An Outcome Is Only as Honest as Its Definition

A few things we believe about outcome-based pricing, up front:

• Outcomes should be defined transparently. You should know exactly what you're paying for before you sign — in plain language, in writing.
• The best outcomes are rigorous and hard, as long as they map to real value. We'd rather be held to a strict bar than a flattering one.
• Not everyone plays it straight. Definitions across the market vary widely, and some of what we're seeing is, frankly, sketchy.
• We keep our platform fee small. Every dollar you pay before an outcome is delivered isn't tied to your result, so a large upfront fee quietly works against the whole model.
• We always make it easy for your customers to reach a human. By default, no exceptions.
• We don't think we have this fully figured out. If you have a sharper take, email me: michael@appliedlabs.ai.

A short history of how software gets priced

Software pricing has moved in waves, and each one matched what the software did at the time:

• Seats. The SaaS era priced per login — a license for every user, paid for whether they used it or not. It fit a world where software was a tool a person operated.
• Usage. As cloud infrastructure took over, pricing followed consumption: API calls, compute, tickets, conversations. Closer to value, but still a proxy for it.
• Outcomes. Now AI can do the work, not just assist with it — so the natural next step is to pay for the result, a resolved issue or a retained customer, rather than the seat or the usage behind it.

That shift is real, but it is not finished and it is not settled. By some industry estimates, only around 5% of software companies price on outcomes today. Attribution is genuinely hard, definitions are still being argued over, and plenty of vendors are using "outcome" as a fresh coat of paint over an old model. We're early — which is exactly why the definitions matter so much right now.

At Scale AI, I led our first price book built on quality rather than raw volume — part of a broader push to move AI companies toward pricing on quality and outcomes. The model itself was rarely the hard part. Defining what you're charging for, rigorously and honestly, always was.

Not all outcomes are equal

When a vendor says "outcome," it can mean very different things. Picture them as a ladder, weakest to strongest:

1. Involvement — the AI was in the conversation: it said something, surfaced an article, fired an automation. Pricing on this is paying for activity with a nicer label.
2. Deflection / containment — the conversation didn't reach a human. Better, but it counts a customer who gave up the same as one who got helped.
3. Resolution — the problem was solved, end to end, with no repeat contact. The first rung that's about the customer rather than the software, and the first that's hard to game — if it's defined strictly and verified, not assumed.
4. Value creation — outcomes that move the business itself: a save, a conversion, a recovered relationship. The hardest to attribute, and the most valuable.

The higher you climb, the more honest the pricing and the tighter your vendor's success is bound to yours. The catch: the lower rungs are the easiest to measure and inflate, so a lot of "outcome-based" pricing quietly sits at the bottom.

What counts as a billable outcome?

Here's the catch underneath all of this: terms like "automation rate," "resolution," and even "outcome" have no standard definition. Every vendor draws the line differently, and the fine print is where the real money is. So the single most useful thing you can do in an evaluation is make each vendor spell out, precisely, what triggers a charge.

The gaps can be wide. Take Fin's model: the headline is "you only pay when it resolves." But read closely, and a billable outcome also includes a multi-step "Procedure" that ends by handing the conversation off to a human — meaning you can be charged even when the AI passed the problem to a person instead of solving it. A number of legacy players do something similar: they market "outcome-based" pricing that, on inspection, is really charging for the AI's involvement, wrapped in an "automation rate" that's hard to decipher.

There's also a quieter trick in the denominator. You can make a resolution rate look great by measuring it only against the easy conversations you were always going to win. We don't — we track total conversations and measure our full AI resolution against all of them. It's a less flattering number, and the only honest one.

Resolution is better — but not bulletproof

Resolution is the most honest of the common metrics, but even it has a gap: most "resolutions" are assumed. The usual test is whether the customer came back. If they didn't, the conversation gets marked resolved. But silence isn't proof — some of those customers gave up, accepted a confident wrong answer, or quietly churned, and each one still lands on the dashboard as a win.

Our first answer to this is almost too simple: we make it easy for your customers to reach a human, by default, always. Burying the exit is the fastest way to manufacture soft resolutions, because a trapped customer who gives up looks identical to one who was actually helped. We'd rather leave the off-ramp wide open and only count what the AI genuinely closed.

Even with that off-ramp open, the numbers hold:

• >70% AI resolution, measured against every conversation
• >90% CSAT on AI conversations
• <0.1% reliability errors and hallucinations

The easy escalation doesn't lower the bar — it's how we keep the number honest.

The definition is a statement of incentives

Step back, and here's the real point: whatever a vendor counts is what it optimizes for. The definition isn't a billing detail — it's what your vendor will spend the next two years getting better at.

• Paid on involvement → it learns to put the AI everywhere, helpful or not.
• Paid on resolution → it learns to close the customer's problem and hand off fast when it can't.

You want a vendor whose only way to win is for your customer to win.

The frontier: raising the bar on outcomes

Resolution is the floor we hold ourselves to, not the ceiling. We want to be on the frontier of what an "outcome" can mean — in two directions.

First, verifying outcomes instead of assuming them: moving resolution from "the customer didn't come back" to something we can actually confirm. We're building toward that now, with more to share over the next few weeks.

Second, climbing the ladder toward the outcomes that grow a business — conversions, customer saves, recovered moments that turn into loyalty. As AI gets more capable, the outcomes worth pricing on get richer, and the vendors who matter will be the ones willing to be measured against them.

It's also why we keep our platform fee deliberately small — nowhere near the six-figure entry tickets common elsewhere. Every dollar you pay before an outcome lands is a dollar of our incentive that isn't tied to your result, and the bigger that upfront number gets, the less "outcome-based" the pricing really is, whatever the label says.

The goal is easy to state and hard to live up to: keep tightening the link between what we earn and the value our customers receive, until there's almost nothing left between them. That's the version of outcome-based pricing worth having — not a friendlier word for activity, but a real promise. If we're not creating value for you and your customers, we shouldn't get paid.

We're early, and we don't have all of this figured out. If you see it differently, I'd like to hear it — michael@appliedlabs.ai.