Many organisations in the UAE now pay for one AI assistant or more, and some pay for three or four. Yet when a CEO asks, “Is our team actually using this, and is it worth it?”, the honest answer is often a shrug. An AI adoption dashboard built on token usage tracking turns that shrug into a clear, weekly picture of who uses AI, for what kind of work, at what cost — and with what result.
Why measure AI adoption?
Licences are easy to buy and hard to evaluate. Seat counts tell you who could use a tool; tokens tell you who does. A token is the unit AI models use to read and write text, so token consumption is the closest thing you have to a meter on the work being done.
Measuring adoption properly helps you to:
- Spot uneven uptake — some teams race ahead while others never log in. Both are signals worth acting on.
- Control spend — identify unused seats, runaway API scripts and duplicated tools before renewal.
- Find your champions — the people whose workflows others should copy.
- Target training — focus enablement, such as corporate coaching programmes, on the teams and use cases where it will make the biggest difference.
- Tell an honest story to leadership — adoption, cost and outcomes on a single page.
The core idea
Tokens measure activity. The goal is outcomes. A good dashboard shows both side by side and never lets the first stand in for the second.
Start from a pre-built kit for your tool
Before you build anything, check what your vendors already provide. Most enterprise AI plans ship with an admin console, usage exports or an analytics API. Starting from these saves weeks and keeps your numbers consistent with the invoice.
- ChatGPT (Team / Enterprise): workspace analytics and user-level activity exports from the admin settings.
- Claude (Team / Enterprise and the API): usage and cost views in the admin console, plus usage reporting for API keys.
- GitHub Copilot and coding agents: seat-level activity and usage metrics available to organisation owners.
- Direct API usage: billing exports from each provider, tagged by project or key.
Exact field names and export formats change as vendors update their products, so treat the vendor's own documentation as the source of truth. Your job is to take whatever each kit gives you and bring it into one shared model.
Distinguish assistant work from computer work
Not all tokens are equal. We find it useful to split usage into two broad categories, because they behave very differently and should be judged differently.
Assistant work
A person in the loop, message by message: chatting, drafting emails and job descriptions, summarising documents, researching a market, preparing for an interview. Token volumes per task are modest, and value shows up as time saved and better quality.
Computer work
AI acting with more autonomy: coding agents writing and testing code, automations processing hundreds of CVs, scripts calling models through the API, agents working across tools. Token volumes can be orders of magnitude higher, and value shows up as throughput — tasks completed without a person doing each step.
If you put both on the same leaderboard, a single engineer running an overnight agent will dwarf a whole HR team using chat thoughtfully all day. Tag every record as assistant or computer work at ingestion, and report them separately.
Build it in six steps
1. Export usage data
Pull usage and billing exports from each AI tool's admin console or API — ChatGPT Enterprise or Team, Claude for Work, GitHub Copilot and any API keys your engineers use. Schedule the exports weekly to start; daily is rarely necessary for decision-making.
2. Normalise per user, team and tool
Create one table with the same columns for every source: date, user_id, team, tool, work_type, input_tokens, output_tokens, requests and cost. Join the team from your HRIS rather than trusting tool metadata, and replace names or emails with a pseudonymous ID wherever the dashboard does not need to identify individuals.
3. Define the key metrics
Keep the list short. These five cover most leadership questions:
| Metric | What it tells you | How to calculate |
|---|---|---|
| Weekly active users | Breadth of adoption | Distinct users with at least one meaningful session in the week ÷ licensed seats |
| Tokens per task | Efficiency of prompts and workflows | Total tokens ÷ completed tasks (for the workflows you can count) |
| Cost per outcome | Value for money | AI cost attributed to a workflow ÷ outcomes delivered (e.g. PRs merged, roles shortlisted) |
| Adoption trend | Momentum | 4-week rolling average of active users and tokens, per team |
| Work-type mix | Maturity | Share of tokens from assistant work vs computer work |
4. Connect outcomes
Join usage to a small number of business outcomes — tickets resolved, pull requests merged, proposals sent, roles filled — so you can see value, not just activity. Start with one outcome per team. In recruitment that might be time-to-shortlist; in engineering, merged pull requests; in finance, month-end close tasks automated.
