How long until something actually works?

The 1-Day Working Session is the smallest useful commitment. It is a $2,500 fixed-price engagement designed for one question: is this idea worth building at all, and if so, what is the shortest path to proof? You leave with a next step you can act on that same week. That might be a scoped Discovery Sprint, a specific data-collection task on your side, or a decision not to build.

It exists because the wrong first spend on AI is a three-month exploratory project that produces slides. A day of senior engineering attention on the actual problem is a better filter.

The 2-Week Discovery Sprint is $18,000 and produces a scoped solution architecture and roadmap. It is the standard front door to a full build. Two weeks is enough to interview the people who will live with the system, look at real data (not a sample), map the integrations, and write a design document that a build team can execute against.

Discovery exists because most failed AI projects fail on the architecture, not the model. Choosing where the model runs, how the data flows, what the human review step looks like, and what "correct" is measured against — those are the decisions that either de-risk the build or condemn it. Two focused weeks with senior people is the right shape for that work.

The 4-Week Prototype Sprint is $48,000 and produces a working prototype against your real data. Not a demo, not a mock, not a wireframe: a functioning system that runs on your actual inputs and produces outputs you can evaluate against your accuracy bar. Four weeks is enough to prove the mechanism, calibrate the model against your data, and put a real reviewer step in front of the output.

By the end of the sprint you have evidence, not conviction, that the full build is worth doing — and if it is not, you have spent $48,000 to learn that, which is cheap. Full custom builds are scoped from what the prototype revealed, never quoted blind.