If agents do the work, who exactly are you charging, and for what? The seat does not answer it, because an agent does not log in, hold a licence, or take coffee breaks — it can run a thousand tasks while a human finishes one.3 The billable hour does not answer it either, because you are no longer selling hours. This is not a marketing question. It is the question that decides whether your company has a business model, and most of the industry is working it out in public, at wildly different prices.
The seat is dead and the hour is dying
The logic is almost too clean. If your software successfully automates the tasks a human used to do, and you price per human seat, you are engineering your own revenue decline — every seat you make redundant is a seat you can no longer bill.3 So the money is moving. Industry estimates attributed to Bloomberg have subscription pricing sliding from roughly 60% of software revenue toward 30% over the coming decade, while outcome-based pricing climbs from about 10% toward 60%.3 The dominant transition state right now is hybrid — a fixed base plus variable consumption — which is where most enterprise renewals in 2025–26 have landed.3 Outcome pricing has become mainstream enough that Deloitte published accounting guidance for it in mid-2026; when the auditors write a manual, the model is real.5
You can watch the confusion in the actual price tags. For the same nominal unit — one resolved customer issue — the published 2026 numbers run from HubSpot's $0.50 to Salesforce Agentforce's $2.00, with Intercom's Fin at $0.99 and Zendesk around $1.50 in between.1 HubSpot moved its Customer Agent from per-use to $0.50 per resolved conversation in April 2026, halving its own headline rate.2 A four-times spread on the same unit is not a market that has agreed what an outcome is worth. It is a market still arguing about what an outcome even is.
The map, and what moves as you walk it
The spectrum itself is settled: access, then usage, then workflow, then outcome — each step tying revenue more tightly to delivered value.6 I walked through it in Services become software. What I want to add here is the part that does not show up on the staircase: every step to the right hands the customer more certainty and loads more variance onto you.
Per seat, the customer carries all the risk — they pay whether or not anyone uses the thing. Per outcome, you carry it — if the result does not land, you do not get paid, regardless of how much compute, engineering and last-mile work you burned getting there. That is the trade nobody puts on the pricing page. Moving right is not just "more aligned." It is you underwriting the customer's result with your own margin.
Outcome pricing is not a free lunch
Three problems bite the moment you try to price a real outcome, and they are why the honest answer is rarely "pure outcome."
Attribution. Real business outcomes have many inputs. If revenue rises or costs fall, your agent, the customer's team, and three other tools all have a claim on the credit — and the dispute is not academic when it decides your invoice.4 The cleaner the outcome you can point to, the fewer people can argue they caused it.
Definition-gaming. The word "resolution" is doing enormous work, and it is not defined the same way twice. A vendor who counts "any conversation the agent handled" as a resolution is running usage pricing in an outcome costume — and it quietly moves the cost of a wrong answer back onto the buyer.4 This cuts both ways: if you are the vendor, the temptation to define the outcome generously is exactly how you lose the trust the model was supposed to buy you.
Complexity variance. Outcome pricing works where the task is high-volume, homogeneous, and binary — support deflection, form processing, data extraction — and breaks where average complexity swings wildly.4 Price a variable-effort outcome at a flat rate and you have written yourself a lottery ticket that only the customer can cash.
How is "resolution" defined, and who eats the cost when the agent is confidently wrong? If the answer is "the customer," it is usage pricing wearing an outcome costume — no matter what the invoice is called.
What the real prices are actually telling you
Look again at the spread with those problems in mind and it stops being noise. Fin, Zendesk and HubSpot bill per resolution — you pay on success — while Agentforce bills per conversation, whether or not the issue is solved.1 That is not a rounding difference; it is a decision about who eats the cost of a wrong answer, priced into the model itself. And at the top of the market the "outcome" pricing sits on top of a floor: Sierra's platform and implementation alone reportedly clears $200,000 a year before a single per-outcome charge.1 Even the outcome-pricing poster children keep a base fee, because the last mile has to be paid for whether or not this month's outcomes show up.
Where a services-as-software firm should actually sit
Not at either pole. Pure access leaves the entire point of taking on the work — the outcome — unpriced, and hands the value back to the customer. Pure outcome, on anything but the most homogeneous task, is you underwriting a result you do not fully control with a margin you cannot fully predict. The defensible place is the hybrid middle, and it has a specific shape: a base that covers the cost of the last mile no matter what, plus an outcome component on the one slice of the result you actually control and can instrument cleanly.
Which turns pricing into an engineering problem before it is a commercial one. You can only price on an outcome you can measure, so the meter has to be built before the deal is signed — and a vendor confident in their agent will happily prove the deflection rate on your traffic before quoting a per-resolution price. A refusal to run that pilot is itself the answer.4 Everything I argued about production loops applies straight to the invoice: no instrumentation, no honest outcome price.
For what it is worth, this is where I land in practice. I want a base that is honest about the last mile — enough to cover the cost of getting an agent to production quality inside a specific customer's mess, because that cost is real whether or not this quarter's outcomes arrive on schedule — and then a component tied to the one result I can both measure cleanly and genuinely move. The base is not a hedge against the outcome; it is the acknowledgement that the last mile has to be paid for. And there is a category I will not price on an outcome for at all: anything whose result depends more on the customer's own business than on our software. Take responsibility for a number you do not control and you have not aligned incentives — you have volunteered to be blamed for someone else's quarter.
What this means if you are building one
Three sentences for anyone setting a price in the agent era. Kill the seat, because pricing per user of software that removes users is a countdown on your own revenue. Never price on an outcome you cannot both measure and materially control — that is not alignment, it is a bet you have handed the customer the only winning ticket to. And do not be embarrassed by a base fee: it is not a failure of nerve, it is the line item that pays for the last mile while the outcomes are still ramping.
The honest caveat: none of this is an argument against outcome pricing. When the outcome is clean, measurable and yours to control, tying your revenue to it is the most powerful alignment a services-as-software company has — it is the whole promise of selling the work instead of the tool. The discipline is not avoiding outcome pricing. It is refusing to price on outcomes that only look clean until the first disputed invoice.