Commercials
When AI does more of the work, what exactly are you paying for?
Rate cards and headcount pricing assumed that effort meant output. AI has broken that link. A look at how software delivery should be priced now, for buyers and for services firms.
For decades, most software services have been bought by the hour or by the head. It was never a perfect proxy for value, but it worked well enough because output roughly tracked effort: more engineers, working more hours, produced more software. AI has broken that relationship, and most commercial models have not caught up.
The link between effort and output is gone
With AI tools, the same engineer can produce several times more code on one task and less than before on another, because of the time spent checking and correcting what the tool produced. Output per hour has become highly variable. That creates problems on both sides of the table.
- Buyers cannot tell how much of what they pay for is AI-assisted, or how well. If a vendor’s team uses AI to finish faster but bills the same hours, the buyer funds the inefficiency.
- Providers that use AI well are penalized. Every hour saved is revenue lost, so the incentive is to adopt AI slowly, or quietly.
- Fixed-price work feels riskier for everyone, because estimates built on historical effort no longer hold.
Who carries the risk under each model
Every pricing model is really a decision about who carries delivery risk. When effort becomes unpredictable, that decision matters more than ever.
Indicative share of delivery risk carried by the buyer under each model.
- Time and materialsYou pay for hours. AI gains go to whoever controls the hours.
- Managed capacityYou pay for a team. Output per team member is rarely measured.
- Fixed priceProvider carries effort risk, and prices it in. Change is painful.
- Outcome-basedRisk is shared and tied to results both sides can verify.
Our view, for illustration. Actual risk depends on contract terms.
| Model | Works well when | Breaks down when |
|---|---|---|
| Time and materials | Scope is genuinely unknown and short-lived. | AI changes productivity and nobody measures output. |
| Managed capacity | You need a stable, long-running team. | Team size is the only lever, and output is not tracked. |
| Fixed price | Scope is clear and stable. | Estimates rest on pre-AI effort data, and change is frequent. |
| Outcome-based | Outcomes can be defined and verified by both sides. | Measures are vague, or one side controls all the evidence. |
Defining a unit of value
Outcome-based pricing only works if both sides agree what they are buying. In software delivery, useful units of value are concrete and verifiable:
- a feature released to production and accepted by the business;
- a module modernized, with tests and documentation in place;
- a defined reduction in lead time, defects or incidents over a period;
- a production-readiness audit completed, with findings fixed and verified.
What matters is that the unit is something the buyer values, and that its delivery can be shown with evidence both sides trust: delivery metrics, quality measures, test results, release records.
How to move from hours to outcomes
Few organizations can switch overnight, and they should not try to. A staged move works better.
- 1Measure firstAgree a baseline for lead time, quality and throughput before changing terms.
- 2Pick a pilotChoose one stream of work with clear, verifiable outcomes.
- 3Price the unitSet a price per unit of value, with a fair band for risk on both sides.
- 4Share the evidenceGive both parties the same view of delivery and quality data.
- 5Scale or adjustExtend to more streams once the model has proven itself.
If you sell services
For IT services firms and capability centers, this is not optional. Clients are already asking why AI has not reduced their bills. The firms that reprice first, around outcomes they can stand behind, will keep their margins and their clients. Those that defend the hour will be repriced by procurement instead.
That means reworking rate cards and offerings, rebuilding estimation on AI-era data, reshaping teams around senior pods rather than large pyramids, and investing in the delivery evidence that outcome-based terms depend on.
The bottom line
Paying for hours made sense when hours predicted output. They no longer do. The organizations, on either side of the table, that move first to outcome-based models backed by shared evidence will set the terms for everyone else.
If this sounds familiar