What you charge for matters more than what you charge

Almost every pricing argument is about the number. Is $20 too low, should the top tier be $99 or $149, how big a discount for annual. Those are real questions, but they're the small ones — the number is a dial you can turn any quarter of any year. The decision that actually sets your trajectory is the one nobody argues about because it feels settled before the conversation starts: what you put the meter on. The unit. The value metric. Get it right and the number becomes a knob you adjust for a decade. Get it wrong and you'll re-price forever and never fix it, because you don't get to re-choose your metric without dragging every existing customer through a migration they didn't ask for.

A value metric has one job: to be the thing your price is denominated in. A seat. A gigabyte. A resolved ticket. A thousand credits. It looks like plumbing, and it is the single most consequential pricing choice you’ll make — because it decides whether your revenue grows when your customer succeeds, whether your buyer can say yes without a fight, and whether you spend the next five years explaining your bill. Three tests tell you whether you’ve chosen well: the metric should track the value the customer gets, scale with that value as it grows, and stay predictable enough that the buyer can forecast it. Software’s best businesses pass all three almost without noticing. AI’s “credit” frequently fails two of them.

Rule one: it has to scale with value — because that’s where growth comes from#

If you internalize one thing here, make it this. In a digital business, the compounding miracle isn’t the first sale; it’s the second, third, and tenth dollar from a customer you already won. The metric for it is net revenue retention (NRR) — what last year’s cohort pays this year vs. last, after churn and after expansion. Above 100% means your existing base grows on its own; the elite run 120%+ and the very best higher still. NRR is the holy grail because it’s what pays back the cost of acquiring a customer and then funds the next one — it’s the engine that lets you spend ahead of growth without lighting cash on fire.

Here’s the part people miss: NRR is mostly a property of your value metric, not your sales team. If the thing you charge for rises naturally as the customer gets more value — more usage, more seats that reflect more people getting paid to use you, more results delivered — then expansion happens by itself, with no new negotiation. The account grows because the customer succeeded, and you’re simply along for the ride. Pick a metric that scales with their value and you’ve built an expansion flywheel into the price itself.

Pick one that doesn’t and you’ve capped your own growth. A flat per-company license that never moves no matter how much value you deliver is a business that has to re-sell for every incremental dollar. And — the AI-specific trap — a meter that scales with your cost rather than their value is almost worse: the credit that ticks up because inference was expensive this month feels to the buyer like a tax, not a fair share of value created. They pay more and resent it, which is the opposite of the flywheel; it’s a churn timer. The cleanest value-scaling metric of all is the delivered outcome — pay per result, so the bill rises exactly when and because the customer got more of what they were buying. That’s not a coincidence; it’s why the outcome frontier and the high-margin frontier are the same address. Most products can’t reach it — but every product can at least choose a metric that moves in the customer’s direction rather than the vendor’s.

Rule two: it has to be predictable — or the buyer can’t say yes#

A value metric also has to be a number the buyer can live with before they’ve spent a dollar. Procurement approves what it can forecast. A finance team can sign off on “$30 a seat for 40 people” in an afternoon because they can do the math in their head and it won’t surprise them. They cannot sign off on “credits, which deplete at a rate that depends on which model runs and how long the task takes, toppable-up at a price we’ll show you once you’re inside.” That’s not a price; it’s a question mark with a logo.

This is where the credit fails hardest. Across the AI Credit Index, the same word — “credit” — buys things that differ by more than thirteen million to one, from a fraction of a sentence in one product to a finished video in another. A unit that means something that wildly different everywhere is a currency each vendor mints in private and never has to peg to anything (the opacity essay is about what that does to a buyer). The buyer can’t compare it across products and can’t predict it within one. So the spend either gets capped — a hard ceiling on how much of you they’ll risk — or it gets rejected upstairs by a CFO who won’t underwrite a variable they can’t bound. Either way, an unpredictable metric is a smaller account than a predictable one, even when your product is better. Predictability isn’t a courtesy you extend to the buyer; it’s a cap you remove from yourself.

