# How Clay Credits Work and How to Stop Burning Them

*How-to — 2026-10-12 — by Mahmoud Zalt*

Actions and Data Credits explained, what is free, where credits actually leak, and the settings Clay documents for spending less on the same result.

**Short answer.** **Clay charges you in two separate currencies, and confusing them is why bills surprise people.** Clay's documentation defines Actions as the platform metric, where each record enriched or exported counts as one Action regardless of which provider served it, and Data Credits as the currency you spend buying the data itself from providers in their marketplace. Getting data into Clay is largely free. Enriching it, and exporting it, is what costs.

**About this guide.** This guide is based on publicly available information and Clay's own documentation, checked on 17 August 2026. Sistava is not affiliated with Clay. Interfaces and pricing change often, so confirm the current steps in their documentation before you rely on them.

Almost everyone who tries Clay has the same first month. The tool does something genuinely impressive, you get excited, you point it at a bigger list, and then the usage screen says a number you were not expecting.

That is not a pricing problem, it is a model problem. Clay's billing is actually quite legible once you know which of the two meters is running, and most of the ways to spend less are documented settings rather than clever tricks.

This guide covers the two currencies, what is genuinely free, the six things Clay themselves publish for saving credits, where to look when the number does not match what you expected, and the leaks nobody warns you about.

## What is the difference between Actions and Data Credits?

Actions measure what you did. Data Credits measure what you bought. Clay's documentation puts it plainly: each record enriched or exported counts as one Action regardless of the data source or provider, while Data Credits are the currency for purchasing data and AI from third-party vendors in Clay's marketplace.

So a typical enrichment charges you on both meters at once. One Action for the platform doing the work, plus a variable number of Data Credits for the data itself. That is why a table can look busy on one meter and quiet on the other.

This split has one immediately practical consequence. Clay's documentation states that connecting your own API keys for a provider removes the Data Credit cost, but the enrichment still consumes one Action. If your team already pays a contact data or AI vendor directly, plugging those keys in cuts half your bill without changing anything about how the table works.

## What is free in Clay?

More than people assume. Clay's documentation lists a specific set of operations that do not consume Actions, and getting data into a table is nearly all of it.

- **Getting data in. **Clay lists sourcing lists of accounts and contacts, CRM and data warehouse imports, webhook imports, and manual data entry as operations that do not consume Actions.
- **Working with data you already have. **Clay formulas and filters, plus scoring and normalisation formulas, are listed as free. Filtering a list down is one of the few things in Clay you can do as much as you like.
- **Some movement between tables. **Clay's list includes Send Table Data and Lookup Single Row in Other Table, alongside CSV export.
- **Testing a Claygent. **Clay's Claygent documentation states that test runs do not cost credits and that you can test up to ten inputs at a time for free.

The list of what does consume an Action is equally clear, and worth reading once because a few entries surprise people. Clay names each enrichment from any provider, AI uses, signals, exports and syncs to a CRM or data warehouse, HTTP API calls, ads audience exports, and one Action per lead added to a campaign in their sequencer.

**Exporting is not free.** This is the one that catches people. Clay's documentation counts a record exported as one Action, the same as a record enriched. A CRM sync across a large table therefore costs, and syncing the same table repeatedly costs repeatedly. CSV export is on the free list, but a live integration is not.

## The six ways Clay documents for saving credits

Clay publishes a credit conservation guide, and it is worth following before you go looking for clever workarounds. These six are their own recommendations.

1. **Turn auto-update off while you build** — Clay recommends disabling auto-update during setup and turning it back on once the table is finalised and you are ready to launch. Auto-update re-runs an enrichment whenever its inputs change, so every correction you make to a column while building can trigger another paid run.
2. **Connect your own API keys** — Clay names OpenAI, Anthropic, Apollo, Findymail, Prospeo, Debounce, and NeverBounce among the providers you can plug your own keys into, using credits you already pay for instead of theirs.
3. **Qualify before you enrich** — Filter disqualified contacts out early using conditional runs or formulas, so only prospects that survived your criteria get enriched. Filtering is free, enriching is not.
4. **Test on about ten rows first** — Clay recommends running a small sample of roughly ten rows before running an entire column, to catch errors and refine prompts before the full run.
5. **Use conditional runs** — Set an enrichment to fire only when a condition is met using the Only run if box in Run settings. This is how you stop paying to re-enrich rows that already hold the answer.
6. **Use lookup columns instead of re-enriching** — Pull data you already have from your CRM or another Clay table with a lookup column rather than buying the same record twice. Clay lists Lookup Single Row in Other Table among the operations that do not consume Actions.

