Repository navigation
Copilot Adoption Interpretation
This page is for Copilot programme leads and anyone who presents Copilot Adoption figures. It defines the terms the page uses, explains how to read each chart, lists the measures that are easiest to over-read, says what each warning means and what to do about it, and answers common questions.
On this page: Glossary · Reading the charts · Measures that are easy to over-read · Warnings you might see · When a figure is blank · FAQ
| Term | Meaning |
|---|---|
| Licensed users | People holding at least one SKU classified as a Microsoft 365 Copilot licence. The true licence count |
| Analysed / scored users | The licensed users this run actually scored. The same as above unless a very large tenant hit the analysis row cap — every rate is of this number, and the page says so when they differ |
| Active | At least one Copilot interaction in the period. A deliberately low bar |
| Habitual | Engagement at or above the Established threshold (50 by default). A weighted behavioural indicator, not measured productivity or value |
| Dormant | Used Copilot before this period, but not once inside it. Needs a conversation |
| Never used | No Copilot activity at all in the history window. Needs onboarding, or the licence back |
| Idle seats | Dormant + Never used. The population the reclaim tiers are drawn from; not the same as reclaimable licences |
| Reclaimable licences | The seats the tool rates safe to take back: the Certain reclaim tier (a disabled account that still holds a Copilot licence, however recently it was used) plus the Probable tier (an enabled account with no Copilot use observed in the imported history, created longer ago than the reclaim grace period). Dormant seats and accounts too new to judge are review-only, and seats an administrator has excluded are held back. So are report-sourced seats when Microsoft's report period doesn't match yours. The figure is therefore usually smaller than the idle seats; the ⓘ on the card reconciles the two. See Reclaim tiers |
| Engagement score | 0–100 from frequency (50%), depth (30%) and breadth (20%) by default, each capped. An administrator can change the weights |
| Bands | By default Champion 75+, Established 50+, Developing 25+, Trialling below that. An administrator can change the three thresholds |
| How often people open Copilot | Infrequent 1–5, Moderate 6–10, Frequent 11–19, Daily 20+ active days a month. Unweighted — deliberately a different measure from "habitual", so the two can be compared |
| Breadth | How many different Copilot surfaces (Teams, Word, Outlook, Chat…) someone uses |
| Business case score | 0–100 for an unlicensed user. Already using Copilot Chat is worth the most, because it is evidence rather than inference |
| Time already saved by seat holders | Modelled hours a month that the Copilot actions of people who already hold a licence may be saving. A range, in hours only, and never added to the licence or Cowork estimates |
| Cowork | Microsoft 365 Copilot Cowork. Only shown if actually detected. Counted as interactions from the Copilot audit log, never as tasks: Microsoft's per-user Cowork usage report has no API, so the tool doesn't import it |
| Chart | Read it as | Common misreading |
|---|---|---|
| Gauges | The rate against a judgement scale | The bands are deliberately stepped, not a gradient — 41% and 43% are the same verdict |
| Funnel | Each stage is a subset of the one above. The % is conversion from the stage immediately above, not from the top | A healthy 90% following a terrible 40% is still a good step |
| Habit tiles | Of the people who use Copilot, how often | Percentages are of active users, not of all licences. Never-used licences are idle seats, assessed by the reclaim tiers instead |
| Engagement mix | The whole licensed population in six exclusive bands | — |
| Treemap | Area = share of interactions | Counts interactions, not people. One heavy user can dominate a surface |
| Frequency vs intensity | Position, with your own medians as the dividing lines | The lines are your medians, not industry targets. Bubble area is licences |
| Radar | The shape of engagement, typical user vs your champions | The gap between the outlines is the finding, not their size |
| Concentration | How much of all activity each cohort accounts for | Only active users are ranked |
| Stacked area | Total volume and its composition | Only the bottom band sits on a flat baseline, so only it reads precisely |
Some measures answer a narrower question than their name suggests. Read these before you quote them.
