Clay Review

Every company selling B2B data claims good coverage, and every one of them is telling the truth about a subset. One is strong on US enterprise and thin on European mid-market. Another has excellent mobile numbers in some sectors and none in others.

The consequence is familiar to anyone who has run enrichment at scale, whether through Bardeen or a data vendor: you buy a provider, run your list through it, and get back a file that is half-filled. The missing half is not random — it is whichever segment that vendor happens to be weak on, which is often the segment you care about.

Clay's answer is to stop choosing. Query one provider, and if it returns nothing, query the next, and the next, until a field is filled or the list is exhausted. That single idea — waterfall enrichment — is what the product is, and everything else is built around it.

What Is Clay For?

• Waterfall enrichment is the core idea — query one data provider, and if it has no answer, try the next, through a marketplace of more than 200.
• Claygent handles what no database holds: AI agents that research a company or person from the open web when structured data runs out.
• It is infrastructure, not a lead list — data flows out to your CRM, to LinkedIn, Meta and Google ads, or through a native email sequencer.
• Signals track change: job moves, promotions and other triggers that make an outreach relevant rather than random.
• Named customers are unusually strong, including Intercom, Anthropic, Vanta, ElevenLabs, Verkada and Rippling, with a 14-day Pro trial and no card required.

Go-To-Market Data Terms

Five terms that decide whether an enrichment stack works or quietly wastes money.

Waterfall enrichment

Querying data providers in sequence until one returns an answer. Raises coverage above any single vendor, and costs on attempts rather than on successes — so the ordering matters as much as the list.

Fill rate

The proportion of records where a field could be populated. The honest measure of a data provider, and the one to test on your worst segment rather than your best.

Signals and intent

Events suggesting timing — a job change, a promotion, a funding round. What separates a relevant outreach from a random one.

Reverse ETL

Pushing enriched data back into operational tools such as a CRM or ad platform, rather than leaving it in a warehouse. Where enrichment becomes usable.

Agentic research

An AI agent visiting sources to answer a question no database holds. More expensive per record than a lookup, so it belongs at the end of a funnel, not the start.

Why Waterfall Enrichment Beats One Provider

Clay describes waterfall enrichment as combining multiple data providers for the best coverage, drawing on a marketplace of more than two hundred.

The arithmetic is what makes this compelling. If one provider fills 60 per cent of a field and a second independently fills 55 per cent, running both in sequence fills considerably more than either alone, because their gaps do not perfectly overlap. Add a third and a fourth and coverage keeps climbing, with diminishing returns.

It also changes coverage from a procurement decision into a runtime one, which no CRM's built-in enrichment — HubSpot AI included — can offer. Under the old model, choosing a vendor meant inheriting its blind spots for the length of the contract. Here, a provider that is weak on your segment simply gets passed over and the next one answers.

There is a cost consequence that follows directly, and it is the thing to design around. Every provider you query costs something, whether or not it returns an answer. A waterfall that always runs all ten providers is far more expensive than one that stops at the first hit — and ordering the sequence so the cheapest, highest-coverage provider goes first is the single biggest lever on your bill.

This is why Clay rewards someone who owns it. A well-ordered waterfall and a badly ordered one produce the same data at very different prices.

One provider's coverage gap is not random. It is whichever segment they are weak on — which is frequently the one you care about.

What Claygent Adds Beyond Structured Data

Claygents are described as AI agents that research target companies and people. They matter because a great deal of what you actually want to know is not in any database.

Whether a company has just opened an office in a new market — the sort of signal an outbound platform such as Artisan acts on but cannot find. Whether their careers page suggests they are building the team you sell to. Whether their pricing page implies they have moved upmarket. None of these are fields a data vendor sells; all of them are readable by something that can visit a website and answer a question.

That is the gap Claygent fills, and it is the more defensible half of the product long term, because structured data is a commodity and judgement about unstructured text is not.

Two cautions. Agent research costs materially more per record than a database lookup, so it belongs at the end of a funnel rather than the start — qualify on cheap data first, then send only the survivors to an agent. And an agent asked an ambiguous question returns a confident, ambiguous answer, so the quality of your prompt is the quality of your data.

Do Those Customer Numbers Mean Anything?

