How to Scope a Fixed-Price AI Project (and Why Most Firms Won't)

Open-ended retainers are easy to sell and easy to extend. They are also why so many AI engagements run long, cost more than expected, and never produce a clear deliverable. Every Aluf engagement is fixed-scope on purpose, and getting the scope right up front is most of the work.

Start with the decision, not the technology

The first question is never which model or which framework. It is: what decision or task does this replace, and who makes that decision today? A scope built around a specific decision, like approving a warranty claim or answering a technician's question, is concrete enough to estimate. A scope built around a vague goal like getting more value from our data almost always grows.

Write down what done looks like before any code gets written

A fixed-scope project needs a finish line everyone agrees on before the build starts. That means specific examples of inputs and the outputs that count as correct, the systems the tool needs to connect to, and who signs off that it works. If that list cannot be written down in a page or two, the project is not ready to be scoped yet, and more discovery is the right next step, not a bigger contract.

Build in the edges, not just the happy path

• What happens when the system is not confident in an answer.
• What happens when the data it needs is missing or out of date.
• Who reviews outputs before they reach a customer or affect a financial decision.
• What the rollback plan is if something needs to come out of production quickly.

Most scope creep does not come from the obvious requirements. It comes from edge cases nobody wrote down, discovered three weeks into a build. Naming them up front, even briefly, keeps the price and timeline honest.

Why a tight scope is a feature, not a limitation

A narrow, well-defined first engagement is not a smaller version of the dream. It is the fastest way to get something real in front of users, learn what was actually needed, and decide what to build next with real information instead of a guess. Most of our second engagements with a client look different from what either of us would have scoped on day one, because the first one taught us both something.