You can watch this happen at the most sophisticated buyers there are. In 2026, Uber’s CTO admitted the company had burned its entire annual AI budget in four months after rolling agentic coding tools out to roughly 5,000 engineers — and responded by capping each employee at $1,500 of tool spend a month. Microsoft hit the same wall, reportedly around $2,000 per engineer in token billing, and moved thousands of developers onto a flat ~$39 seat instead. That’s a two-front verdict on an unpredictable meter: an ROI question on one side (at $2,000 a head, what exactly are we getting?) and a predictability one on the other (you can’t run a budget on a variable that eats the year in a quarter). Faced with a unit they couldn’t forecast, two of the most sophisticated buyers on earth did the same thing — forced predictability back on, one with a hard monthly cap, the other with a flat seat. When your largest customers are capping you just to stay legible, the metric is telling on itself.

The shell didn’t change — and you shouldn’t change it either#

Now the reassuring part, and the most counterintuitive advice in this whole curriculum: while the meter was mutating, the packaging barely moved, and you should be glad. Read 51 AI pricing pages and the scaffolding is almost monotonous — still roughly three tiers in good-better-best order, still an annual plan at a discount, still “contact sales” sitting above the published top. The revolution happened inside the meter; the shell around it is the same shell SaaS shipped a decade ago.

That persistence isn’t laziness or a failure of imagination. Those conventions survived because they’re battle-tested fixtures of how buyers actually decide — the accumulated equilibrium of millions of purchases. Three options make a person choose instead of freeze, and quietly steer them to the middle. An annual discount trades a little price for a lot of commitment, and both sides know the deal. A “contact us” tier tells an enterprise buyer they’ll be handled, not self-served. These aren’t arbitrary; they’re load-bearing because they fit the grain of how people evaluate and approve a purchase.

Which is exactly why you shouldn’t touch them. You are already asking the buyer to absorb something genuinely new and slightly unnerving — a usage meter, a unit they’ve never budgeted for, a bill that moves. That is a large withdrawal from their tolerance for novelty. Don’t make a second one. Innovate on the metric, where the value and the money actually are, and conform on everything else, so the buyer has precisely one new thing to learn instead of five. Spend your whole novelty budget on the meter and pay retail-familiar everywhere else. The companies that try to reinvent the metric and the tiering and the billing motion all at once don’t read as innovative; they read as exhausting, and exhausting loses deals. When the substance is changing this much, the packaging is where you offer the buyer the mercy of the familiar.

Two moves inside the shell#

Two tactics live in that familiar shell and are worth naming, because both cut in either direction depending on whether you’re selling or buying.

The anchor. A cluster of incumbents — GitHub Copilot, Atlassian’s Rovo, monday.com, HubSpot — have independently settled on the same reassuring convention: one credit equals one cent. It’s a clean, roundable anchor, and it does real work: it makes an alien unit feel legible. But pegging the dollar value of a credit says nothing about how many credits an action burns — and that second number is the one that moves. monday.com, hardening its meter, actually cut its per-credit price from roughly eight cents to one while raising a standard action from one credit to eight; the headline got cheaper-sounding while the real price held (the mechanics are their own essay). So: as a seller, a clean anchor is a legitimate way to make a meter approachable — use it, and keep it honest by being equally clear about the burn. As a buyer, read a competitor’s round-number peg as a flag, not a comfort, and go check the burn rate behind it.

Where to go dark. Enterprise opacity isn’t going anywhere — a whole tier of category leaders publish no price at all, and “contact us” is a deliberate instrument, not a confession. Going dark lets you discriminate by willingness to pay, signal a high-touch motion, and avoid anchoring a big deal to a small published number. All real. But opacity has a cost the seller rarely prices in: every “contact us” is a buyer you’ve made work harder, and some of them simply leave. The rule of thumb is to publish where you’re self-serve and go dark only where the deal size genuinely pays for the friction — because below that line, the hidden price isn’t protecting your margin, it’s quietly shrinking your funnel.

The decision you can’t unmake#

Set the number wrong and you fix it next quarter. Choose the metric wrong and you live with it, or you put every customer you have through a migration to escape it — which is why so few ever do, and so many spend years apologizing for a unit they chose in a hurry. So spend the disproportionate effort here, at the start, on the question that feels like plumbing: not what’s the price, but what are we pricing? Choose a unit that rises when your customer wins, that a buyer can forecast before they commit, and drop it into the proven shell rather than a clever new one. Do that and the price takes care of itself for years.

What you charge for matters more than what you charge. The number is a decision you get to keep making. The metric is the one you have to get right.