| Where to find it | What to do | Do not conclude |
|---|---|---|
| Executive / Analyst view — Microsoft's prompt line | Read Microsoft's prompt count, average, report date/version and period alongside—not inside—the audit totals | A prompt is an audit interaction, missing values are zero, or filtered reports narrow this tenant-wide line |
| Executive / Analyst view — managers' own use | Compare active/inactive managers' teams, including the split by managers' licence status; inspect Unknown and withheld groups | That managers' use caused the difference, or Unknown means inactive |
| Analyst view — department agent columns | Use customer-built-agent breadth/depth and independent builder counts to decide where to enable more workflows | That Microsoft/unknown-origin agents were counted, or a builder was identified as the owner of a particular agent |
| Agents — Type, reach and growth | Distinguish Origin from Kind; inspect unknown-origin exclusions, departments reached and the closed 28-day growth windows | That raw audit rows equal logical turns, maker testing is adoption, billing is autonomous-run counts, or Microsoft's 15× is your target |
| Analyst view — leadership comparison | Ask an administrator to configure a direct-member Entra group, then wait for the importer outcome | That membership is historical, or See PII can override small-cohort/complement suppression |
| Administration — Copilot Adoption settings | Change one coherent group of rules at a time, record the settings version, and refresh before comparing. The 33 governed settings include weights, bands, engagement targets, habit buckets, opportunity and Cowork decisions, agent thresholds, reclaim grace and the segment minimum | That history windows, row caps, the 28-day habit normalisation, the default time-saved credits or activation windows are editable: they are fixed. (A reader can change time-saved assumptions for their own session only.) |
The ⓘ help beside a figure gives its definition, formula, source and coverage. SQL is available in Analyst view where offered. Portal labels/help and the built-in guidance titles are translated in English and Spanish; imported names and administrator-authored taxonomy labels are not translated. Workbook methodology and guidance titles remain English.
The warnings appear at the top of the page and on the Excel report's Report sheet. They are grouped here by what you do about them. Checking Administration → Service health needs the Administration permission, so a programme lead without it should ask an administrator.
| Warning | What it means | What to do |
|---|---|---|
| "No licence information has been imported" | The user metadata import has not run | Enable it; the tool cannot work without it |
| "No Microsoft 365 Copilot licences were found in this tenant" | The analysis ran and found nobody holding a Copilot licence | On a new installation the user import usually hasn't finished yet. Otherwise, check that Copilot licences are assigned |
| "Neither the Copilot audit import nor Microsoft's Copilot usage report has any data for this period" | Every user will look unused | Ask an administrator to check Administration → Service health before acting on anything here |
| "Some Copilot interactions have not finished being upgraded to the new reporting format" | Some stored interactions are still being converted after an upgrade, so every Copilot figure is too low for now | Nothing to fix: it repairs itself over the next few import cycles. Run the report again once the importer has caught up. If the warning stays, check that the Office 365 activity importer web job is running |
| "The Copilot audit import has no data for this period, so per-user engagement is derived from Microsoft's own usage report" | Falling back. That report covers Microsoft's window, not yours, and excludes unlicensed use | Treat unlicensed figures as unavailable, not as zero |
| "This tenant has 'concealed user information' enabled" | Microsoft's per-user report returns hashed identities | Audit-derived figures are unaffected — this is why the audit import matters |
| "Purchased and unassigned Copilot seats are unknown" | The tenant's purchased-licence inventory hasn't been imported, usually because Organization.Read.All isn't granted. Assigned seats are still counted |
Grant Organization.Read.All and let the user import run again. See Copilot Adoption Setup
|
| "Purchased and assigned Copilot seats disagree for …" | Graph's purchased count for that licence doesn't reconcile with the seats assigned, so its unassigned seats show as Unknown rather than zero | Check that licence's purchase and assignment counts in the Microsoft 365 admin center. Nothing needs fixing in this tool |
| "The Microsoft 365 usage reports are not available, so licence candidates are ranked only on unlicensed Copilot Chat use" | The Microsoft 365 usage-report import is off or has no data | Heavy Microsoft 365 users who have never tried Copilot won't appear on Licence opportunities. Turn on the usage-report import; see Copilot Adoption Setup |
| "Licence opportunities need either the Copilot audit import or the Microsoft 365 usage reports" | Neither source has data, so no licence candidates can be identified | Enable at least one of the two imports |
| "The Microsoft 365 usage reports are not available, so coordination load cannot be measured" | Cowork readiness has no workload evidence, so nobody is identified as a Cowork candidate | Turn on the usage-report import |
| "The Copilot audit import has no data for this period. Cowork interactions come only from the Copilot audit log" | Nobody can be shown as already using Cowork; readiness uses workload and Copilot fluency only | Check the Copilot audit import |
| "Cowork readiness needs the Copilot audit import or the Microsoft 365 usage reports" | Neither source has data, so there is no readiness assessment | Enable at least one of the two imports |
| "Cowork readiness was measured, but the licensed-user analysis it takes Copilot fluency from did not complete" | The licensed-user analysis failed, so the Cowork tab couldn't be scored. It is not a missing import | Ask an administrator to check Administration → Service health, then run the report again |
| "Could not load …" | One query failed; the rest of the report is fine | The named chart is missing. Ask an administrator to check Administration → Service health, then retry |
| Warning | What it means | What to do |