5. Visualise
Use whatever your company already has — Power BI, Looker Studio, Tableau, Metabase or even a well-built spreadsheet. The layout matters more than the tool: a headline row of four numbers, a team-by-tool breakdown, a trend line and an outcomes panel. If it does not fit on one screen, it will not be read.
6. Run a weekly review ritual
A dashboard nobody discusses is decoration. Book 20 minutes a week with team leads. Look at the trend, ask one person to demo a workflow that saved real time, agree one experiment for next week, and note which teams need support. Over a quarter, this ritual does more for adoption than any launch email.
A sample dashboard
Here is the shape we recommend. The figures are illustrative only, to show layout rather than benchmarks.
Active users
74%
▲ vs last week
Tokens / task
5.2k
median
Cost / outcome
AED 3.10
blended
Tools in use
4
across teams
Weekly active users by team (%)
Engineering
Talent Acquisition
Finance
Marketing
Operations
Outcomes this week
- Roles shortlisted9
- Pull requests merged37
- Proposals drafted14
Needs attention
- 11 seats unused for 30 days
- One API key above budget
- Operations trending up
Ranking methods that reward outcomes, not volume
Leaderboards are tempting and can motivate people — but ranking by raw tokens rewards waste. Someone pasting a 200-page PDF into chat five times will “beat” someone who solved the same problem with a sharp, well-structured prompt. Better approaches:
- Outcome per token: rank workflows, not people, by outcomes delivered per thousand tokens or per dirham spent.
- Consistency over spikes: credit teams for the number of weeks with healthy active usage, not a single record week.
- Reuse: recognise people whose prompts, templates or agents are adopted by colleagues.
- Improvement: celebrate the biggest improvement in cost per outcome, which levels the field between large and small teams.
- Peer-nominated wins: pair the numbers with a short weekly story of real impact.
Rule of thumb
If a metric can be improved by doing the same work less efficiently, do not rank people on it.
Pitfalls to avoid
Incentivising token volume
Once people know tokens are counted, some will generate them for their own sake. Keep token totals as a diagnostic for spend and adoption, never as a target.
Vanity metrics
“Two million tokens this month” sounds impressive and means nothing on its own. Every headline number should answer a decision: renew, expand, train, reallocate or stop.
Privacy and the PDPL
Usage logs are personal data when they can be linked to an employee. Under the UAE Personal Data Protection Law (Federal Decree-Law No. 45 of 2021) — and the DIFC Data Protection Law if you operate in the DIFC — you should be clear about why you collect them, limit access, and keep them no longer than necessary.
- Tell employees what is tracked and why, in plain language.
- Report at team level by default; restrict individual views to people with a genuine need.
- Never ingest prompt or response content into the dashboard — counts and metadata are enough.
- Set a retention period and apply it automatically.
Comparing apples with agents
Mixing assistant and computer work on one chart produces misleading comparisons. Keep them apart, as described above.
Privacy checklist
Purpose written down · team-level by default · no prompt content · role-based access · automatic deletion. If you can tick all five, you are off to a responsible start.
Where to go next
Start small: one tool, one team, one outcome, one weekly meeting. Once the ritual sticks, add the next tool. Within a quarter you will have a clear picture of where AI is creating value in your organisation — and where it needs more support.
Adoption data often shows that some roles have changed shape. When that happens, re-score them with our free job evaluation tool so grades and pay keep pace with the work. In recruitment, the same principle applies to hiring software: our AI-assisted applicant tracking system is designed so recruiters, not algorithms, make the final call.
NamasteAi is a DIFC-based recruitment and HR partner that helps UAE employers hire, upskill and organise teams for the AI era, from HR consultancy on roles and policies to AI-assisted recruitment. If you would like a second pair of eyes on your AI adoption programme, get in touch with our Dubai team.