Measure first, then build. Working together has four phases, and each ends with something you keep even if we stop there.
1. Measurement (1–2 weeks)
I go through the process, a data sample and the constraints. Together we build a test set of real questions or cases that everything after it is measured on.
- Test set with correct answers
- Today's state in numbers: time, error rate, volume
- Written plan: what to build, what not to build and what it will cost
2. Prototype on your data (2–4 weeks)
I build a first version on a sample of your data and measure it on the test set. You see the numbers before the solution touches a customer.
- Working prototype
- Before-and-after numbers on the test set
- List of cases where the solution still fails
3. Deployment with approval (2–6 weeks)
I connect the solution to your systems with the rights it needs. The AI suggests; a person approves anything that touches money, customers or can't be undone.
- Solution in production
- Log of every AI step
- Documentation and team training
4. Operations (optional, ongoing)
I monitor quality and cost and keep tuning. A new model is measured on the same set first. We widen autonomy only where the numbers allow it.
- Monthly quality and cost report
- Model swaps only after verification
- Extension to further cases
What holds in every phase
- You see a working version every week.
- You get the price of the next phase in writing before it starts.
- The code, test set and documentation are yours.
- I flag scope changes before the budget moves.