Most Copilot adoption dashboards track the wrong things. The metrics that actually predict value are MAU (monthly active users on the Copilot SKU), weekly active and sticky daily-over-weekly ratio, action-per-session (the count of meaningful AI-completed tasks per logged session), prompt depth (single-shot vs multi-turn), and the SKU-burn ratio (licensed users where 30-day MAU sits below 50% — the rationalisation candidates). Vanity metrics — total prompts, total characters generated — correlate with neither retention nor value. The 2026 enterprise pattern: a five-metric dashboard tied to a 90-day rationalisation cycle that reclaims 12–25% of seats inside the first year and locks the savings in at renewal.
Why Copilot adoption metrics matter to procurement
Copilot adoption metrics for enterprise are a procurement question because every Copilot seat carries a 12-month minimum commitment at $30 per user per month. A 10,000-seat M365 Copilot deployment is $3.6M annualised. If 30% of those seats are unused or under-used at 90 days — the rate we see across more than 60 enterprise deployments — the procurement exposure is $1.08M per year of stranded spend. The measurement framework exists to surface the rationalisation candidates inside the cancellation or reassignment window, not to celebrate rollout vanity metrics in steering committees.
The structural mistake we see is the Microsoft Customer Success-led measurement framework. The Microsoft default dashboards emphasise reach ("60% of users have launched Copilot at least once") rather than depth ("how many users are 75%+ MAU after 90 days"). The Microsoft-favourable framing is true and useless; the procurement-favourable framing is the basis for rationalisation. The Microsoft Licensing Center of Excellence owns the measurement definition — not the LSP, not Microsoft FastTrack.
The five metrics that matter
The Copilot adoption metric set we recommend for any enterprise running M365 Copilot at scale:
- 30-day MAU on the Copilot SKU. The count of licensed users with at least one Copilot interaction in the trailing 30 days. Target: >75% at 90 days post-provision. Below 50% at day 90 is the rationalisation threshold.
- Sticky ratio (DAU/WAU/MAU). Daily active divided by weekly active divided by monthly. A healthy ratio for an information-worker tool is 0.20–0.35 (DAU/MAU). Below 0.10 is a habit problem; the user has the licence but does not use it.
- Action-per-session. The count of distinct AI-completed actions per logged session (drafting an email, generating a slide, answering a question over a document). Healthy: 3–6 per session. Below 1.5 indicates trial behaviour, not adopted behaviour.
- Prompt depth. Average prompts per Copilot conversation. Single-shot interactions (depth 1.0) often indicate users treating Copilot as autocomplete; depth 2.5–4.0 indicates true reasoning sessions. Depth predicts retention better than volume.
- SKU-burn ratio. The percentage of licensed seats with 30-day MAU below 50%. The rationalisation pool. Track over time and feed into the quarterly rationalisation review.
Where the data comes from
The five metrics are computable from native Microsoft surfaces:
- Copilot dashboard in the M365 Admin Center. The primary source. Per-user MAU, weekly active, action counts, and SKU assignment.
- Microsoft 365 Usage Analytics (Power BI). The secondary source for cross-product activity correlation — users with high Copilot usage tend to be high Office and Teams users; the cross-product join surfaces the cohort patterns.
- Microsoft Viva Insights (advanced). Optional layer for sentiment and time-saved attribution. Requires a Viva Insights licence on the measurement cohort.
- HR data (Workday, SuccessFactors). Joiners-movers-leavers and role / department mapping. Without HR join you cannot do cohort analysis or rationalisation by population segment.
The Copilot dashboard plus M365 Usage Analytics plus HR join covers the full measurement framework. There is no enterprise need for a custom telemetry platform — see our coverage of the license optimization toolset for the broader analytical context.
Thresholds — what "good" looks like
| Metric | Healthy | Watch | Rationalisation candidate |
|---|---|---|---|
| 30-day MAU (day 90) | >75% | 50–75% | <50% |
| Sticky ratio (DAU/MAU) | 0.20–0.35 | 0.10–0.20 | <0.10 |
| Action-per-session | 3–6 | 1.5–3 | <1.5 |
| Prompt depth | 2.5–4.0 | 1.5–2.5 | <1.5 |
| SKU-burn ratio (cohort) | <15% | 15–30% | >30% |
The 90-day rationalisation cycle
The cycle has four phases on a 90-day repeating cadence:
- Days 0–30: provision. Roll out Copilot to the target cohort. Set up the dashboard. Baseline the five metrics.
- Days 31–60: training and nudge. Cohorts below the watch threshold receive targeted enablement (skill cards, prompt libraries, manager nudges). Do not rationalise on month-one data — the habit-formation curve is longer than that.
- Days 61–90: triage. Cohorts still below the rationalisation threshold are flagged. Manager-level conversation: is this a habit problem (retain and re-engage) or a fit problem (reassign or reclaim the seat)?
- Day 90: rationalise. Reclaim the seats flagged for rationalisation. Reassign to the next cohort waiting for provision. Update the SKU-burn ratio.
Microsoft 365 Copilot seats can be reassigned within the M365 Admin Center; the 12-month commitment is per seat, not per user, so a reassigned seat does not break the EA commitment. This is the operational lever that makes the rationalisation cycle work without re-opening the EA. For the full commitment-side context see our analysis of CCCU economics and the broader Copilot Studio licensing pillar.
Vanity metrics to ignore
The metrics that show up in Microsoft FastTrack reports and steering committee decks but do not predict value:
- Total prompts. A user who runs 200 trivial prompts in a week is not more adopted than a user running 30 substantive prompts.
- Total characters generated. Length is not value. Long verbose generations often correlate with low-quality use cases.
- Reach (% of organisation provisioned). 100% reach with 30% MAU is a worse outcome than 70% reach with 80% MAU on the right population.
- Survey-only NPS. Useful as a triangulation signal; never as the primary metric. Behavioural metrics beat self-reported metrics in adoption analysis.
- Number of features used. Information workers cluster on 3–5 features that fit their work. "Feature breadth" is a Microsoft-favourable metric that does not predict retention.
Anonymised case study: 1,840 seats reclaimed at $662K annualised
An 8,300-employee retail enterprise deployed M365 Copilot across the full information-worker base in late 2025. By month four the Microsoft-led adoption review reported “72% of users have launched Copilot at least once” and recommended additional training spend. We rebuilt the measurement framework on the five-metric model and ran the 90-day rationalisation cycle. The findings: 1,840 licensed users at 30-day MAU below 25%, of whom 1,200 had not opened Copilot in 60 days. After targeted re-engagement 380 users moved back above the watch threshold; the remaining 1,460 seats were reclaimed at day 90 and reassigned to a waiting cohort of 1,200 newly identified high-fit users plus 260 seats cancelled at the anniversary. Net annualised saving: $662K (260 seats × $30 × 12 plus avoided expansion). The five-metric dashboard now runs continuously and drives the renewal-cycle Copilot sizing.
Copilot adoption metrics are the difference between $30 per user of value and $30 per user of stranded spend. Build the five-metric dashboard, run the 90-day rationalisation cycle, and treat the SKU-burn ratio as the controlling metric. Pair the discipline with the broader enterprise spend reduction playbook and the EA tier-collapse renewal context, and Copilot stops being a vanity rollout and starts behaving like the managed SKU it is.