both (2026) [1]. Coverage and accuracy trade off against each other, and a waterfall is the mechanism for refusing to choose.

How a waterfall works

Query provider one. If it returns a usable value, stop. If not, fall through to provider two, and so on. Every record traverses the chain until something answers, which means the final coverage is determined by the whole set of providers you include — reordering them does not change coverage, only what you pay to reach it [1].

That is the practical rule. Since most providers charge per successful match, put the cheapest acceptable source first and the expensive one last, so the expensive provider is only asked about records nobody else could resolve.

What to enrich, and what not to

Enrich the fields that change a decision or a sentence. In practice that is a short list: a deliverable email address, the company attributes your segmentation depends on (Segmentation and tiering), and whatever specific detail your personalization actually uses (Personalization at scale).

Everything else is cost without benefit. Enriched fields that appear in no filter and no sentence are pure spend, and they carry an ongoing accuracy liability besides — a wrong value in an unused field is harmless right up until someone uses it.

Note also that field quality varies sharply within a provider: work emails are the best-covered attribute, while phone numbers and detailed firmographics are markedly weaker [1]. Do not infer a provider's phone accuracy from its email accuracy.

Enrichment is not verification

A waterfall returns an address that some source believes is right. Verification tests whether the mailbox accepts mail today. They are different operations with different failure modes, and a confident enrichment result is not a substitute for the second step — Email finding and verification.

Generated fields

Enrichment increasingly includes derived values — a summary of what a company does, a one-line observation from its site, a guess at which of your segments it belongs to. These are useful and they fail differently from looked-up facts: a wrong lookup is missing data, a wrong inference is confident and wrong, and it goes into a sentence in your email.

Two guards. Sample the output by hand before it reaches a send, and write sentences that stay true when the inference is generic — the technique in Personalization at scale.

References

  1. Clay — Waterfall enrichment (vendor guide with 2025 work-email benchmark, Apr 2026)