Skip to content
ArticleUpdated 5 min read

Where your Clay credits actually go

Clay no longer charges for an enrichment that returns nothing. Credits go on two meters: data credits when a provider returns data, and actions for nearly everything else, including AI calls, HTTP requests, exports and CRM pushes. Tables cost more than planned when steps per row multiply, sample runs are skipped, or top-ups fill gaps at a premium.

The old complaint, and why it is out of date

For years the standard complaint about Clay was that a failed lookup still cost you. If you read an older review that says a waterfall trying six providers bills six times for nothing, it describes Clay before March 2026. Clay's pricing page now says it plainly: if an enrichment returns no result, you are not charged data credits or actions.

That is a fair policy, and it is the same one Apollo, Instantly and Lemlist follow. So if your Clay balance still drops faster than you planned, misses are not the reason. The reasons are structural and mostly predictable, which means you can do something about them.

Two meters, not one

Since its March 2026 repricing, Clay bills in two currencies.

  • Data credits pay for data bought from Clay's provider marketplace: an email, a phone number, company details. They are spent when a provider returns something.
  • Actions pay for the platform's work: running a step, calling a provider, an AI model call, an HTTP request, an export, a push to your CRM or sequencer.

As of October 2026, Free includes 500 actions and 100 data credits a month, Launch ($167 a month) 15,000 actions and 2,500 data credits, and Growth ($446 a month) 40,000 actions and 6,000 data credits. Seats are unlimited.

Two meters make cost per contact hard to see in advance. A row can be cheap in data credits and expensive in actions, or the other way round, depending on what your table does to it.

Where the credits actually go

1. Steps that are not enrichments. Every AI column, formula that calls out, HTTP request, export and CRM push uses actions. According to Databar's write-up of the change, HTTP functions were free before March and are now paid actions. A table that classifies each company with an AI prompt, checks a website, then pushes to HubSpot can spend more actions per row on that work than on finding the contact.

2. Waterfall steps that do return data. A waterfall stops at the first provider that returns a result, and that result is charged. But a result is not always the result you wanted. A provider can return a generic company address, or an email you then pay to validate in a further step. Each of those is a real return, so each is billed.

3. Steps per row multiply. Ten columns across 5,000 rows is 50,000 step runs. A table that cost a few hundred actions on a 30-row test can use a month's allowance on the full list. Auto-running columns when new rows arrive does the same thing quietly.

4. Top-ups. Actions reset each month and do not roll over. When you run out, Clay's pricing page lists top-ups for Launch and Growth at a 30% premium over the plan rate. Data credits roll over up to twice the monthly amount, which cushions uneven months, but a big table late in the month can still push you into top-ups.

5. The repricing itself, depending on your plan. Reports disagree on whether March 2026 made Clay cheaper. Databar reports marketplace data prices dropping 50 to 90% and the top-up premium falling from 50% to 30%. Other third-party analyses report mid-tier costs rising about 28% with roughly 40% fewer credits. Both can be true for different workloads: a table heavy on marketplace data probably got cheaper, one heavy on HTTP calls and AI steps may not have. If you sized your plan before March, re-measure.

How to estimate a table before you run it

  1. Build the table on 50 rows that look like your full list, not your best 50.
  2. Run it once and note the actions and data credits used.
  3. Work out per-row figures. Divide each meter by 50.
  4. Multiply by your full row count, separately for each meter.
  5. Add a margin of 20 to 30% for rows that go further down the waterfall than the sample did.
  6. Compare with what you have left this month, remembering actions do not roll over.

A worked example, with made-up sample results: say the 50-row run used 400 actions and 60 data credits. That is 8 actions and 1.2 data credits per row. For 3,000 rows, expect about 24,000 actions and 3,600 data credits, or 29,000 and 4,500 with a 25% margin. That is more than Launch includes in one month on both meters, so either split the job across two months, cut steps, or plan for top-ups.

How to keep it down

  • Filter before you enrich. Put cheap qualifying steps first and stop rows that fail before they reach paid providers.
  • Order the waterfall cheapest first, and remove providers that rarely win for your market.
  • Bring your own API keys where you already pay a provider: Clay then charges actions only, no data credits.
  • Turn off auto-run on expensive columns while you are still building.
  • Deduplicate rows before enrichment, especially when merging lists.
  • Keep AI steps short. One prompt that returns three fields is cheaper than three prompts.
  • Measure cost per usable record, not cost per credit: total spend divided by rows that came back with the fields you needed. Our piece on what a lead actually costs shows the method.

When a different tool is cheaper

Clay is worth its price when someone on the team will build and maintain tables, and the flexibility pays for itself. If what you want is a ranked list of people worth contacting, with no table to build, a narrower tool may cost less. Sluice, which we make, uses one currency (a credit is one cent), bills only lookups that returned something usable, and shows the worst-case price of each pass before it runs. It does far less than Clay. The Sluice vs Clay comparison sets out who should pick which, and our pricing is public.

Questions people ask

Does Clay charge for failed enrichments?
No. Clay's pricing page says that if an enrichment returns no result, you are not charged data credits or actions. Articles saying otherwise describe the model before the March 2026 repricing.
What uses Clay actions?
Platform work: running workflow steps, calling providers, AI model calls, HTTP requests, exports and sending data to other tools. Actions reset each month and do not roll over.
Do Clay data credits roll over?
On Launch and Growth, unused data credits accumulate up to twice the monthly amount. Actions do not roll over.
How do I estimate what a Clay table will cost?
Run it on a sample of about 50 representative rows, read the actions and data credits used, divide by 50 and multiply by your full row count, then add a margin for rows that take longer paths through the waterfall.
Does using my own API keys in Clay save credits?
Yes. With your own provider keys, Clay charges actions only and no data credits; you pay the provider directly instead.

Sources

  1. Clay pricing
  2. Databar: Clay pricing 2026, March updates (third-party)

Try it on your own market

Sluice quotes the worst-case price before anything runs and charges only for lookups that found something, so finding out costs close to nothing.

Get started