Turn an intention into testable hypotheses.
Describe the task, the people concerned, possible errors, review workload and responsibilities.
Field use · Deployment support
Deploy AI without discovering its effects after the fact.
Define the use before implementation, organise a reversible pilot, then compare promises with effects actually observed in the work.
18 questions · about 10 minData stored locallyArchivable report
An equilibrium still open
The Arcadian or Pastoral State · 1834
The landscape seems peaceful and already organised, yet its equilibrium is still being built. The work reminds us that, before deployment, the choices still open — need, participation, limits, follow-up and stopping criteria — determine the strength of what follows. View the artwork ↗
Test on a small scale, observe the real work and remain able to reverse course.
A useful pilot has a defined scope, a baseline, success criteria and reasons for stopping.
Compare promises with the work actually performed.
Measure final quality, time saved or shifted, autonomy, skills, team dynamics and warning signals.
01 / Two moments, two tools
Document the decision beforehand. Verify the effects afterwards.
The two questionnaires answer different questions. The first identifies what the project may change; the second examines what actually changed once the tool was in use.
Before deployment18 questions · ≈ 10 min
Assess the project’s potential impact.
Use it when defining the project, selecting a solution or preparing a pilot. The questionnaire combines the six psychosocial risk factors in the Gollac framework with three AI-specific mechanisms.
Output 01Nine-dimension profile
Output 02Priority issues
Output 03Exportable action plan
After deploymentDated follow-up
Observe the effects in real work.
Use it during a pilot or after go-live. Follow-up compares the initial objectives with the experience of people who use, review or are affected by the system’s outputs.
ObserveQuality and errors
CompareWorkload and autonomy
DecideContinue, correct or stop
02 / Before deployment
Six decisions to make before launching a pilot.
The questionnaire does not replace collective decision-making. It makes explicit the choices too often buried in a vendor presentation or productivity promise.
| Decision | Question to resolve | Record to keep |
|---|---|---|
| 01Actual need | What work problem are we trying to solve, for whom and with what expected outcome? | Description of the current task and project objective |
| 02Scope | What the tool may do, what will remain human, and which uses are explicitly excluded. | Authorised use cases, limits and population concerned |
| 03Data and version | Which data enters the system, where it flows, and which exact version will be tested? | Sources, retention rules and version record |
| 04Review | Who will review the outputs, with what skills, information and available time? | Named owner, review procedure and planned workload |
| 05Participation and appeal | How will workers, their representatives and occupational health be involved? Who can challenge an output? | Participation, reporting and appeal arrangements |
| 06Reversibility | Which errors or effects will trigger correction, suspension or termination of the project? | Decision thresholds and fallback to the previous way of working |
The record that matters
At the end of project definition, everyone should understand what will and will not be tested, who decides, and under which conditions the experiment can be stopped.
03 / Organise the pilot
A pilot is not a demonstration under favourable conditions.
It tests a hypothesis in real work, including incomplete information, atypical cases and time pressure.
Describe the situation without the tool
Quality, lead times, workload, incidents, room for manoeuvre and existing difficulties. Without this baseline, no gain can be attributed to the project.
Include ordinary work and exceptions
Test frequent, rare, ambiguous and potentially serious cases, as well as situations in which the system should recognise that it cannot answer.
Set a duration and scope
Limit the population, data and decisions concerned. Maintain a fallback and avoid irreversible dependency during the trial.
Review before expanding
Bring together results, user experience and occupational health signals to decide whether to continue, modify, suspend or abandon the project.
What a vendor demonstration cannot measure.
- Time required to review and correct
- Additional work created by exceptions
- Effects on mutual support and skills
- The practical ability to challenge an answer
- Errors specific to your data and occupation
- How objectives change once the tool is installed
04 / After deployment
Track effects hidden by an average performance figure.
An overall improvement can hide workload shifted onto certain occupations, loss of autonomy or rare but serious errors. Follow-up must therefore combine measures with field feedback.
DimensionWhat to observeSignal requiring action
01Final quality
Errors, omissions, rework, incidents prevented or created, and variation between cases.
A serious, repeated or hard-to-detect error.02Time and workload
Production time, but also review, correction, learning and exception-handling time.
Review cancels out the gain or is performed under time pressure.03Autonomy and appeal
Ability to depart from the output, explain a decision and obtain a review.
The recommendation becomes an instruction that cannot be challenged.04Skills
Opportunities to learn, maintenance of independent practice and ability to identify an incorrect output.
People can no longer act without the tool or check its output.05Work collective
Task allocation, cooperation, knowledge transfer and clarity of responsibilities.
Difficult work becomes concentrated among a few people or becomes invisible.06Warnings and health
Reported difficulties, fatigue, tensions, sickness absence and situations in which use must be interrupted.
Warnings increase without a response or a way to stop.Indicative schedule to be adapted to the pilot duration, risk level and pace of change in the work.
05 / Use the tool in your organisation
A shared basis for discussion, not a score imposed on people.
The questionnaire can be used directly. For a collective approach, a pilot or adaptation to your context, an initial discussion helps select the right scope and participants.
Employers and executives
Document the decision, resources, responsibilities, pilot conditions and criteria for continuing or stopping.
HR and transformation teams
Examine redesigned tasks, actual workload, skills to preserve, training and employment consequences.
Employee representatives and works councils
Use a shared framework to examine working conditions, request relevant information and agree follow-up.
OHS services and prevention professionals
Become involved during project definition, identify exposure situations and help choose collective prevention indicators.
Guided access · Collaboration
Would you like to use the tool in a company, works council or OHS service?
Briefly describe the project, its stage, the population concerned and the intended use. Together, we can determine the most appropriate way to use the tool.
Contact Dr Charles Broutin on LinkedIn ↗06 / Method & limits
What the questionnaire can do — and what it does not claim to do.
The tool’s value depends on the quality of the discussion and information entered. Its result is neither a compliance label nor automatic authorisation to deploy.
Exploratory tool
It structures analysis and highlights points requiring attention. It has not been validated as a psychometric test.
No identifying data
The questionnaire asks for no person, company or provider names. Answers remain on the device.
No worker scoring
The profile concerns a project and work situation. It must not be used to rank or individually assess workers.
No substitute
It replaces neither prevention duties, social dialogue nor specialised legal, clinical or technical analysis.