|---|---|---|
| "… licensed users (…%) were scored from Microsoft's Copilot usage report because the audit import had no per-user signal for them" | Those people are scored from Microsoft's report instead. Their prompt counts are kept out of interaction totals, concentration and intensity | Expected for a few people. A large share suggests the audit import is incomplete |
| "Microsoft's pinned Copilot usage-report period is D…, but this analysis window is D…" | Report-sourced seats that would otherwise be Probable reclaims are held back from the reclaimable figure, because the two periods differ. Certain (disabled-account) seats are never held back | Expected. The ⓘ on Reclaimable licences shows how many were held back for window mismatch. A period the same length as Microsoft's report avoids it |
| Warning | What it means | What to do |
|---|---|---|
| "only N users could be analysed" or "Only the first N licensed users were analysed" | A very large tenant hit the row cap (50,000 licensed users by default) | Rates describe those N users, not the whole tenant. Do not quote them as tenant-wide |
| "The analysis covers only the first … of this tenant's Copilot licence holders, and this view selects licence holders beyond that limit" | A filtered view includes people outside the analysed users | Treat the figures as a partial count of this view |
| "The agent inventory was capped at … agents" | The tenant used more agents than the inventory holds (5,000 by default) | Agent figures are a floor, not a total |
| "More than … combinations of agent and person were used in this period" | Over 1,000,000 agent–person pairs | Agent breadth, depth and reach are left out rather than calculated from part of the data |
| "Unlicensed Copilot usage was capped at … users" | More unlicensed users than the cap (50,000 by default) | Unlicensed figures are a floor, not a total |
| "Cowork readiness scores at most … seat holders" | The Cowork assessment reached its cap (50,000 by default). People already using Cowork are taken first, then those with the most coordination load | Counts on the Cowork tab describe the seat holders who made the list. If the warning says nobody in your filtered view made the list, widen the filter or read the tab for the whole tenant |
| "Cowork readiness: … seat holders were scored without a Copilot fluency figure" or "Cowork readiness can't be scored for this view" | Those seat holders fall outside the licensed-user analysis cap, so their fluency reads as 0 | Treat their readiness tier as unreliable, or read the tab for the whole tenant |
| Warning | What it means | What to do |
|---|---|---|
| "Cowork use is shown as a number of people, not as a percentage" | The percentage needs the number of people allowed to use Cowork. Your spending policies decide that, and the tool can't read them | Nothing. It appears on most tenants with Cowork use |
| "Licence assignment history is not available yet …" or "Licence assignment history starts on …" | A past date range is scored against today's seat holders, or partly reconstructed from the first licence refresh | Read its seat holders as approximate. See Choosing the reporting period |
| "This period does not end today …", "… are today's values …" or "The agent inventory remains an as-of-now view …" | Shown for every past date range | Expected. Choose a rolling period to see named lists |
Do not replace a dash with zero in a presentation or spreadsheet.
Since Stable build 1860 (PR #877, PR #884): for readers without See PII, idle seats are withheld for a Copilot licence that fewer than 5 people hold, and for the one other Copilot licence that would give the figure away. On the email-domain table, each population (seat holders, unlicensed Copilot Chat users and licence candidates) is withheld when it's one to four people, and agent builders by department are withheld for a department with fewer than 5 builders. When a department, country or accountability breakdown leaves out a group too small to show, it also leaves out its smallest listed group, in every view: in a tenant with any department under 5 seat holders, these readers usually see one department fewer. The workbook, including its Cowork sheet, follows the same rules. See Portal permissions → Small groups withheld from readers without See PII.
| Symptom | Check first | Reader-safe conclusion |
|---|---|---|
| No scored seat population | Completed directory/licence import, included seat SKUs, scope and history | No usable denominator yet, not “nobody uses AI” |
| Growth window blank | Retained history reaches its first day, some Copilot data inside, successful source read | That window is unmeasured; no growth ratio |
| Breadth/depth/builders blank | Agent–person cap, origin read, period coverage, separate authoring import | Evidence missing/withheld, not zero workflows or makers |
| Since Stable 1859 (PR #821): Active agents or Interactions per agent user shows a dash, Not measured for this period | A past range: the page's warnings say whether the audit log holds nothing for it or its read failed | The range's agent use is unmeasured, not zero. The agent list, health verdicts and agent use by department are still as of now |
| Manager or leadership comparison blank | Unknown/capped manager evidence; leadership pending/failed/stale read; small cohort/complement; narrowed scope | The comparison is not reportable; roles cannot override it |
| Category chart empty or gapped | History import and service plan, durable opt-in, compatible model, immutable version, ten-person floors and run counters | The sample is insufficient/withheld; not a complete zero-intent population |
| Microsoft prompt field blank | Source report version/date and a successful summary import | Microsoft did not supply that field; do not derive it from audit totals |
| Adoption report refuses after a settings read | Table endpoint/DNS/runtime permissions and ten-second deadline | Effective rules unavailable; cached/default substitution would be unsafe |
| Billing connection required or failed | The optional delegated administrator connection, consent, renewal and completed import phases | No usable consumption evidence, not zero billed use |
For an operational investigation, ask an administrator to start with Copilot Adoption Troubleshooting. Do not post production workbooks, tenant identifiers, raw prompts, payloads, logs or unreviewed screenshots in public issues.
