Should you meter at all?

The instinct, the moment you ship anything built on a model, is to put a meter on it. Every call costs you something real, so charging by the call feels like simple hygiene — you wouldn't give away electricity, and inference is the new electricity. That instinct is half right and almost always overapplied. A meter is not free; it carries a bill nobody quotes you up front. And the flat seat — the thing the meter was supposed to replace — keeps winning in places the reflex would never expect. Before you decide what to charge for or how to move the number, you have a prior decision to make, and most teams skip it: whether to put a meter on at all.

There are really only four ways to price the thing: a flat seat (one price, use it as much as you like), a pure meter (you pay for what you consume), a hybrid (a flat base plus a meter on the heavy stuff), and the outcome (you pay for a result). The whole rest of this playbook lives downstream of one fork in that list — meter, or don’t. Get this call right and the later questions get easier. Get it wrong and no amount of clever tiering rescues you, because you’ve taxed the wrong thing.

The reflex — and why it’s only half right#

Start with why the instinct exists, because it isn’t foolish. Classic software had a marginal cost near zero; one more user cost the company almost nothing, which is exactly why a flat price worked so beautifully and the gross margin sat at 75%-plus. AI broke that. Now every active user runs up a real, variable cost in compute every time they use the product — the COGS that software never had, showing up on the bill. Put a flat price over a variable cost and you’ve built a margin landmine: a thin tail of heavy users can consume far more than they pay, and the harder they lean on the product — the more they love it — the more money you lose on them. That is the legitimate, and real, case for a meter. When your cost genuinely tracks usage, charging by usage keeps you from being punished for your own success.

So far the reflex holds. The error is treating that argument as universal — assuming that because you can meter, and because some cost is variable, you must meter everything. That’s where the hidden bill comes in.

The bill the meter doesn’t print#

A meter looks free to install and isn’t. It charges you in four currencies that never appear in the pricing model.

It suppresses the usage you’re trying to create. This is the big one, and the most underpriced. A user on a flat plan explores — they try the feature, run it again, build it into their day, become dependent. A user watching a meter rations. Every action carries a little tax and a little anxiety, so they do less, discover less, and form a weaker habit. In a market where the entire game is to become indispensable before someone else does, a meter that makes people think twice before pressing the button is taxing the one asset you most need to accumulate. You are charging admission to the thing that was supposed to make them stay.

It pushes a forecasting burden onto the buyer — who may refuse it. A flat seat is a number procurement can approve in an afternoon. A meter is a question mark: how much will we use, what will that cost, what happens in a heavy month. Plenty of buyers simply will not sign a variable they can’t bound — they cap the spend, or they pick the predictable competitor, or the deal dies upstairs. An unpredictable meter is a smaller account than a predictable one, even when the product is better.

It is a standing trust liability. A flat price is set once. A meter is a relationship you have to keep running honestly, and every adjustment to how fast it burns is a trust event you have to survive. You’re signing up for a recurring obligation, not a one-time decision.

It is operational drag. Metering means metering infrastructure, usage billing, the support load of disputed charges, the dashboards, the explaining. Real cost, real engineering, real headcount — spent on the plumbing of charging rather than on the product.

None of these kills the case for a meter. They just mean the meter has to clear a bar, not show up by default.

The three questions that actually decide it#

Here’s the honest test. Reach for the meter only when all three of these are true at once.

Does the value track usage? If a heavy user genuinely gets proportionally more value — more results, more output they’d have paid for anyway — then charging by usage is fair, and it builds an expansion engine into the price. If usage and value have come unstuck — if the meter just punishes engagement that doesn’t make the customer richer — then you’re taxing love, and they’ll feel it.

Is your cost high, variable, and concentrated? A meter is a margin governor; it’s worth installing when there’s a margin to protect. If a thin tail of power users would torch your unit economics under a flat price, you need the governor. If your per-user cost is modest and fairly even — bounded by something physical, like hours in a meeting or words a person can actually dictate — a flat price is safe, and the governor is just friction you’ve inflicted on yourself.

Can the buyer still forecast it? Even when the first two say “meter,” you only get to do it in a form the buyer can live with. A meter denominated in a unit they can predict is a price. A meter denominated in something opaque and jumpy is a question mark with a logo — and you’ll lose the buyers who won’t sign question marks, which is most of the serious ones.

Meter when value scales and cost is dangerous and the buyer can forecast it. Miss any one and the flat seat is probably the stronger move — and choosing it is discipline, not timidity.

Why the answer is usually “both”#

Run the three questions over a real product and you rarely get a clean yes or a clean no. You get: most of what the user does is cheap, bounded, and habit-forming, and a slice of it is genuinely expensive, variable, and optional. Which is exactly why most of the market has landed on the hybrid — a flat seat for access and predictability, with a meter bolted on only to the heavy actions that actually threaten the margin.

You can read this convergence across the 51 products in the Credit Index: the dominant shape is a base subscription plus credits for the expensive work, not a pure meter on everything. The discipline is in the boundary — meter the part that’s dangerous, leave the rest flat so adoption and forecasting survive. ChatGPT draws exactly that line: everyday chat stays flat under a rate limit, while agent runs — the genuinely heavy, optional thing — are capped separately. Flat for the habit, a governor on the costly edge. The mistake the reflex makes is metering the whole product when only a fraction of it ever needed a meter.

The holdouts, and what they know#

The clearest teachers are the products that looked at the meter and walked away. Granola, the meeting notetaker, charges a flat $14 a seat (and $35 at enterprise) for unlimited meetings — and argues in public for per-seat over metered pricing, because a notetaker’s cost per user is bounded and a meter would just make people hesitate before hitting record, which is the one behavior the product cannot afford to suppress. Wispr Flow, voice dictation, is a flat $15 a month for unlimited words: same logic, bounded per-user load, a daily-habit tool that dies the instant using it feels expensive. And ChatGPT — the product that defined the category, and whose heavy users are genuinely costly — still sells a flat seat ($20 for Plus, around $25 a seat for Business) and governs with rate limits, a soft ceiling, rather than a credit meter you watch deplete and refill.

What all three know is the thing the reflex forgets: a meter isn’t only a way to capture cost, it’s a behavior you’re installing in your user — hesitation. Where the product’s value is a daily habit and the cost is bounded, hesitation is the last thing you want, and the flat seat the buyer never has to think about is the most powerful instrument you’ve got.

What stage you’re in sets the dial#

The last input is when. Early, when adoption is the entire game and you’re trying to become the default before a competitor does, lean flat and generous — a meter that makes people ration is self-sabotage when you haven’t yet earned the indispensability that would let you charge for intensity. Later, once you’re entrenched and that heavy tail has become real money, you can introduce a meter where the cost actually lives, from a position of strength. The common, expensive error is metering on day one out of reflex — taxing the adoption you haven’t won yet, to protect a margin you don’t yet have.

So decide what you’re optimizing before you decide how to price. Adoption, or margin. Land grab, or harvest. The meter is a tool for the harvest, and a liability during the grab.

The meter is a tool, not a default#

Put the meter on where your cost is genuinely at risk and your customer’s value genuinely scales — and even then, only in a unit they can forecast, and ideally only on the slice of the product that needs it. Everywhere else, the flat seat the buyer can approve without a spreadsheet is not the cautious option; it’s frequently the winning one. The question is never “can we meter this” — you almost always can. It’s “should we” — and the honest answer, more often than the reflex admits, is not yet, or not here, or not all of it.

Once you’ve decided there will be a meter, the next decision is the one that outlasts every number you’ll ever set: what you put the meter on.