Measuring AI Adoption in Your Teams: The Indicators That Matter
To measure AI adoption in a team, track three things: who uses it each week, for which tasks, and how independently. The number of licenses bought and the number of logins say almost nothing, and tracking each employee by name creates more problems than it solves.
This article complements the complete guide to AI in business for SMEs and builds on the article about AI KPIs in business, which proposes six management indicators. Here, we look only at adoption: what teams actually do, and how to measure it without turning the tool into a surveillance system.
Why aren't public figures enough?
Because they measure organized use at company level, not what your teams do day to day. According to Insee (Insee Première no. 2120, published 21 July 2026), 18% of companies based in France reported using at least one AI technology in 2025, 8 points more than in 2024. The scope is specific: companies with 10 or more employees, excluding agriculture and the financial sector, and the survey counts organized use of AI, not occasional individual use by employees.
That detail matters for you. In an SME, a large share of usage goes through personal accounts, tabs open next to the work, free tools. No national statistic will tell you whether the sales team uses AI three times a day while accounting has never touched it. You have to measure it yourself, starting by deciding what "adopting" means for you.
What does "adopting AI" mean for a team?
Adopting means building the tool into a recurring task, with a result the person checks and takes responsibility for. Opening a chatbot once a month to reword an email is not adoption. Using an assistant every week to prepare meeting minutes, reviewing the output and sending it as is, is.
Three levels help you place a team:
- Trial: the person has tested the tool, sometimes once. This is a starting point, not a result.
- Regular use: they use it every week for at least one identified task.
- Autonomous use: they can adapt their instructions (see prompt engineering), spot a doubtful answer (see AI hallucination) and decide when not to use the tool.
This grid is deliberately simple. You can fill it in ten minutes with a team lead, no software needed. It draws on the logic of AI maturity levels, applied to a team rather than the whole company.
Which indicators should you track?
Five indicators cover most of it: active usage rate, regularity, depth, autonomy and perceived value. The table below shows how each is calculated and where the data comes from.
| Indicator | What it tells you | Calculation | Data source |
|---|---|---|---|
| Active usage rate | How much of the team uses it | People who used the tool this week ÷ people trained | Admin console (aggregated figure) or short survey |
| Regularity | Whether usage lasts | Share of people active at least 3 weeks out of 4 | Same source, monthly follow-up |
| Depth | For which tasks | Average number of different tasks per active person | 5-question survey |
| Autonomy | Whether the person manages alone | Share of people who adapt their instructions and check results | Team discussion, three-level grid |
| Perceived value | What helps or blocks | Score from 1 to 5 on usefulness and trust, plus free comment | Anonymous survey |
Active usage rate is the easiest to collect, and also the most misleading on its own. A team can show 80% active use with a single automated task: everyone rewords emails, nobody has changed how they work. Depth is what reveals it.
For the calculation, use an honest denominator. If 12 people out of 40 have been trained, the rate is computed on 12, not 40. Otherwise the figure drops mechanically with each new training cohort and you will believe in a decline that does not exist.
How do you measure without monitoring each employee?
Measure at team level, with aggregated data, and say clearly what it is for. It is the safest choice legally, and also the most effective: someone who feels watched hides their usage, which worsens the shadow AI problem you are trying to reduce.
The framework comes from the French data protection authority (CNIL) for any system monitoring employee activity. A system must be proportionate and pursue a defined purpose, with no less intrusive means available. The CNIL cites Article L. 1121-1 of the French Labour Code on this point. Employees must be informed before it is introduced (Article L. 1222-4), and the works council (CSE) must be consulted beforehand (Article L. 2312-38), which is mandatory from 50 employees. The CNIL generally considers constant monitoring excessive. If you are unsure how usage logs from an AI tool are treated in your company, check with your DPO or legal counsel.
In practice, four simple rules:
- Prefer dashboards aggregated by team over per-person reports.
- Never use these figures in an individual performance review without announcing and framing it first.
- Announce what you measure, why, and how long you keep the data.
- In an organization with 50 or more employees, inform and consult the CSE before starting. Your AI policy is the right place to put it in writing.
How do you gather team feedback?
With five questions, asked every two months, anonymously. Beyond that, response rates drop. Here is a starting template to adapt:
- Which tasks did you use AI for this month?
- Which task would you like to hand over to it but don't dare to, or don't know how?
