S2 · The Credit Economy — how Clay actually charges you

(paid-course opener: orient → stakes → in. No tease grammar.) Okay — this video is about how Clay charges you money. Before you build anything, before you approve anything, you need to be able to read this ledger — because I'll be honest with you, almost every Clay horror story I've seen comes down to somebody who couldn't.

And I've seen it go wrong in both directions. I've seen a marketing team inherit Clay, run one big audience — a hundred and fifty thousand rows — watch it nearly eat their whole credit balance, and then freeze. Just stop touching the thing. Every click felt like it might cost a thousand dollars. And I've seen the opposite: a company where nobody was watching at all, and hundreds of thousands of credits — paid for, approved, sitting right there in the account — just expired at the end of the year, never used.

Same root cause both times: nobody could read the bill. So that's what we're fixing right now.

(Master map appears: Strategy Track highlighted.) Quick note on where we are — this is the strategy layer. We're not building any tables today. Today you learn to read the bill.

Okay, so Clay charges you on two meters — just two, that's the whole system. (two gauges appear)

Meter number one is Data Credits. You spend these whenever you're buying information you don't already have — an email lookup, company data, an AI agent going out and researching the web. The price depends on what you're buying, so think of this one as the grocery bill.

Meter number two is Actions. You spend these whenever Clay does something — runs an enrichment step, calls an API, syncs a record to your CRM. And these are small — roughly a tenth of a credit each. Think of that one as the delivery fee.

And then there's the part nobody tells you. (third column stacks up, line by line) Formulas. Filters. Sorting, deduping. Importing your data. Looking up data you already own — from your CRM, from another table. Manual edits. Exporting a CSV. All of that is free. Free, free, free. Logic is free. Moving your own data around is free.

And one more rule — honestly, this is the one that should change how you feel about this whole tool: you only pay for results that come back. A provider searches, finds nothing? Costs you nothing.

Let me give you an example of this. A real conversation.

I was advising a founder — she had a list of ten thousand companies, and she wanted to run an AI agent on all of them to qualify them. Read each company, decide if it fits. Reasonable idea, right? So I asked her: okay — what does that cost?

If you run the agent across all ten thousand rows, that's ten thousand actions — plus the research cost, on every single row. (board: 10,000 rows → agent → 10,000 paid runs)

But here's the thing — she already knew half those companies were the wrong size and the wrong geography. And knowing that isn't research, right? That's just a filter, and filters are free. So you filter first — that costs you nothing — you get it down to the two hundred companies that actually deserve the research, and you run the agent on just those. (board: 10,000 → free filter → 200 → agent) Same output, at a fraction of the cost. And she got it in one line. She said: "So if I filter it myself first and give Clay the two hundred names — that's two hundred actions." That is correct. That's the whole game.

Build the free part first, and the paid part last. Every Clay horror story you've heard — and look, the data says teams underestimate their Clay costs by forty to sixty percent, and new users burn like seven hundred to a thousand dollars in their first two weeks — almost all of it is this one mistake, multiplied. Paid columns doing a job the free ones do better.

Second judgment call: whose key pays?

So when Clay's marketplace buys you data — emails, phone numbers, company records — you pay data credits. And honestly? Usually a good deal. Clay negotiated rates with a hundred and fifty providers so you don't need fifteen contracts. So: buy your data through Clay.

But when the work is intelligence — an AI agent researching, classifying, writing — you can plug in your own API key. Your own OpenAI account, your own Anthropic account, does the thinking. You pay them directly, at cost, and Clay just charges the action. No data credits at all. Bring your own intelligence. (board: two lanes — "DATA → marketplace, credits" / "INTELLIGENCE → your key, actions only")

Now — one more thing, and this matters if you're the one signing the check. In March 2026, Clay rebuilt this whole pricing model. They split the meters. They cut marketplace data prices — some by fifty, some by ninety percent. And they started charging an action on every API call, which used to be free. Older contracts got grandfathered. So if your plan is from before March 2026 — don't touch it until you've done the math, right? And if it's not, just know this: the new model rewards exactly one kind of team. The kind that can read this ledger. Which is now you.

Okay, now let's read your actual bill. (one screen capture: a plan & usage page — sandbox account; the only screen in this video)

Your plan is two allowances — so many data credits, so many actions, every month. And there are three questions I ask against this page. Every time.

One: what did we buy? That's the plan line. And look at the rollover rule while you're there — unused credits carry over, but only up to a cap. That cap is exactly how those credits in my cold open died, by the way. They paid for them, the cap hit, and they just… expired.

Two: what did we use — and on what? Usage, by table. I'll be honest with you — in one workspace I audited, one single table had burned forty-two percent of everything that company had ever spent. One table. And more than half the tables in there had never used a single credit. The bill was not a mystery. Nobody had looked.

Three: what happens when we run out? Top-ups — and top-ups come at a premium. So if you're buying overage credits every month, that's not a usage problem. That's a planning problem. It means nobody priced the runs before they ran.

So here's what you do with all of this. Five questions. They're printed on the back of the ledger card, and they turn any burn, any invoice line, any "hey, should we run this?" into a two-minute conversation. (board builds the list)

One: which meter did this touch — data, actions, or neither? Two: did we pay only for successes? Because if something billed on failure, something's set up wrong. Three: could a free mechanism have done it? A formula, a filter, a lookup on data we already own? Four: whose key paid for the intelligence — the marketplace's, or ours? And five — this is the one that prevents the disaster instead of explaining it afterwards: what does this cost at the full row count? You ask that one before it runs. Because a reasonable-looking table, multiplied by fifty thousand rows — that's how budgets die.

Ask those five out loud in a meeting and watch what happens. You haven't built a single table yet, and you're already the most dangerous person in the room about this tool.

Everything in this video is one idea, so let me just say it plainly: you only pay for results that come back, and logic is free. That's the whole economy. And once you can read it, both of the situations I told you about at the start — the team that froze, the company that wasted — both of those disappear. You stop being scared of the cheap things, and you stop ignoring the expensive ones. The tool was never the problem, right? Nobody could read the bill.

This is Artifact F-02 — the Credit Ledger Card. (card on screen) Everything free, everything paid, both meters, the five questions on the back. Print it. Genuinely — print it. It makes reading a Clay bill a ten-second job.

Now, one thing this card cannot do: it can't stop one untrained person from burning an entire balance while seven other people watch. That's not economics. That's governance. And that's the next video.

Your exercise: the two-meter quiz. Ten scenarios — like, "your SDR imports five thousand rows and runs two formulas: what does it cost?" — and you answer each one in meters: credits, actions, or free. Then take your own plan and decode it into the worksheet: what you're buying, what rolls over, what it costs when you run out. Pass = eight out of ten on the quiz, plus the completed worksheet. Submit in Skool.

Next up: S3, Credit Governance — and the true story of one person, five hundred thousand credits, and the seven people who never got to use them.