Why is my adoption rate 100% but habit rate near zero? Everyone opened Copilot at least once and almost nobody kept using it. This is the most common and most expensive pattern, and it is exactly what the adoption rate on its own conceals.
Why do the numbers differ from the Microsoft 365 admin centre? Different sources and windows. This tool primarily uses the Copilot audit log, matching your chosen period exactly and including unlicensed users; Microsoft's dashboard uses its own aggregation window and licensed users only. The How this is calculated tab names the source for every figure.
Why is a Copilot-branded SKU not counted as a Copilot licence? Microsoft ships Copilot-branded products that are not a Microsoft 365 Copilot licence — Copilot Studio, Copilot for Sales. Counting them would inflate the licensed population and make adoption look worse than it is. Every SKU found is listed under How this is calculated, including the excluded ones.
Someone is on the "never used" list but says they use Copilot daily. Ask an administrator to check the audit import on Administration → Service health, and check whether they hold a licence at all — heavy Copilot Chat use without a licence appears under Unlicensed usage, not in the licensed list.
Can I change the thresholds? Yes: an administrator can change 33 governed settings, including engagement targets, habit buckets, opportunity/Cowork decisions, agent lifecycle, reclaim grace and segment privacy as well as weights/bands. See Copilot Adoption Settings for defaults, validation and seat-impact confirmation. Activation windows and row/performance caps are not editable. Screens/workbooks quote effective rules; two snapshots are comparable only when those rules match.
Why does the agent list not show an agent I know exists? Agents appear in the usage inventory only after observed use. An agent built but never run, or only tested by its maker in the Studio test pane, is not listed as adoption usage. The separate optional Agent 365 Package Catalog can describe packages that exist; it is not proof they ran or were billed. Check scope/history/caps before concluding an agent is absent.
Why did an agent's interaction count drop after upgrading? A turn with a Copilot Studio agent in Microsoft 365 Copilot used to count twice, because it is audited twice. It now counts once, for interactions imported after the upgrade (and the few days before it that the importer re-reads). See how a Copilot Studio agent's turns are counted.
Can I put a money value on the time saved? Not on this page. Every modelled figure here is in hours, because pricing a modelled saving needs an hourly rate the product can't evidence. The estimates are also never added together: time already saved by seat holders, the licence estimate and the Cowork estimate each answer a different question. For a modelled money view, use Copilot investment, which applies your organisation's own saved assumptions; it is never an invoice or a measured saving. For the credits people actually consumed, see Copilot spend.
The Copilot section
- Copilot analytics — what Copilot data is collected, and when
- Copilot Adoption specification — what the tool is for, and the design principles behind every figure; every formula is in the metrics reference
- Leadership comparison — the optional comparison of one admin-named group with the tenant; totals only, with cohort and complement suppression
- Agent costs — what your Copilot Studio agents cost, and what they were charged for
- Copilot investment and Copilot spend — modelled money from your saved assumptions, and the Copilot credits people consumed
- Work Trend Index — which of Microsoft's research questions these figures can and can't answer
- Optional features on other pages: DLP governance signals on Insights → DLP impact, the prompt-category mix and trend on Insights → Copilot → Prompt insights, delegated billing and the Agent 365 catalog under Administration. Prompt categories need interaction history and a durable opt-in, not the audit toggle; catalog metadata is not billed usage.
Elsewhere
- Monitoring and Health alerts — checking the imports are running
Previous: Copilot Adoption Tabs · Next: Copilot Adoption Reviews and Exports: how to act on the figures.
Programme leads' path: User guide → Tabs → Interpretation → Reviews and exports
- Copilot analytics overview
- Copilot Adoption user guide
- Copilot Adoption setup (admins)
- Work Trend Index
- Copilot investment
- Copilot spend
- Agent costs
- Prompt categories
- Copilot credit history (preview)
- Agent 365 package catalog
- The Copilot data in depth
- Prerequisites
- App registrations setup
- Install with the installer
- Verify the deployment
- Security & permissions model
- Private endpoints (optional)
- Certificate authentication (optional)
- Manual installation
- Activity & usage analytics (Power BI)
- SharePoint page tracking