Services-as-software: when you buy the outcome, not the seat

The head of customer operations at a subscription retailer told me she had cut her cost per ticket by 62 per cent. Then she showed me the contract. Her vendor bills per “resolution”, defined as any conversation the customer does not reopen within 72 hours. Tickets abandoned by frustrated customers count as resolutions. When I re-ran her numbers excluding conversations with no customer reply and no CSAT response, the saving was 31 per cent. Still good. Not what the slide said.

This is the defining commercial shift of the year. A growing class of vendors no longer sells software by the seat; it sells completed work units — a resolved ticket, a reviewed contract, a booked meeting, a merged pull request. The category has picked up the name services-as-software, and the pitch is seductive: you are buying labour output with software margins, and the vendor carries the delivery risk.

What the market charges

Public pricing is thin in this category, which is itself informative. Intercom’s Fin is the transparency outlier at roughly $0.99 per resolution, with resolution defined and credited in the documentation. Sierra prices per resolution under enterprise agreements and does not publish; the range I have heard consistently from buyers is $1 to $2.50 depending on complexity and volume commitment. Decagon sits in similar territory. In legal, Harvey and its peers price per seat with heavy minimums — six figures annually for a mid-size firm is normal — while claims-drafting tools like EvenUp bill per matter.

In engineering, Cognition’s Devin moved from a $500-a-month floor to a consumption model with a low entry point and metered compute units, which is closer to honest. Coding assistants that operate agentically — Claude Code, Cursor, GitHub Copilot’s agent modes — sit between $20 and $200 per developer per month, and the top tiers are consumption-limited in ways the marketing does not lead with.

The clauses that decide whether it works

I have now read nine of these contracts. Four things determine whether outcome pricing is genuinely a transfer of risk or an accounting costume.

The definition of the unit. If “resolution” includes deflections where the customer simply left, you are paying for failure. Insist that a resolution requires either an explicit customer confirmation, a completed action in your own system, or a positive CSAT — and that the definition is in the contract, not the onboarding deck.

The escalation boundary. Every vendor has a hand-off to humans. Ask what percentage of volume escalates and whether you are billed for the attempt. One vendor I reviewed billed a resolution when its agent successfully routed a ticket to a human. That is not a resolution; that is a router charging like a solution.

The floor. Almost every outcome-priced deal carries an annual minimum. At the volumes most mid-market buyers actually have, that minimum is the real price and the per-unit rate is decoration. Divide the minimum by your realistic annual volume before you compare anything.

The data exit. These vendors accumulate your resolution history, your tone, your edge cases. Ask, in writing, what you get on termination: the transcripts, the labels, the taxonomy, in what format, within how many days. Two vendors gave me a straight answer. One offered “a commercially reasonable export”, which means nothing.

Where I would buy, and where I would build

Buy when the work is high-volume, externally facing, well-bounded and boring: password resets, order status, refund eligibility, appointment rescheduling. The vendors have seen millions of these conversations and your in-house team has not. A realistic first-year outcome, based on the four deployments I have data for, is 35 to 55 per cent full deflection on tier-one volume, not the 80 per cent in the case studies.

Build when the work touches your differentiating logic, your pricing rules, or regulated decisions you must be able to explain to an auditor line by line. And build when your volume is low — below roughly 3,000 contacts a month, the annual minimum on most outcome deals makes the per-unit maths irrelevant.

One structural warning. Outcome pricing quietly aligns the vendor with volume, not with reduction. If your product improves and ticket volume falls, your vendor’s revenue falls too, and the commercial pressure runs the wrong way at renewal. The best contract I saw this quarter handled that with a declining-minimum clause tied to a jointly agreed root-cause reduction programme. It took the buyer four months to negotiate. It was worth it.