Pricing when the seat stops working

A seat was always a proxy. Nobody ever wanted to buy a login; they wanted the work that a person with a login could do. Counting people was simply the cheapest way to approximate how much work your software absorbed.

When software does the work directly, the proxy inverts. Your best customer now delivers more output with fewer people, so your revenue falls at the exact moment you are creating the most value. That is not a market being unfair to you. That is a pricing model measuring the wrong thing.

This is no longer a fringe view. Andreessen Horowitz has been writing about the shift toward outcome-based pricing since late 2024, and has since written about the price wars that follow when everyone moves at once. Zuora's guide for finance leaders lays out the per-agent, per-activity, per-output and per-outcome variants now being used in practice. The direction is settled; what is not settled is how a company with an installed base gets there without breaking itself.

Every model is a bet on a condition being true. Pick the one whose condition you can actually meet.

Pull the last four quarters. For every account, plot seats against actual usage of whatever your product does — documents processed, tickets resolved, transactions handled, hours saved.

If those two lines move together, you have time. If they have separated — usage climbing while seats flatten — you have a pricing problem dressed up as a sales problem, and it will show up next as renewals that hold flat while your delivery costs rise.

Most founders have never run this query. It takes an afternoon and it settles the argument.

Outcome pricing is the model everyone wants and the one most companies are not ready for. It requires three things simultaneously: the outcome must be countable, you must be able to attribute it to your product without an argument, and you must know your own cost per outcome well enough that a heavy user cannot bankrupt you.

Miss the first and you get disputes. Miss the second and you get renewals where the customer credits their own team. Miss the third and you have written an unlimited cheque against inference costs you do not control — a16z's piece on AI price wars is largely about companies that did exactly that.

The honest intermediate step for most companies is a platform fee plus usage. It gives you floor revenue, it gives the customer a forecastable bill, and it lets you learn your unit economics before you promise a result.

This is the part that gets skipped, and it is the part that costs money.

The failure I see most is not a pricing model chosen badly. It is a good model introduced late — after customers have already built next year's budget on the old one. A limit that arrives after the budget is set costs goodwill you cannot buy back, and it is remembered far longer than the extra revenue lasts.

The second failure is repricing without changing the sales conversation. If your team still opens with how many users the buyer has, the new model is a spreadsheet change and nothing more.

I once sat in a room at a company worth around $2B where a very expensive idea was being presented to the CEO paying for it. Everyone nodded. I said, plainly, that it would never work, and why. The room went quiet, and afterward he pulled up a chair next to his and asked me to stay close and keep saying what his executives were afraid to say.

Repricing dies in that same silence. Somebody in your company already knows the seat model has stopped tracking value — usually the person closest to the customers who are over-consuming. They will not raise it in a meeting where the pricing page is treated as a founder decision that has already been made.

Before you model anything, ask the two people nearest the customer what they would charge and why. You are not looking for agreement. You are looking for the objection nobody has said out loud yet.

Run the seats-versus-usage query. If the lines have separated, add one usage-based tier next to your existing plan and sell it to the five customers who are already over-consuming. That is a two-week project, it is reversible, and it produces the only evidence that matters: whether someone will pay the new way.