Method

How we work

Four phases with a reviewable deliverable in each one. We publish no timelines or prices: they are agreed in writing according to scope.

Four phases

Assessment

We map the current process, the data available and the real constraints. The result is a documented decision about what should be automated and what should not.

Deliverables
- Map of the process and its data
- List of risks and constraints
Design

We define the use case, the success metric and the system's autonomy limits. No agreed metric, no implementation.

Deliverables
- Use case and success metric
- Autonomy and cost limits
Implementation

We build in increments, with automated evaluation from the first day and human review at the points the process requires.

Deliverables
- Evaluated working version
- Quality report for the test set
Operations

We measure quality and cost in real use, review regressions and leave the contingency and rollback plan in writing.

Deliverables
- Dashboard of agreed metrics
- Contingency and rollback plan

Working criteria

Measured or not published

Every claim about quality rests on a defined test set.
Autonomy has limits

Every automated action declares its scope, its cost and its rollback path.
Data is treated as an asset

The data plan is agreed before implementation, not afterwards.

We start with the assessment

A first assessment clarifies what should be done and what should not.