If you do only one of these, do the conditional runs. Every other item on the list saves you a percentage. Conditional logic changes how many rows the meter ever sees, which is a different order of saving.

## What is sandbox mode, and does it stop the meter?

Sandbox mode lets you rebuild a table's configuration on a small slice of rows without touching the live one. Clay describes it as a table mode for safely building, testing, and publishing table configurations using a subset of rows without affecting your running workflow.

The mechanics, according to Clay's documentation: you turn it on with the Sandbox mode button in the toolbar and leave with Exit sandbox. A sandbox table holds a maximum of 50 rows and starts by duplicating the top ten from your table. Your production table becomes read-only while it is active, and outbound actions such as exporting or writing to another table are automatically disabled.

The part that actually saves money is at the end. Clay states that enrichment results from rows run in sandbox mode are copied to your production table when you publish, so you do not need to re-run them. You publish with either Publish and don't Run, which syncs the configuration only, or Publish and run, which syncs and runs the affected columns across all rows.

**Sandbox is a safety net, not a spending freeze.** Clay's documentation notes that you can still manually run individual cells or columns while in sandbox mode. The mode protects your production table from configuration mistakes and blocks outbound actions. It does not put a wall between you and the meter, so a manual run is still a run.

## Where do you see what you actually spent?

Clay's usage reporting is more detailed than most people realise, and finding it is the difference between guessing and knowing. Their documentation describes reaching it by clicking your account name, going to Settings, then Usage in the sidebar.

From there Clay documents several breakdowns. An Overview tab with workspace balances for Data credits and Actions. A Workbooks tab breaking spend down by folder, workbook, and table. An Integrations tab grouping spend by integration across the workspace. Separate tabs for signals, for API spend, and for programmatic spend from team members using Clay through an AI assistant.

Individual tables have their own dashboard too. Clay documents a time view showing spend over time with hourly, daily, weekly, or monthly buckets, a column view breaking spend down per column, and a run view grouping spend by individual run. There is an Export option for pulling a CSV out for offline analysis.

Two features here deserve to be better known. Clay documents a cost estimate shown before a run, giving total estimated credits and a breakdown by column, and an alert that fires when a run would exceed 5 percent of your monthly credit budget and cost more than 500 credits. If you have never seen those, it is because your runs have been small, not because they do not exist.

**The other way to do this.** Watching two meters, tuning conditional logic, and auditing spend per column is real operational work, and it is exactly what we removed at Sistava. Rather than budgeting lookups, you hire an AI Employee for a flat monthly price and tell it in plain English what you want handled. It has the tools connected already, it asks you when it is unsure, and it sets itself up. No credit maths and nothing to audit.

## Where credits actually leak

The documented settings cover the obvious cases. These are the ones people learn about after the fact.

- **Prompt iteration on a live column. **Running an AI column, reading the output, changing a word and running it again is the classic drain. Clay's Claygent documentation says test runs do not cost credits and allow up to ten inputs at a time, so the same loop in the test panel is free. Iterate there.
- **Auto-update left on during a big edit. **Clay's enrichment documentation says auto-update re-executes an enrichment when its input values change. Tidying a domain column across three thousand rows can therefore trigger three thousand re-runs.
- **Re-syncing the same records. **Exports count as Actions per record. A CRM sync that runs on a schedule over an unchanged table is paying for the same records again.
- **Enriching rows that were never qualified. **Independent write-ups on Clay spend repeatedly point at unchecked automation logic as the cause of an unexpectedly large single run. Conditional runs and early filters are the fix, and both are free.
- **Doing person-level work before company-level work. **If twenty contacts share an employer, researching that employer once at the account level costs one lookup instead of twenty. Order of operations is a spending decision.

## Do unused credits roll over?

The two currencies behave differently here too. Clay's documentation states that Actions do not roll over and unused capacity expires monthly, while Data Credits do roll over on monthly plans up to twice the monthly limit.

