From Seats to Outcomes

The seat deserves more respect than it gets. For thirty years it made software the easiest line item in the budget, because it tied cost to the one thing every company already counts: people. Finance could forecast a seat-based bill to the decimal, procurement could benchmark it against a market rate, and nobody had to answer the awkward question of what the software actually produced, because a seat was permission, not production. No one ever audited a seat. You checked that the people existed, not that the value did. That was the trade, cost made predictable, value left conveniently vague, and everyone accepted it because measuring value was harder than counting heads. The vagueness was not a flaw in the model. It was the feature that let the model scale.

Agents break the link that made the seat coherent, because an agent's output no longer scales with anyone's headcount, and pricing is following the work. Salesforce launched Agentforce at two dollars per conversation in October 2024. Intercom prices its Fin agent at 99 cents per resolution, the figure Bessemer's monetisation playbook, a venture firm reading its own market, treats as the reference point for outcome pricing. a16z's December 2024 enterprise newsletter put it in writing in December 2024: per-seat is no longer the atomic unit of software. Sierra sits at the far end of the ladder. Bret Taylor describes a model where a resolved case bills at a pre-negotiated rate and an escalation to a human costs nothing. Each step down that ladder, from access to activity to outcome, prices software less like a licence and more like labour.

Outcome pricing sounds like the buyer's dream, which is exactly why it deserves suspicion. Pay only for results reads as the vendor absorbing all the risk. Look at what actually moved. Under seats, value was vague but cost was bounded; a licence count cannot surprise you in month seven. Under outcomes, value is finally legible and cost is unbounded, because the meter runs on volume you do not fully control, and it is read by the party whose revenue rises with every unit it declares complete. The definition of a resolved case stops being a support metric and becomes a commercial boundary worth real money. Was the customer's problem solved, or was the ticket closed? Under seat pricing that distinction cost nothing. Under outcome pricing it is the entire invoice.

This is where procurement discovers it has no muscle. Buying outcome-priced software is not licensing a tool. It is closer to managing an outsourced workforce, and it demands the disciplines that come with one: a written definition of done for every outcome you will pay on, verification independent of the vendor's own meter, sampling that catches quiet drift, a dispute path, stop conditions for when volume spikes, and budgets that cap the month before the month caps you. None of that apparatus exists in a standard software buying process, because none of it was ever needed. The renewal meeting was the only audit a seat ever faced. An outcome deserves the scrutiny you would apply to an invoice from a contractor who grades their own work.

The scarce work, then, is not choosing a pricing model. It is building the capacity to buy under one. Defining the outcome so it cannot be gamed is judgment work: too loose and you pay for closed tickets, too tight and the agent escalates everything and delivers nothing. Verifying that an outcome actually occurred is engineering work: an independent record of what happened, reconciled against the vendor's meter, at a cost that does not swallow the savings. Writing the stop conditions is governance work: the point at which the meter pauses and a person asks why volume doubled. Vendors are racing each other toward the outcome. Almost nobody is building the buyer's side of the table, and the gap between those two speeds is where the expensive surprises will live.