One of the world's largest consultancies now takes roughly a quarter of its global AI client fees on an outcome basis rather than time and materials. Two years ago that figure was close to zero. It is the clearest evidence yet that clients have stopped accepting the traditional AI engagement structure - pay for hours, hope for value.
We think this is the right shift. We also think most organisations evaluating an outcome-based AI proposal for the first time are about to sign something they haven't properly scrutinised.
Why it's happening now
Two things converged. First, clients got burned. Enough £5 million programmes delivered nothing measurable that procurement teams started asking why they were paying for effort rather than results. Second, AI-accelerated delivery made time-billing actively perverse: a consultancy that ships a working model in six weeks instead of six months earns less under T&M, not more. That's a broken incentive, and the market noticed.
The "AI factory" model - repeatable delivery units with pre-built accelerators that stand up a programme in weeks rather than months - only really makes commercial sense under an outcome or platform-subscription structure. Nobody wants to bill fewer hours for doing better work.
What "outcome-based" actually requires
The label gets applied loosely. A genuine outcome-based AI contract needs three things a conventional SOW doesn't, and the absence of any one of them is where these deals go wrong.
1. A rigorous, agreed baseline
You cannot get paid for a 20% reduction in cost-to-serve if nobody agreed what the starting cost-to-serve actually was, measured how, and over what period. We've seen baseline disputes consume more negotiating time than the statement of work itself. Fix the baseline methodology in writing before the programme starts, not after the first invoice is disputed.
2. A metric that can't be gamed
Chatbot containment rate is the industry's cautionary tale here - trivially easy to inflate, and we've written before about what that costs you. An outcome-based fee needs a metric resistant to the obvious gaming move: measure resolution quality alongside volume, not volume alone, and make sure the incentive rewards what you actually want.
3. Honest attribution of control
If the outcome depends on data quality, change management, or a legacy system integration the vendor doesn't control, the contract needs to say so explicitly and price accordingly. Vendors who accept full outcome risk on variables they don't control either pad the fee heavily to compensate, or quietly under-deliver on the parts they can't influence.
What to do before you sign
- Insist on a fixed-fee discovery phase before any outcome clock starts - you cannot price outcome risk on a system nobody has assessed yet
- Get the baseline methodology in writing, agreed by both sides, before implementation begins
- Ask explicitly what happens if the outcome depends on something outside the vendor's control, and get that risk allocation in the contract, not implied
- Check the metric against the obvious way someone could hit the target without delivering real value
The bottom line
Outcome-based pricing is the right direction for AI programmes, and it's coming to your next renewal whether you've prepared for it or not. The firms that get value from it will be the ones that treat the baseline and the metric definition as seriously as the technical delivery plan - not the ones seduced by a headline that says "we only get paid if it works."