Clay publishes specific results against named customers: Intercom at 140 per cent more outbound pipeline, Anthropic at a threefold enrichment rate improvement, Vanta above 80 per cent enrichment coverage, ElevenLabs at 50 per cent more qualified leads, Verkada at three times the reply rate.

These are worth separating into two kinds, because they are not equally meaningful.

The enrichment figures — Anthropic's threefold improvement, Vanta's coverage percentage — are directly attributable and highly credible. Enrichment rate is a measurable property of a tool: you run the same list through the old process and the new one and compare fill rates. If waterfall enrichment does what it claims, exactly this result should appear.

The pipeline and reply-rate figures are weaker evidence, not because they are untrue but because they are overdetermined. A 140 per cent pipeline increase involves the messaging, the offer, the team, the market and the season alongside the data. Better data plausibly contributed; it did not act alone.

What makes the whole set unusually credible is the naming. Anthropic, Intercom, Rippling and Vanta are companies with communications functions that approve this kind of citation. Named, approved customer results are meaningfully stronger evidence than any anonymous percentage, and this is a stronger reference list than most tools in the category can show.

• Trust enrichment-coverage figures most — they are a direct property of the tool and easy to replicate.
• Treat pipeline and reply-rate gains as contributed to rather than caused by better data.
• Weight named references above any statistic; a company that let its name be used has checked the claim.
• Replicate the enrichment test yourself: run one real list through your current process and through Clay, and compare fill rates.
• Judge on your own hardest segment, not on US enterprise where every provider looks good.

Where Credits Go, And How To Waste Fewer

Clay runs on credits and offers a 14-day Pro trial with no card required. The credit model is where teams most often get an unpleasant surprise, and it is entirely avoidable.

The mechanism to understand is that a waterfall spends on attempts, not on answers. Ten providers queried for one record is ten costs whether the first one answered or the tenth did. Multiply that across a list of ten thousand and the difference between a good waterfall and a careless one is not marginal.

The three levers, in order of impact. Order your waterfall so the provider with the best coverage-to-cost ratio for your segment runs first. Set the waterfall to stop on first success rather than continuing. And filter the list before enriching at all — there is no reason to spend on records you would discard afterwards.

Claygent sits on top of this as the most expensive step per record, which makes it the last thing you should run rather than the first.

We do not publish amounts, since credit rates change. Use the trial to run one real list end to end and record the credits consumed; that figure, extrapolated to your monthly volume, is the only cost estimate worth having.

Running The Trial Properly

The 14-day Pro trial is genuinely enough to answer the only question that matters, provided you test on the right list.

The bar is step three. If Clay does not beat your current fill rate on your hardest segment, the waterfall is not reaching providers that cover it.

1. Take a real list you have already enriched elsewhere, and pick the segment where your current provider performs worst — not your best market.
2. Build a waterfall of three or four providers and run the same list through it, stopping on first success.
3. Compare fill rates field by field against your existing result. That delta is the entire value proposition and it takes an hour to measure.
4. Check accuracy, not just coverage, by verifying twenty filled records by hand. A filled field that is wrong is worse than an empty one.
5. Now add a Claygent step on the qualified subset only, asking one specific question your data cannot answer, and judge the answers.
6. Record the credits each stage consumed. Reorder the waterfall cheapest-effective-first and run it again — the difference between the two runs is your ongoing saving.

Who Actually Operates It

Clay is powerful and it is not simple, and that combination has a predictable failure mode worth naming before purchase.

The tool assumes someone will design waterfalls, write agent prompts, manage credit consumption and maintain the workflows as sources change. In practice that becomes a role — the person who owns Clay — and teams that buy it without assigning one tend to use a fraction of it while paying for all of it.

That is not a criticism of the product so much as a description of what it is. Infrastructure needs an operator; a lead list does not. The company's own community and education programmes exist largely because this is true.

The question to settle before buying is therefore organisational rather than technical: who on your team will own this, and do they have the hours? If the answer is nobody, a simpler enrichment tool will deliver more of its value than Clay will of its own.

Where that person does exist, the ceiling is high — which is why the named customers are the companies they are.

Clay: The Data Infrastructure Facts