Clay's credit counter is easy to ignore until it is empty. The habit that fixes it is simple: look at the credit estimate before you press run.
Here is what things cost, from tables I built this week on the free plan, and the habits that keep a build cheap. Costs are what Clay displayed in my workspace at the time and they change, so check yours.
What things actually cost per row
| Step | What Clay showed me |
|---|---|
| Find companies or jobs (the search itself) | Free to preview, rows count when you save them |
| Enrich company (size, industry, revenue) | about 1 credit per row |
| Research company ICP with AI | about 0.1 credit per row |
| AI column on a small model | about 0.1 credit per row |
| Find people at a company | about 0.1 credit per person returned |
| Company contacts lookup | about 1 credit per row |
| Latest funding waterfall | about 5.8 credits per row, up to 12 |
The pattern is obvious once you see it in one place. Small-model AI columns are almost free. Waterfalls and agent-style research are where the money goes.
On a full pipeline run across three rows, Clay's own estimate was about 31 credits. That is roughly ten credits per company. On a free plan with 100 credits a month, that is ten companies, not ten thousand.
Habits that keep it cheap
1. Turn auto-run off before you add a single column
When auto-run is on, any new row or changed column starts spending on its own. Turn it off first. It is the single most useful setting in the product when you are building.
2. Test on three rows
Run every new column on three rows, read the output, and only then widen. A broken prompt on three rows costs a few credits. The same prompt on three thousand rows costs a week.
3. Read the estimate before every run
Clay shows a credit estimate in the run menu. It takes one second to read and it has saved me more than any other habit.
4. Put the expensive step last
Order the table so cheap columns filter first. If an AI column can disqualify half the companies for a tenth of a credit, run it before the waterfall that costs six. You only pay for enrichment on rows that survived.
5. Use run conditions
You can tell a column to run only if another cell meets a condition, for example "only if the company is B2B software". The waterfall then skips everything else. In my funding table I could see "run condition not met" on rows that did not need the lookup, which is exactly what you want.
6. Pick the smallest model that does the job
Scoring a company from a description does not need the biggest model. A small one handles it at a tenth of a credit. Reserve bigger models for writing that has to sound right.
What to do when you hit the limit
If the credits are gone, resist the urge to upgrade immediately. Ask first whether the expensive column was necessary. For a lot of work, a cheaper data source or a smaller waterfall gets you 90% of the result.
I wrote about how waterfalls work and when they pay for themselves in Clay waterfall enrichment explained.
The short version
Auto-run off. Three rows. Read the estimate. Cheap filters first. Small models by default. You will spend a fraction of what you would otherwise, and the table works the same.
Clay's newer features change where some of this control lives. See Clay Functions replace Recipes and Clay Sculpt 2026: what changes.