Waterfall Enrichment: When It Earns Its Keep

Waterfall enrichment is the practice of chaining several data providers behind one lookup: try the cheapest first, and only fall through to the next if the first returns nothing. Clay popularised the pattern and most enrichment platforms now ship a version of it. The pitch is straightforward, no single provider has complete coverage, so querying several in sequence lifts your total hit rate.

That pitch is true for one category of data and mostly false for another, and the distinction is worth being precise about before building a chain of five vendors behind something that did not need one.

Waterfalling pays off when provider coverage is both incomplete and poorly correlated, meaning the records provider A misses are substantially different from the ones provider B misses. Personal contact details are the clearest example. Whether any given provider holds a working direct-dial number or a personal email for a specific individual is close to arbitrary, driven by which sources each one happened to acquire. Chaining four providers on a phone number lookup can genuinely take coverage from something like 30 percent to something like 60 percent, because each one contributes records the others simply do not have.

The economics reinforce it there too: a failed lookup against most contact-data providers costs nothing, since credits are typically consumed only on a match. That makes the marginal cost of adding one more step in the chain close to zero for the records that already failed.

Profile and firmographic data behaves differently. Whether a mid-size software company exists in a dataset, and what its industry and approximate headcount are, is not arbitrary between providers, it is highly correlated. Everyone building this kind of dataset works from broadly the same public footprint, so the companies missing from provider A are largely the same ones missing from provider B: very small, very new, or barely present online.

Waterfalling there produces a chain where step two answers questions step one already answered, and the records that fell through keep falling through. You have added latency, four sets of credentials, four rate limits, and four field-mapping layers, in exchange for a coverage lift that is often a couple of percentage points.

The way to find out which situation you are in takes an afternoon and is worth more than any vendor benchmark: take 200 records that your current provider failed on, run them through a second provider, and count how many resolve. If it is a handful, the correlation is high and a waterfall will not fix your coverage problem. If it is a third of them, build the chain.

A multi-provider chain has a consequence that shows up months later: you no longer know where any given field came from unless you deliberately recorded it. When a value turns out to be wrong, and someone asks which source produced it, a waterfall without per-field provenance cannot answer. That matters for debugging, and it matters considerably more for compliance, since a deletion or correction request you cannot trace to a source is a request you cannot fully honour.

If you build a waterfall, store the winning provider per field alongside the value. It is a small schema addition that turns an unanswerable question into a lookup.

Provenance is the cost that bites hardest, but it is not the only one. Three others are worth budgeting for before the chain exists rather than discovering them in production:

Waterfalling is partly a response to metered pricing. When every attempt has a marginal cost, ordering providers cheapest-first is rational, and the whole architecture follows from that constraint. Remove the per-call cost and the ordering question disappears for that provider: there is no reason to hold back a lookup you have already paid for, so it belongs at the front of any chain rather than at the end.

Kooperativa is priced that way, $499/mo or $449/mo annually per workspace, unlimited requests inside a shared 500 requests per minute limit, which makes it a sensible first step in a waterfall on profile and company data specifically. The honest limit of that argument: we do not return email addresses or phone numbers at all. On contact data, which is where waterfalling actually earns its reputation, we are not a step in the chain, and a team whose real gap is direct-dial coverage should build that chain out of providers that sell it.