Method
Every stage ends with an outcome management can verify.
No stage begins before the previous one is approved. Investment grows only as certainty about the result grows.
The four stages
01
Assessment
Processes, data and existing systems, mapped with the people who manage them. Use cases are identified, and those that do not hold up are discarded.
OutcomeUse case map and priorities
02
Design
Architecture, approval workflows, success indicators and running costs, defined before any code is written.
OutcomeSpecification approved by management
03
Delivery
Short releases, each tested on real cases against an acceptance criterion agreed in advance.
OutcomeSystem accepted on the company's data
04
Operation
Quality monitoring, action logs, updates as models, processes and regulations change.
OutcomeSystem governed over time
Commitments
The rules that do not change from one project to the next.
01
A person approves
Every output that leaves the company goes through an accountable person. No exceptions.
02
Declared sources
Every answer states the document it comes from. An answer without a source is treated as a hypothesis.
03
Measured before use
Every system is tested on real cases with expected outcomes, and goes into service only at the agreed level.
04
Every action logged
Requests, sources, actions and corrections are logged. It is always possible to reconstruct why the system acted as it did.
05
Data within the perimeter
Documents and archives stay in environments chosen by the company. No data is used to train third-party models.
06
Owned by the company
Code, configurations, instructions and test cases belong to the company, by contract.
Security and data
Designed to pass review by the IT department and the legal team.
- Environments
- Systems run in the company's own cloud infrastructure or in dedicated environments, as decided by the IT department.
- Models
- Model providers selected with a contractual commitment not to use data for training. The choice of model is justified for each use case.
- Access
- Corporate authentication and permissions inherited from the source systems. Every access is logged.
- Exit
- No vendor lock-in: every system is documented so that it can be maintained by others.
First meeting
A confidential conversation with senior management.
To understand where artificial intelligence can improve your operations, and where it is not worth it. A non-disclosure agreement can be signed before the meeting on request.