- On a scale of 1 to 5, how far do you trust the results?
- What holds you back most: lack of time, training, a suitable tool, clear rules?
- Have you ever drafted something with AI that you ended up redoing completely?
Question 5 is the most useful and the least asked. It measures the hidden cost of adoption: time spent fixing unusable output. If half the group answers "yes", the problem is not adoption but the quality of instructions, which means training.
How do you read the results and act on them?
Read indicators in pairs: a single value tells you little. Here are the combinations that come up most often.
- High active use, low depth: the tool serves a single gesture. Pick two other high-volume tasks and show them in a workshop.
- Low active use, positive feedback: the people trained are convinced but few. Extend training before changing tool.
- Active use falling after the first month: novelty effect. Look for what blocks people (access, time, unclear rules) before launching another communication campaign. The topic is covered in detail in the article on AI change management.
- High use, low autonomy, overconfidence: a risky situation. Teams use the tool without checking. Add a review rule and a session on common errors.
Do not look for a universal threshold. Figures sometimes circulate in the trade press, but they depend on the job, the tool and how people count. Set your own baseline before the pilot, then compare yourself with yourself month on month.
What rhythm and who owns this tracking?
A twenty-minute monthly review, led by an identified owner, is enough for an SME. More often, you spend more time measuring than acting. Less often, you miss the drop-off in month two.
The owner does not need to be technical. They need to know the teams, ask the right questions and report to management. In small organizations it is often the manager or the person in charge of HR. If you are unsure about this role, the article on hiring an AI lead details the profile, and the one on AI governance in SMEs shows how to fit it in without adding weight.
A sample agenda for the monthly review: read the five indicators (5 minutes), cross-reading (5 minutes), one decision (training, rule, tool) and an owner to carry it out (10 minutes). One decision a month beats a twelve-action plan nobody follows.
Which common mistakes distort the measurement?
The three most common: counting licenses, measuring too early and measuring without a baseline.
- Counting licenses: 30 licenses bought does not mean 30 users. Only real usage counts, measured over a given week.
- Measuring too early: one week after training, everyone is trying it out. The useful figure comes at 6 to 8 weeks, once the novelty has worn off.
- Measuring without a baseline: without an initial reading (time spent on the task, existing usage including undeclared use), you cannot show what changed, or link it to return on investment from AI.
Another, quieter trap: confusing adoption with results. A team can adopt a tool massively and gain nothing. Adoption is a condition, not proof. To link the two, see the article on AI productivity figures in SMEs and the definition of AI ROI.
Frequently asked questions
What is a good AI adoption rate in a team?
There is no reliable benchmark that fits every SME. Set your own target after a first reading: for example, moving from half of trained people active each week to a large majority within three months. A 100% rate is neither realistic nor desirable, some tasks do not need AI.
Do you need a tool to track AI usage?
Not at the start. The admin consoles of professional tools often provide an aggregated usage rate, and a five-question survey completes the picture. Only add a tracking tool beyond about twenty users, and after framing the information given to employees and the CSE.
Can you track each employee's AI usage individually?
It is technically possible, but risky. The CNIL requires a proportionate system, prior information for employees and, where applicable, consultation of the CSE. Named tracking also pushes teams to hide their usage. Favor team-level data.
How often should you measure adoption?
Once a month for usage indicators, every two months for perceived value. The first useful readings come 6 to 8 weeks after a training session, not before.
What if adoption stalls after training?
Find the cause before acting: lack of time, unclear usage rules, a tool poorly suited to the tasks, or weak command of instructions. A refresher session on the team's own cases often works better than another general training.
How do you measure adoption if teams use unapproved tools?
Simply ask, in the anonymous survey, and treat the answers without sanctions. These uses reveal real needs. The article on shadow AI explains how to frame them rather than ban them.
How GrowthPerf helps
GrowthPerf is a Qualiopi-certified training provider. We help SMEs and nonprofits define what "adopting AI" means for each team, set up a simple first reading, and adjust AI training for business according to what the figures show. Hands-on programs such as operational AI for SMEs are built around participants' real tasks, which makes measurement easier afterward.
To move forward, the simplest way is to start from your teams and your current tools. Find the overall framework in our guide to AI in business, then use the list of indicators above for a first reading.
Next step: get the adoption survey (5 questions, ready to send) and take 30 minutes with us to review it, no commitment.