A survey
Asks people how often they use AI.
A snapshot of what people remember and are willing to admit. Send it out again next quarter.
AI usage monitoring
AI usage monitoring: measure AI adoption without tracking people.
AI usage monitoring is software that shows whether AI is being used in an organisation: how many people, how often, with which tools and for what. That question is about the organisation, not about a person, and the answer should be too. BeeSensible counts on the device itself and reports without a user id. An administrator sees that three people use AI daily, never who. Anyone searching for AI adoption software lands on the same question: are the licences being used, and what else is going on.
app.beesensible.eu/shadow-ai
AI tools in use
7
Newly discovered
2
The problem
Licences were bought. Whether they are used, nobody knows.
A board wants to know whether AI has arrived in the organisation and where the work is changing. The figures that exist come from one vendor's admin centre or from a survey. What is missing is a view of the whole organisation, and a way to get it without building a file on every employee.
The admin centre counts licences, not work
The Microsoft admin centre shows who has activated Copilot. It says nothing about ChatGPT, Perplexity or the AI feature inside a tool the team already had.
The survey measures what people remember
Whoever asked in spring which AI tools are in use has an answer about spring. And about what people were willing to write down.
Measuring looks like tracking, and tracking is not free
Many tools that measure AI use do so per person. That turns the measurement into a monitoring facility, with the GDPR and the works council attached. To know whether AI is used, none of that is needed.
Features
What to expect from AI usage monitoring.
The term covers everything from a licence counter to employee monitoring. These are the features that belong to measuring AI use, with an honest note on what BeeSensible does fully, partly and not at all.
| Feature | What it does | BeeSensible |
|---|---|---|
| Share of employees using AI | How many people use AI, set against the number of seats. | Yes The Adoption screen. Computed on the device, reported once a month as an anonymous count. |
| Frequency of use | How many people use AI daily, weekly or occasionally. | Yes Without a user id. The organisation sees "three people use AI daily", never who. |
| Prompts over time | How the number of prompts develops from week to week. | Yes Counts, no content. |
| Which tools | Which AI tools are in use, including the ones nobody requested. | Yes Domain only, at most once a day per device. |
| What AI is used for | The split across seven kinds of work: writing communication, marketing and content, summarising, rewriting and editing, translating, coding, knowledge and analysis. | Yes Cloud mode only. The first prompt of a conversation is classified and discarded straight away; only the label is counted. In local mode this never runs. |
| Share of use through allowed tools | How much of the AI use runs through tools the organisation has marked "allowed", with its trend. | Yes On the Risk screen. The figure that shows whether the policy lands. |
| Per-employee view | See who uses which tool and how often. | No Deliberately not. The usage streams carry no user id, so it cannot be reconstructed later either. |
| Productivity score per person | A number per employee for how much AI someone uses or produces. | No Deliberately not. The figures describe the organisation, never a person. |
| Exports and API | Pull the figures as CSV or PDF, or fetch them through an API. | No Not available. The figures live in the dashboard. |
What it does not do
Measure what the organisation does, not what a person does.
The line between measuring AI use and tracking employees sits in the design, not in a setting. This is where BeeSensible stops.
- No user id in the figures on AI use. Frequency is computed on the device and reported as an anonymous count; there is no table that says who.
- No content. No prompts, no page titles, no paths. For what AI is used for, only the label is counted.
- No per-person view and no productivity score. An administrator cannot click through to an employee anywhere.
- No per-team breakdown of AI use.
- No exports, no API, no digest emails. The figures live in the dashboard.
- No blocking. Measuring changes nothing about what someone can open.
The approach
See what is there first, then attach a number to a goal.
Measuring AI use is the first of four steps. The number only becomes worth something once a decision, a notice and substantiation follow it.
1 See
Measure what happens across the organisation
How many people use AI, how often, with which tools and for what. Counted on the device, reported without a user id.
2 Decide
A decision per tool
What you see asks for a choice: allowed, limited use, not allowed, or deliberately no decision yet. The organisation chooses.
3 Help
The decision at the moment of work
Someone who opens a tool with a decision behind it sees that decision in the browser, alternative first. Continuing is possible.
4 Substantiate
The trend after the decision
The share of AI use through allowed tools, with its trend, and how often a notice was followed. That is the figure a board can use.
and again How the cycle works
Why this approach
What usually happens today, and what it misses.
Four ways organisations currently try to get AI use into view.
Licence use from the Microsoft admin centre
Counts who has activated Copilot.
Sees one tool. The others stay out of view, and those are exactly the tools nothing has been decided about.
Employee monitoring
Records per person which tools someone opens.
Delivers more than the question needs, and collides with the GDPR and the works council.
Measuring nothing
Assuming the licences are being used.
Works until someone asks what the investment delivered.
Frequently asked questions
What people ask before they choose.
What is AI usage monitoring?
Software that shows whether and how AI is used in an organisation: how many people, how often, with which tools and for what. The term is also used for tools that record per employee what someone does. BeeSensible does the first and deliberately not the second.
How does BeeSensible measure how often people use AI?
The device itself keeps track of whether an AI tool was opened daily, weekly or occasionally this month. Once a month it reports an anonymous +1 for that bucket, with no user id. The organisation sees the distribution, never who is in it.
Can an administrator see who uses ChatGPT?
No. The figures on tool use and frequency reach the servers without a user id, and the dashboard has no drill-down per person. An administrator sees that a tool is in use and how much, not by whom.
How does BeeSensible know what AI is used for?
In cloud mode, the first prompt of a conversation in a major AI tool is classified into one of seven categories, such as summarising, translating or coding. The text is used only for that classification and discarded straight away; only the label is counted. In local mode this never runs, because no text leaves the device. The organisation decides with its analytics setting whether this measurement is on.
Is an employer allowed to measure AI use?
Yes, with conditions. The GDPR asks for a purpose, proportionality and transparency, and a works council has a say over facilities suited to monitoring. Measuring without a user id and without content makes that conversation a lot shorter. The article on what an employer may check goes into the detail.
Further reading
The reasoning, one question at a time.
See which AI tools are in use, and how much of the use runs through the allowed route.
Fourteen days free, every module, no credit card. No user id in the figures, from day one.
14 days free, no credit card. Rolled out across your organisation within the hour.
