In March Forrester's Ted Schadler wrote that the bell tolls for time-and-materials pricing. Hourly billing has always rewarded providers for stretching a project out, he argues, and once AI does a growing share of the delivery, the hours stop measuring the work. "Time no longer has meaning as a unit of work," he wrote, "only results matter."
A provider that starts using AI and finishes in one day what used to take a five-day week has, under an hourly contract, just cut its revenue for that task by four-fifths.
Two models that point the incentive the right way
Fixed capacity. You pay a flat monthly fee for a team of a known size, and you set what it works on from a prioritised queue. Slow work earns the provider nothing extra.
Outcome-based. Payment depends on agreed results, such as hours of manual work removed or invoices processed without a person touching them.
Many contracts mix them. A capacity fee covers the team, and a smaller share depends on a few outcomes both sides can measure.
What makes capacity pricing work
Without a written queue, a flat fee drifts. Within a few months nobody remembers what the team was meant to deliver, and the fee starts to look like a cost with no product attached.
You own a queue of work items in priority order. Each item has acceptance criteria agreed before work starts, so "done" means the same thing to both sides. Someone reviews what shipped against the queue every month. And the exit terms are short and plain, so either side can stop without lawyers.
What makes outcome pricing work
Latent Bridge's write-up of HFS Research's 2026 New York summit lists where outcome pricing breaks: ambiguous requirements, scope that grows without anyone revisiting the price, fewer checkpoints along the way, and outcomes that depend on the client's own people.
The last of those is the easiest to miss. If an agent can cut support response times, but only once your team rewrites the help-centre articles it answers from, the provider doesn't control the result. Paying on that outcome puts the provider's fee in your team's hands, and a sensible provider will either refuse or price the risk in.
Pay on outcomes the provider mostly controls, and measure the baseline in writing before work starts.
Questions to ask before you sign
- If this work takes half as long as planned, who benefits?
- What does "done" mean for each item, and who decides?
- How will we see what shipped each month?
- Which outcomes depend on our own staff, and what does the price assume about them?
- How do we stop if it isn't working?
If a proposal can't answer the first question in your favour, ask for a different pricing model.
Frequently asked questions
Why is time-and-materials pricing a problem for AI work?
Hourly billing pays more when work takes longer. When AI shortens the work, a provider billing by the hour earns less for getting faster, so the incentive runs against the buyer.
What are the alternatives to hourly pricing for AI engineering?
Fixed capacity, where you pay a monthly fee for a team and set its priorities from a queue, and outcome-based pricing, where payment depends on agreed results. Many contracts mix the two.
When does outcome-based pricing work?
When the provider controls most of what produces a measurable outcome, and both sides wrote down the baseline before work started. If the result depends mainly on your own staff, the provider shouldn't be asked to carry the risk.
What should a fixed-capacity AI contract include?
A prioritised queue you control, acceptance criteria for each item, a regular review of what shipped, and simple exit terms. Those keep a flat fee tied to delivered work.