Mentivis OS OpenMentivis OS EntrepriseMentivis OS Edu
Contact usDemo
All articles
M · AI & TrainingGeneral

Published 9 September 2026 · Updated 10 October 2026 · By Marie Castelli, Partner, Mentivis

How to choose an AI training system: the right questions to ask

Choosing an AI-based training system means examining four dimensions: functional (does the system generate learning paths or select from a catalogue?), technical (integrations, hosting, model used), legal (GDPR, AI Act, informing employee representatives) and organisational (deployment time, maintenance, management). The most telling question concerns personalisation: real recalculation based on what has been learned, or simple predefined paths.

Pillar: Beyond the LMS: understanding and choosing your training tools

Why is this market hard to read?

The same promises (personalised paths, adaptive AI, skills measurement) cover very different technical realities from one vendor to another, and demonstrations are built on favourable cases. A rigorous choice means distinguishing what is demonstrated from what is announced, by asking precise questions and checking the answers on your own data.

Which functional questions to ask?

Does the system generate content or merely distribute it?
A system that produces content from a competency framework differs from a platform that distributes an existing library. Generation makes it possible to cover specific needs, but requires content to be validated by experts.

Is personalisation dynamic or based on predefined paths?
Conditional branching ("if the learner gets it wrong, they go to module B") remains a fixed path. Dynamic personalisation recalculates the path at each assessment according to what has been learned. The demonstration should show this mechanism on a real case.

Is the measurement of learning built into the path?
An end-of-module quiz mainly measures immediate recall. Measurement spread over time, which triggers an adjustment of the path, gives better information on actual acquisition.

Can the system take on your existing competency frameworks?
Ask how long it takes and who does it.

Which technical questions to ask?

  • Integrations: which interfaces with the HR information system and existing tools are used in production by customers?
  • Hosting: where is the data hosted, and which subcontractors does it pass through?
  • Language model: which one is used, and is learners' data used to train it?
  • Availability: what service level is contractually guaranteed?

Which legal questions to ask?

GDPR. The vendor normally acts as a processor: Article 28 of the GDPR requires a contract governing the processing and allows only processors that provide sufficient guarantees. The CNIL (the French data protection authority) has published guidance on applying the GDPR to AI systems, notably on data minimisation and automated decisions.

AI Act. Systems intended to evaluate learning outcomes, steer the learning process or determine admission in education and vocational training institutions are among the high-risk systems listed in Annex III, with obligations applying from 2 December 2027. Ask the vendor how it classifies its system and what human oversight it provides. Since 2 August 2026, learners must also know that they are interacting with an AI (Article 50).

Labour law. In French companies with at least 50 employees, the introduction of new technologies and tools that allow employees' activity to be monitored fall within the information and consultation remit of the CSE, the employee representative body (Articles L2312-8 and L2312-38 of the Labour Code). A system that tracks learners' activity may be concerned.

Qualiopi. For a training provider holding Qualiopi certification (the French quality label for training providers), ask how the system produces the data needed for outcome indicators.

Which organisational questions to ask?

  • Actual deployment time and cost, observed at comparable customers rather than estimated.
  • Configuration and maintenance workload: who updates the frameworks and validates the content?
  • Day-to-day management: who in the organisation administers the platform?

What should be avoided?

Choosing on the basis of a demonstration prepared by the vendor: insist on a demonstration using your own data and frameworks. Choosing on the number of features announced without checking which are in production. Choosing without involving IT, HR, finance and, where applicable, the data protection officer.

Frequently asked questions

Is a call for tenders necessary?

For a public buyer, public procurement rules apply. For a private company, nothing requires it, but it usefully structures the comparison.

How can the soundness of a young vendor be assessed?

By asking for customer references in production, case studies with measured indicators and information on its financial position.

Is a pilot needed before a full rollout?

It is recommended. A pilot on a limited scope reveals difficulties a demonstration does not show: quality of generated content, integrations, learner buy-in.

Is an AI training system necessarily "high-risk" under the AI Act?

No. Classification depends on use: Annex III covers, in particular, the evaluation of learning outcomes and admission in education and vocational training institutions. The analysis must be made case by case with the vendor.

Sources

Further reading

Related products: MentivisOS Entreprise, MentivisOS Education

What AI cannot do in training
Adaptive learning: definition, mechanisms and evidence