How to buy AI without overpaying

Every other chapter in this playbook is written from the seller's chair — what to charge for, how to move a meter, how to defend a margin. But half of this audience sits on the other side of the table, buying these tools and watching the bill move. This chapter is for you. You don't get to set the meter. You do get to refuse to pay for the opacity around it — and most of the leverage you have is in the questions you ask before you sign, not the ones you ask when the invoice surprises you.

Most of what follows is read straight off the AI Credit Index — the part of a buyer’s checklist that almost nobody else can give you, because it comes from looking at how 51 products actually meter their pricing rather than from general advice. The seven price moves are the core, and they’re where the money is. After that, briefly and honestly, the rest of the checklist — the things every buyer should weigh that sit outside what a pricing benchmark can measure.

The seven price moves#

These are the levers you control as a buyer of a metered product. None of them require negotiating leverage you may not have; most are just questions asked at the right moment, before the answer stops being free.

Find the refill price before you commit. The top-up, not the sticker, is where your real spend lands. If you’re buying self-serve and the refill rate is only shown once you’re inside the app, surface it during the trial and weight that silence in your decision; if you’re buying for a team, make the refill rate a written contract term. A vendor who won’t show you what overage costs until you’re already a customer has told you something about how the rest of the relationship will go.

Translate credits into your work, not the demo’s. Ignore “1 credit = 1 image.” Across the Index a single “credit” buys things that differ by more than thirteen million to one, so the vendor’s example tells you nothing about your bill. Run ten of your real jobs in a trial, watch the meter move, and divide — that’s your true cost per task. Multiply by your real monthly volume against the plan’s allotment and you’ll know, before you sign, whether you’re a top-up customer and at what price.

Size to a spiky month, not an average one. You buy the plan for a normal month and the top-ups in your worst week — the launch, the board deck, the incident. The average month is a comfortable lie; the spike is where the expensive math actually bites. Forecast the peak and price the plan against that, or the overage will find you exactly when you’re busiest.

Decide top-up vs. upgrade with the dollar math. When you’re regularly over your allotment, there’s a break-even between buying refills and moving up a tier — and it rarely sits where the vendor’s UI nudges you. Do the division yourself; the Index works the mechanics out so you can see which side of the line your usage falls on, instead of defaulting to whichever path the product makes easiest to click.

Get exhaustion behavior in writing. When the meter hits zero, does the product hard-stop, throttle, or quietly bill you for overage by default? “Auto-overage” is a budget you don’t control — a bill that grows while you’re not looking. Know which one you bought, and prefer the soft landing you switch on over the automatic charge you have to switch off.

Re-check the credit burn rate every quarter. A static dollar price is not a static price. Track how much real work your plan’s credits buy now versus last quarter; if the same money buys less, you’ve been repriced — whatever the pricing page still says. This is the single most common way a metered bill climbs without anyone deciding to raise the price, and the only defense is measuring it yourself.

Cap what you can’t forecast. When a meter is genuinely unpredictable, do what the most sophisticated buyers on earth did: in 2026, after burning its annual AI budget in four months, Uber capped each engineer at $1,500 of agentic-coding spend a month; Microsoft hit the same wall and moved thousands of developers onto a flat ~$39 seat instead. A spend you can’t bound is one you shouldn’t sign. If the vendor won’t give you a predictable unit, impose the predictability yourself with a hard ceiling.

Beyond the meter#

Now the honest part. The seven moves above are the buying decisions the Index can actually inform — price, metering, overage. They are not the whole of buying AI well. The rest of the checklist sits outside what a pricing benchmark can measure, and it’s table stakes everyone writes about — but no buyer’s guide is complete without it, so here it is, tight.

Pilot before you commit — and use the pilot to learn your burn rate. This one is pricing-adjacent enough to belong with the seven: run the tool for one to three months on a monthly plan before you sign anything annual. The annual prepay is where the two quiet risks compound — credits that expire unused at cycle end, and a burn rate that can drift down mid-term — and a short pilot is how you measure your real cost on your real work before you commit a year of budget to a meter you’ve never watched run. Don’t let an annual discount talk you past the one period where you’d actually find out what the thing costs you.

Read the data terms before you’re locked in. Outside the Index’s lane entirely, but on every real checklist: is your data used to train the vendor’s model, and can you opt out? How is PII handled, and where does it physically land? How deep do the integrations reach into your systems, and what permissions do they actually require versus request? These don’t show up on a pricing page, and they’re far harder to renegotiate after you’ve wired the tool into your stack than before. Ask while you’re still a prospect, when your questions still get answered.

Know how hard it is to leave. The last quiet cost is exit. How portable is your data and your configuration if you switch? How much of your workflow gets rebuilt around this one vendor’s way of doing things? Lock-in isn’t a reason not to buy — but it is a reason to know, going in, what walking away would cost, because that number is the real ceiling on your leverage at every renewal that follows.

The one lever you have#

You don’t set the meter, the burn rate, or the bill. What you set is whether you pay for the dark around them. Make the vendor show you the refill price. Translate their unit into your work. Size to the spike, not the average. Get exhaustion behavior in writing, pilot before you sign, and read the terms while your questions are still cheap to ask. None of it requires leverage you don’t have — it requires asking the questions at the one moment the answers are free, which is before you commit, not after the invoice arrives.

And then keep watching. The Index refreshes quarterly precisely because a price you checked once is a price you’ve stopped checking. The seller’s job is to run the meter. Yours is to keep reading it.