AI usage across your organisation
Who is using AI, what they use, and how usage is changing. A synthetic rollout across M365 Copilot, Cowork, agents and GitHub Copilot. Usage intensity is recorded activity per person per week, including everyone in the roster, not just AI users.
Is AI usage growing?
Weekly people with recorded activity. Each person appears once in the overall line, even when they use several services.
These are observed users, not a complete census of every AI tool. Missing source records can make counts lower bounds. Service lines overlap and must not be added.
Weekly counts
How are people using AI?
Distinct users show reach; recorded activity per person per week shows intensity. All rates use the full organisation roster, including non-users. “At least” marks incomplete observations. Product units differ and must not be added.
Which agents are being used?
How are developers using GitHub Copilot?
Feature uses are a separate breakdown. Do not add them to chat requests or accepted completions. The fixture’s allocations are illustrative.
How much AI is being consumed?
Credits describe metered consumption. They are not currency, productivity or business value.
M365 credits per person per week
Each week’s recorded credits divided by all roster people that week.
GitHub credits per person per week
A different credit unit, shown separately. No cross-product total is calculated.
Which metered M365 services drive consumption?
Current limits do not establish a budget period or forecast. This dashboard does not calculate limit exhaustion, monetary cost or future spend.
Audit details: recorded totals and weekly credit figures
Where does AI usage differ?
Compare observed reach alongside group size. Higher recorded usage is not automatically better performance.
Groups use each person’s latest attributes in the selected window. Reach is recorded users divided by group headcount. Actions per person per week use all group members, including non-users and people with activity gaps. “At least” marks a lower bound where records are incomplete. A dash means unavailable or withheld.
Collaboration activity
Do people with higher recorded credit consumption also have higher collaboration activity? Compare groups in the same period, then explore their weekly context.
Activity by credit level
Each activity rate uses all people in that credit group and all weeks in the selected period, including zeros. No recorded credits does not mean no AI use.
Credit groups and exact values
Credits over time
Recorded credits per person per week for the selected group.
Collaboration activity over time
Selected activity per person per week for the same people and weeks.
The charts use separate scales. A point may be a recorded lower bound if observations are missing; see the weekly values. A gap means unavailable or withheld.
Weekly values
Definitions & coverage
Scope: the Person Query roster and the declared 13-week synthetic product window, 5 April to 4 July 2026. Headcount is the roster, not the sum of service user counts. Non-roster agent records are excluded from the dashboard.
Missing coverage: no confirmed unlicensed Copilot Chat activity, other AI vendors, satisfaction survey, task-completion outcomes or output-success measures are supplied. “AI users” means users observed in the supplied sources, not all AI use in the organisation.
Synthetic scenario: reproducible staggered team rollouts, changing weekly participation and demand, and a temporary dip with recovery. These changes are applied to source records, not chart lines. Access and HR attributes are unchanged. Existing licensed-activity gaps remain unresolved. The agent scenario assumes no use before its first week; committed CSVs and other demos are unchanged.
Metric definitions and denominator rules
- Observed AI users
- A distinct person with positive M365 Copilot actions or active days, Cowork tasks or active days, agent sessions, metered M365 sessions or credits, or GitHub activity or credits during the window. A person using multiple services is counted once. A positive observation remains positive even if another source is unresolved.
- Recorded activity per person per week
- Recorded actions, tasks, sessions, feature uses or credits divided by all roster person-weeks in the selected period: headcount multiplied by weeks. Non-users, unlicensed people and person-weeks with activity gaps stay in the denominator. Only received activity contributes to the numerator; rates are recorded lower bounds where coverage is incomplete, not estimates of missing usage. Group rates use that group’s full roster; agent, feature and service breakdowns use the full organisation roster, not only that category’s users. Raw totals remain in audit details.
- M365 Copilot access and adoption
- Access uses positive Total Copilot enabled days in the latest week, not activity. Licensed zeros remain in the denominator. The latest-week adoption rate is published only when activity is known for every licensed person and the licence and activity counts meet disclosure thresholds. Native licensed-only M365 nulls mean unlicensed under Microsoft’s documented rules, but these historical fixtures deliberately contain processing gaps in licensed activity. Those gaps remain unresolved. Their unlicensed zeros are fixture conventions, not the native-export null convention.
- GitHub use and acceptance
- Recorded use includes positive suggested or accepted completions, user-initiated chats, the Agent adoption indicator, or GitHub AI credits. A missing activity row is not a licence record or proof of non-use. Acceptance is total accepted completions divided by total suggested completions over the received records, not an average of person-level percentages. Zero suggestions produce no rate.
- Returning agent users
- The share of distinct active agent users in the latest week whose exported seven-day returning indicator is one. Repeated person-level indicators are deduplicated across agents. They are never added across weeks. This is not retention across all AI services.
- Credit volume
- M365 Copilot credits and GitHub AI credits retain separate units. Credits per roster person-week divide recorded credits by all roster person-weeks in the selected window, including people with no activity or no licence. Daily records are aggregated to Sunday-start weeks before joining to the roster. Feature and agent facts remain separate so they cannot multiply the person-week panel. No currency conversion, licence fees, policy-limit totals or savings are inferred.
- Collaboration activity and credit levels
- Emails sent, Teams chats sent, meetings attended and collaboration hours come from the Person Query. Collaboration hours include overlapping meeting and other collaboration activity only once; do not add hours to the other measures. Metric rules can vary by organisation. Credit groups use each person’s average recorded weekly credits in the selected period. People with zero recorded credits form one group; positive values are divided using organisation-wide tercile cutoffs, with ties kept together. A band can be empty. Filters retain those cutoffs and use each person’s latest group in the selected period. M365 and GitHub credits are never combined. Missing credit observations make that product’s group comparison unavailable. Missing activity remains in the denominator: “At least” marks a recorded lower bound, while all-missing values are unavailable. These are associations within the same period, not causal or predictive results.
- Disclosure and interpretation
- Some aggregates are withheld under a minimum group size of 10 people and complementary suppression rules. These rules do not prevent inference by comparing overlapping time windows, credit bands, teams or other linked aggregates. This synthetic example is not an approved real-export adapter or a privacy guarantee. Real use requires mapping, coverage, organisation-specific disclosure review and suitable controls. Tool activity does not establish productivity, satisfaction or causal impact.