Outcome-Based Pricing: Can You Define the Outcome?

A customer opens a chat at 11 p.m., fires off three increasingly annoyed messages, gets a confident-sounding answer, and closes the tab. Seventy-two hours pass. Nobody reopens the ticket. Your AI agent marks it "resolved" and bills the customer fifty cents.

Did you solve their problem, or did they give up? The invoice can't tell the difference — and that gap, between a real resolution and a quiet surrender that looks exactly like one, is the whole game in AI pricing right now.

Outcome-based pricing has gone from a contrarian bet to the default answer, and for good reason. But almost everyone talking about it is arguing about the wrong thing. The hard part was never deciding to charge for outcomes. The hard part is defining what an outcome is — and that turns out to be a product problem, not a pricing one.

The model everyone is switching to

The logic is clean enough to fit on a slide. Software used to be sold by the seat because a seat was a decent proxy for value. AI agents break that proxy: one agent can do the work of ten people, so ten seats' worth of value gets billed as one login. So you stop charging for access and start charging for results.

The market has moved fast. Intercom's Fin agent says it was "first to introduce outcome based pricing to the market", on the principle that "you should only pay for Fin when it delivers value." Zendesk charges around $2 per automated resolution, or $1.50 on committed volume. HubSpot's Breeze agent bills $0.50 per resolved conversation and $1 per recommended lead. Salesforce Agentforce runs a menu — roughly $2 per conversation, about $0.10 per action, or the old $125 per user per month if you still want a seat. Support tooling vendor Lago calls performance pricing the model "most aligned with customer value", and the analysts agree: the whole industry is drifting from pay for access to pay for results.

I'm convinced the direction is right. If you've read my take on why per-seat pricing is dying, none of this surprises you. The trouble starts one layer down, in a place most pricing decks skip entirely.

"Outcome" is a definition, not a fact

Here's the sentence that should stop you cold. Zendesk infers a resolution partly from 72 hours of inactivity — and, as the pricing analyst Dan Balcauski points out, "a customer who simply gave up the conversation as futile and a customer who was assisted will both appear as inactive for 72 hours."

Read that twice. The billable event — the thing your revenue now depends on — cannot distinguish success from failure. You are charging for a signal that fires identically whether you helped or drove someone away.

It gets worse where the outcome is even softer. HubSpot's prospecting agent bills $1 per "recommended lead," but the qualification is the agent's own judgment, "without any weight given to whether the prospect actually turns into a customer." So you're paying for the agent's opinion that it did something valuable. That's not an outcome. That's activity wearing an outcome's clothes.

This is the trap. Outcome-based pricing sounds like you're finally aligning price with value. But you only get that alignment if the outcome you meter is the outcome the customer actually cares about. Meter the wrong thing and you've built a machine that bills confidently for work it didn't do — and customers figure that out faster than you'd like.

The three questions before you bill a single outcome

Lago is blunt that outcome pricing is the hardest model to operationalize: you have to define success, measure it reliably, and settle disputes when the AI half-succeeds. Before you ship a price, force your team through three questions.

Can the customer verify it themselves? The best billable outcomes are ones the buyer can check without trusting you. "Ticket closed and not reopened in 14 days" is verifiable. "Lead our model believes is qualified" is not. If verifying the outcome requires the customer to take your word for it, you don't have a pricing model — you have a trust exercise, and you'll lose it during the first invoice dispute.

Does the metric track value, or just track easily? There's a permanent tension here that Balcauski frames perfectly: should the price be tied to something the customer can easily assess, or to something the customer actually values? A resolved-ticket count is easy to measure and only loosely tied to value. Revenue influenced, hours saved, or churn avoided are what the customer actually buys — and they're brutal to attribute. Pick the metric that's a little harder to measure but a lot closer to value. Easy-but-hollow metrics erode trust every billing cycle.

What happens on partial success? Real work is rarely a clean win or loss. The agent resolves 80% of a ticket and hands off the rest. It books the meeting but with the wrong stakeholder. Decide the partial-credit rule before you have a customer arguing about it, because you will, and "we'll figure it out case by case" is how you turn every renewal into a negotiation.

Design the outcome like a north star metric

The mental shift that fixes this: stop treating the billable outcome as a pricing detail and start treating it as a metric you're designing on purpose. It's the same discipline as choosing a north star metric — a single number that has to be honest, gameable in only harmless ways, and genuinely correlated with the customer getting what they came for.

Your billing metric will get gamed. Not maliciously — structurally. The moment revenue depends on "resolutions," every fuzzy edge case gets nudged toward "resolved." If your definition can't survive your own incentive to inflate it, it definitely won't survive an auditor or a skeptical CFO on the other side. The industry is maturing into this reality: Product School now lists "outcome-based accountability replacing feature-shipping culture" as a defining 2026 shift, quoting Dow Jones's Lisa Kamm — "if you promote only for launches, you get launches." Bill only for resolutions counted loosely, and you get loose resolutions.

It's a margin problem, too

One more reason to get the definition right: outcomes cost you money whether or not you get to bill for them. Every attempt burns tokens — the resolved tickets and the abandoned ones alike. If your metered outcome is generous to the customer, you're eating inference cost on all the near-misses. If it's stingy, churn does the collecting for them. This is exactly the token-cost math I've written about before: a pricing metric that ignores your cost of goods is a great way to grow revenue and lose money at the same time. Lago notes that hybrid models — a subscription floor plus a usage or outcome component — post 21% higher median growth than pure-play pricing, and part of why is that a floor keeps the near-misses from bleeding you dry while you tune the definition.

The takeaway

"We charge for outcomes" is not a pricing strategy. It's the beginning of one. The strategy is the definition underneath it — the specific, verifiable, value-tracking, partial-credit-aware description of what your agent did that a customer will happily pay for and defend to their boss.

Write that definition first. Pressure-test it against the give-up case, the partial win, and your own incentive to fudge it. Then, and only then, put a dollar sign in front of it. The winners in AI pricing won't be the ones who switched to outcome-based pricing earliest. They'll be the ones who could say, in one honest sentence, exactly what an outcome is — and mean it.