For annual plans Clay documents a 15 percent rollover when renewing at the same or a higher tier, and for trial or free accounts they state that 100 credits roll over while the rest expire. Clay also notes that if a provider refunds them because of invalid data, they refund those Data Credits to you.

Check the current terms on Clay's own documentation before you plan around any of this. Rollover rules are exactly the kind of thing that changes between plan revisions.

## Meter you manage, or employee you brief

Clay's model is honest about what it is. You are buying data at wholesale and orchestration on top, and the two meters reflect two genuinely different costs. For a team running large volumes with a dedicated operator, that granularity is a feature: you can see exactly which column is expensive and decide whether it is worth it.

Sistava is built for the person who does not want that job. You hire an AI Employee, tell it what you need handled, and it works out how. Neither model is universally better. If you want line-item control over every lookup you buy, learn the meters properly. If you would rather not think about them at all, brief the employee and spend the attention somewhere else.

## FAQ

### What is the difference between Actions and Data Credits in Clay?

Clay's documentation defines Actions as the platform metric, where each record enriched or exported counts as one Action regardless of the data source or provider. Data Credits are the separate currency used to purchase data and AI from third-party vendors in Clay's marketplace. A typical enrichment consumes one Action plus a variable number of Data Credits.

### What does not cost credits in Clay?

Clay lists several operations that do not consume Actions, including sourcing lists of accounts and contacts, CRM and data warehouse imports, webhook imports, manual data entry, Clay formulas and filters, scoring and normalisation formulas, CSV export, Send Table Data, and Lookup Single Row in Other Table. Clay also states that Claygent test runs do not cost credits, with up to ten test inputs at a time.

### How do I stop wasting Clay credits?

Clay's own credit conservation guide gives six methods: turn auto-update off while building, connect your own provider API keys, qualify leads before enriching, test on about ten rows first, use conditional runs through the Only run if box in Run settings, and use lookup columns to reuse data you already have instead of re-enriching it.

### Does connecting my own API key save credits?

According to Clay's documentation, using your own API keys removes the Data Credit cost for that provider, but the enrichment still consumes one Action. Clay names OpenAI, Anthropic, Apollo, Findymail, Prospeo, Debounce, and NeverBounce among the providers this applies to. For a team already paying those vendors directly, it removes one of the two meters.

### Does sandbox mode stop me spending credits?

Not entirely. Clay describes sandbox mode as a table mode for building and testing on a subset of rows, capped at 50 rows, with your production table read-only and outbound actions disabled while it is active. But their documentation also notes you can still manually run individual cells or columns, so a manual run in sandbox is still a run. The saving comes at publish time, when results from sandbox rows are copied across rather than re-run.

### Do Clay credits roll over?

Clay's documentation states that Actions do not roll over and expire monthly, while Data Credits roll over on monthly plans up to twice the monthly limit. For annual plans they document a 15 percent rollover when renewing at the same or a higher tier, and for trial or free accounts 100 credits roll over while the remainder expire. Check Clay's current terms before planning around this.

### Where can I see what my credits were spent on?

Clay documents a usage area reached by clicking your account name, then Settings, then Usage. It includes workspace balances, spend broken down by workbook and table, spend grouped by integration, and per-table views showing spend over time, by column, and by individual run, with an option to export a CSV.

## Sources

- [Clay documentation, Actions and Data Credits](https://university.clay.com/docs/actions-data-credits)
- [Clay documentation, Ways to save Clay credits](https://university.clay.com/docs/clay-credit-conservation)
- [Clay documentation, Credit usage](https://university.clay.com/docs/credit-usage)
- [Clay documentation, Sandbox mode](https://university.clay.com/docs/sandbox-mode)
- [Clay documentation, Enrichments](https://university.clay.com/docs/enrichments)
- [Clay documentation, Claygent Builder](https://university.clay.com/docs/claygent-builder)
- [Clay FAQ, What are Clay credits and how do they work](https://www.clay.com/faq/what-are-clay-credits-and-how-do-they-work)
- [CloudEagle, Clay Pricing Guide](https://www.cloudeagle.ai/blogs/blogs-clay-pricing-guide)

**Tags:** clay, clay-credits, gtm-automation, data-enrichment, cost-control, b2b-prospecting