Business analysis · AI projects
AI changes the solution.Not the analytical rigour.
How I read an AI project as a Business & Functional Analyst.
Projects that embed AI raise new questions: uncertainty, quality of outputs, human oversight, data, risks and monitoring over time.
Here is the framework I would use to address them, drawing on my experience as a Business & Functional Analyst.
My approach
5 steps, from business need to operations.
The framework I would use to structure a project that embeds AI, with decision points at every step.
- 01
Scope
Does the problem really call for AI?
- 02
Design
How do we fit it into the process?
- 03
Evaluate
How do we define what “good enough” means?
- 04
Deploy
How do we bring it into service under control?
- 05
Operate
How do we check that it stays good over time?
Zoom 01
Designing AI into the process.
An AI solution is more than the model. The analysis must specify where AI steps in, where people stay in control, which rules apply and what happens in case of doubt or unavailability.
Zoom 02
Defining and measuring what “good enough” means.
With a probabilistic system, acceptance testing cannot rely on ✓ / ✗ cases alone. Quality must be measured on a representative set of situations, with metrics, thresholds and some zero-tolerance cases.
Example acceptance criterion
| ID | AI-AC-014 |
|---|---|
| Objective | Correct routing of requests |
| Metric | ≥ 95% correct answers |
| Evaluation set | 300 real, representative cases |
| Segmentation | FR / NL measured separately |
| Critical cases | 0 errors tolerated (to be escalated) |
Indicative gain example (per case)
Actual gain: 3 min per case (−25%)
Sector focus
In insurance, in practice.
AI cannot be analysed outside its regulatory context. Depending on the use case (underwriting, claims handling, assistance, document processing…), the bar can be higher: sensitive data, bias, explainability, human oversight and obligations under the AI Act and sector requirements (EIOPA, etc.).
- Use casesUnderwriting, claims handling, assistance, document processing
- Key concernsData, bias, explainability, human oversight, traceability
- Regulatory frameworkAI Act (risk classification), GDPR, sector requirements (EIOPA, governance, etc.)
In summary
An analysis framework, not a claim of technical expertise.
This framework draws on the principles and good practices I have studied and consider relevant for approaching a project that embeds AI. It translates the specifics of AI into my own field: understanding the need, designing the process and the functional solution, defining verifiable requirements and supporting delivery, drawing on my experience as a Business & Functional Analyst, particularly in insurance.