Measure progress in training, not employees: track how people meet their own practice goals and which skills the team needs to rehearse, and do not use this as a basis for pay, promotion or disciplinary steps. The difference matters both in practice and in law.
Please note. This text is general orientation, not legal advice. Before you start tracking data about employees, consult the responsible person in your company.
Why the line matters
- The legal view. Under the EU Artificial Intelligence Act (AI Act), AI systems intended to make decisions on terms of work, or to monitor and evaluate the performance and behaviour of people in work relationships, are classed as high-risk (Annex III, point 4(b); text of the annex). The Act also prohibits inferring employees' emotions in the workplace, apart from medical and safety purposes (Article 5(1)(f)). A training tool used as an employee assessment can end up in this category.
- The GDPR. Article 22 protects people from decisions based solely on automated processing that have legal or similarly significant effects. Let people decide about people.
- The practical view. When a number becomes the target, people start optimising the number instead of the skill. Training that turns into an exam stops being a safe place to make mistakes.
What to measure for learning
The “four levels” framework for evaluating training, attributed to Donald Kirkpatrick (reaction, learning, behaviour, results), helps to separate what exactly you are tracking. For training salespeople two levels matter most: learning and behaviour.
| Indicator | Who sees it | What it is for |
|---|---|---|
| Own practice goal met (yes/no, short description) | The salesperson | Focuses practice on one observable skill |
| Number and regularity of practice rounds | The salesperson, possibly the team in aggregate | Whether practice actually happens; not a measure of quality |
| Self-rated confidence before and after practice | The salesperson | Shows where support is missing |
| Observed behaviour (checklist: a clarifying question asked, a summary given, a refusal respected) | The salesperson and mentor | A concrete, checkable change |
| Topics the team practises most (for example price, postponing a decision) | Managers, in aggregate | Where to focus training; no names |
| Feedback on the training itself | The training lead | Improving the training, not rating people |
Be careful especially with small teams: even an aggregate can reveal individuals. Better not to show a breakdown if it would show who is who.
How to put it in place
- Say in advance what the data is for. And what it is not for. Write it down and tell people.
- Let the salesperson own their data. They see their own progress first and decide what to share with the mentor.
- Share in aggregate. Managers see topics and participation of the team, not rankings of people.
- Do not rank people by practice score. A leaderboard turns training into a contest and rewards playing the score.
- Do not link training to pay and career. If you decide about people, rely on human judgement and several sources, not a single number from practice.
- Collect as little as possible. Do not store what you do not need. Set a retention period.
- Involve employee representatives where the rules require it.
Employment law and monitoring
If training processes data about identifiable employees, rules on workplace control and monitoring may apply, and employee representatives may have information or co-determination rights. These differ from country to country. Check the employment law and consultation duties where you operate, and take advice. For the general principles see also How we score.
How DIRI AI handles this
DIRI AI is intended for training. It is not offered or designed for recruitment, pay, promotion, disciplinary action, dismissal or the monitoring and evaluation of employees' work performance; a practice score is not a basis for decisions about people. The participant sees their own report and can ask for human review. Details are on the How we score page.
Related articles: How to practise sales objection handling, Turn feedback into a practice goal and AI role-play vs practising with a colleague.
Frequently asked questions
Can a practice score be used in employee reviews?
We do not recommend it and DIRI AI is not intended for it. The score describes one practice call against fixed rules and has not been validated as a basis for decisions about people. Such use is outside the intended purpose.
What do I tell management that wants to see “training results”?
Show participation, topics and changes in observed behaviour in aggregate and without names. If the company wants to follow sales results, it does so within ordinary sales management, separately from training and with the relevant legal steps.
How do I measure whether improvement carries over into live calls?
A sensible way is human observation: a mentor listens to a live call, where this is legally and organisationally in order and the employee knows about it, and compares it with the practice goal. It is judgement, not an algorithmic score.
Do I have to inform employees?
Yes, as good practice always; in many countries the rules also require it. The rules differ by country and type of data; check them with the responsible person in your company.
A note on sources
References to the AI Act are for orientation and the rules may change. The Kirkpatrick levels are a general model for evaluating training, not evidence of effectiveness. The table of indicators is an editorial proposal. This article does not replace legal advice.
DIRI AI trains communication and provides a report after each practice call; the participant sees their report and the company controls who else may see it. Use for training, not for evaluating employees, is the intended purpose.
Try one call on the free Free plan
The Free plan includes one voice call a day of up to 5 minutes, no payment card; e-mail confirmation is required. Check current terms before you start.