Hourly billing rewards the wrong thing. It rewards hours, not outcomes — and in AI work, the gap between hours and outcomes is wider than in any other kind of software. A prompt iteration that takes ten minutes can be worth more than a week of plumbing, and a week of plumbing can be necessitated by a ten-minute decision. Billing the hours misprices both.
So we work fixed-scope, milestone-based. The client knows the total cost before we start, and we get paid when working software lands — not when the clock runs. This changes the incentives in ways clients feel immediately. We're motivated to kill bad ideas fast instead of billing through them. The client is motivated to make decisions instead of deferring them, because the scope is the contract.
It also forces honesty upfront. You can't price a fixed scope without understanding the problem, so the scoping phase becomes real work: we interrogate the data, the integrations, the success criteria. Agencies that bill hourly can afford to "discover" all this mid-project — on your dime. We can't, so we do it before you pay for build.
The objection we hear: "But AI is experimental — how can you fix the scope?" Fair. The answer is that we fix the scope around the slice, not the science. Milestone one is always the riskiest assumption, attacked first. If the data turns out to be unusable or the accuracy bar unreachable, you learn that in weeks, for a fraction of the budget — not in month five. Fixed scope doesn't mean fixed thinking; it means the experiment has a budget and a deadline.
There's a second, subtler effect: fixed pricing makes us say no. We've turned down projects where the scope couldn't be made concrete, because we'd be gambling with our own margin. That "no" is valuable information for the client — it means "this isn't ready to build yet," which is cheaper to hear than to discover mid-project.
If you want an agency incentivized to ship rather than to linger, that's how QuantaloomAI works: fixed scope, milestones, no hourly meter running.
Field note: A typical 14-week engagement breaks into four milestones: prototype the riskiest assumption (weeks 1–3), build the working slice (weeks 4–7), harden and integrate (weeks 8–11), deploy with evals and handover (weeks 12–14). Payment tracks milestones, so if we ever parted ways mid-project, the client keeps working software for what they paid. When scope changes mid-flight — and it does — it's a written change order with a price and a timeline impact, decided in days. Clients tell us this discipline actually speeds them up: knowing changes cost something makes them prioritize, which is the behavior that ships products.