Outcome-based pricing for AI systems: how the number is set and measured

The short answer

Every Outwerk engagement has a named metric, a baseline, a target and a measurement method written into the statement of work before anything is built. The run fee for a month is due when the target is met. When it is missed for reasons within our control, that month's run fee is credited. The build fee, half on signing and half on acceptance on your real data, is separate and is not refunded.

Why price on a number at all

There are three ways to buy AI for operations. Tools and templates are cheap, and you run them. Consultants are priced on hours, so the incentive runs the wrong way: the longer it takes, the more it costs. An outcome price puts the incentive where the client wants it. We are paid when the hours come back, and not when they do not.

It also forces a discipline that benefits both sides. A workflow that cannot be measured cannot be priced this way, so it does not get built. That rule has saved more clients from bad projects than any technology choice.

Step one: the metric

A good metric is one the business already cares about, that the system can move on its own, and that can be read from the client's systems without judgment. Four we use most:

We avoid metrics the system does not control. Revenue is the clearest example: a system can make quoting faster, but the market decides whether the quote is accepted.

Step two: the baseline

The baseline is measured at the audit and confirmed in the first week of the build, from the client's own data. For hours, we time real occurrences rather than asking people to estimate; estimates run about half of the measured figure. For days and rates, we take the trailing three months from the system of record.

The baseline is written into the statement of work with the method used to measure it. If the method changes later, both sides sign the change.

Step three: the target

The target is set below what we expect the system to achieve, not at it. A workflow that should return 40 hours a month gets a target of 30. That margin covers the exceptions, which in our experience are 10 to 20 percent of occurrences (estimate), and it means the number is one we expect to meet every month, not most months.

For the operations retainer the target is set across the whole operation and grows as each monthly workflow is added. Month one might be 30 hours; month six, 180. The retainer fee is set against the running target, which is why the fee can step up as the scope does, and why the floor is USD 5,000 a month.

Step four: measurement

Measurement is monthly, from the client's systems, by a method both sides can repeat. The monthly report shows one line per workflow: the metric, the target, the month's figure, and whether it was met. A person on the client side can check every figure against the source.

What the report does not contain is a story. If the number was missed, it says so, and the credit appears on the next invoice.

The carve-outs, in plain terms

A missed month is credited when the miss was within our control. It is not credited when:

• the client did not provide data, access or approvals the statement of work required, and the delay caused the miss;
• a third-party system the workflow depends on was unavailable, and the outage caused the miss;
• the client changed the process the workflow was built around without telling us;
• the baseline itself was wrong because the data we were given was wrong.

Each of these is listed in the services agreement, and each has to be shown, not asserted. In practice the conversation at the end of a missed month is short, because the measurement method was agreed before the month began.

Two worked definitions

A distributor, quote to invoice. Metric: days from accepted quote to invoice sent. Baseline: 4.2 days, measured from the last 90 days of orders in the accounting system. Target: under 3 days on the monthly median. Method: timestamps on the quote record and the invoice record, median across the month's orders. Illustrative figures.

A law firm, intake. Metric: hours returned to fee earners a month. Baseline: 25 enquiries a week at 40 minutes of admin each, timed across ten real intakes, about 72 hours a month. Target: 50 hours returned. Method: enquiries handled end to end by the system multiplied by the baseline minutes, less the time logged on exceptions. Illustrative figures.

What this is not

It is not a share of savings. We do not take a percentage of what the system saves, because that requires auditing the client's finances and arguing about attribution. It is a fixed monthly fee with a condition attached.

It is not a guarantee that the technology works in general. It is a guarantee that this workflow, in this company, meets this number this month, or the month is on us.

How to start

The number is set at the busywork audit: thirty minutes, free, and you leave with the metric, the baseline and the target we would propose, in writing within 24 hours.