Read the Copilot usage report and the Agent usage report
You've picked the Microsoft 365 admin center's usage report to check basic adoption. That choice was correct—but it's also exactly where the Introduction's mismatch happens. Two people at Relecloud pulled two different, equally correct numbers for the same month, for the same company. Here's the precise answer: those numbers almost certainly came from two different reports measuring two different populations—the general Copilot usage report and the separate Agent usage report. Both are right. They're just not answering the same question.
Two reports, one shared blind spot
The Copilot usage report covers how people across Relecloud adopt, retain, and engage with Microsoft 365 Copilot and its enabled apps, and it includes a summary agent metric as part of that broader picture. The Agent usage report, by contrast, is a separate, dedicated surface built only to track agent adoption. If one colleague reads the agent number embedded in the general usage report and another reads the number in the dedicated Agent usage report, they land on different figures—not because either person made a mistake, but because the two reports scope "agent usage" differently, refresh on different schedules, and cover different sets of agents. That's the resolution to the puzzle from the Introduction: same month, same company, two legitimately different measurements.
With that resolved, the rest of this unit walks through what each report actually tells you, starting with the metric leadership cares about most.
Enabled users vs. active users: the gap that matters
The Copilot usage report typically becomes available within 72 hours of the end of the day it covers (UTC), and its hero metrics separate licensing from behavior. Enabled Users counts every unique person in Relecloud with a Microsoft 365 Copilot license over your selected timeframe—7, 30, 90, or 180 days. That's a licensing figure, not a usage figure. Active Users counts only the enabled users who tried a user-initiated Copilot feature in at least one app during that same window. Divide Active Users by Enabled Users and you get the active users rate—the single number that tells you what fraction of paid seats are actually in use.
Enabled and Active aren't the same population, and assuming they are is the first trap in reading this report. A license grants access; it doesn't guarantee anyone opened Copilot at all. The gap between the two counts is exactly what a rollout's "operate" phase exists to close, and it's a far more useful budget-review number than the Enabled count alone, because Enabled only tells leadership what was purchased.
The report adds two more layers of depth beyond that headline gap. Total prompts submitted counts every prompt sent to Microsoft 365 Copilot Chat during the timeframe, plus an average-prompts-per-user figure—a rough proxy for how deeply, not just how often, people engage. Active agent users counts unique Copilot users who used agents built by Relecloud, including admin-approved agents and agents built with Agent Builder and shared inside the org. That agent figure is useful, but it's scoped narrowly: it covers only organization-built agents, agent data only exists from November 1, 2024 forward, and usage of agents built by Microsoft or Microsoft partners isn't included yet. That scope limitation is one half of the puzzle's answer—it's a headline count inside a broader report, not the full agent picture.
Reflect: Suppose Relecloud shows 4,000 Enabled Users this month but only 900 Active Users. What does that 3,100-user gap suggest about where the rollout's "operate" phase should focus next—and would the Enabled count alone have told you that?
Where adoption lands: the workload breakdown
Knowing that people use Copilot is only half the picture; knowing where they use it is the other half. The Copilot usage report breaks activity down by host app—Word, Excel, PowerPoint, Teams, Outlook, and others—and using any Copilot feature in an app counts toward that user's Active Users status and updates their per-app Last activity date. This workload-level view turns a flat "adoption is 60%" into something you can act on: adoption strong in Teams, flat in Excel, worth a targeted nudge.
One distinction here is worth reading carefully before you trust the per-user prompt columns. In Word, the "Edit with Word" feature counts toward "Prompts submitted in Copilot Chat (work)." The equivalent features in Excel and PowerPoint—"Edit with Excel" and "Edit with PowerPoint"—don't count toward that same prompts metric. Two features that look parallel across three apps roll up differently in the numbers. If you're comparing prompt counts across Word, Excel, and PowerPoint side by side, that asymmetry can make Word's Copilot Chat prompts figure look inflated relative to the other two, purely because of how the metric counts, not because Word users engage more.
The per-user table itself lists User name, Display name, Prompts submitted (any app)—a total across all in-scope host apps—along with separate columns for Copilot Chat (work) prompts and Copilot Chat (web) prompts. Reading a single user's row means checking which column you're actually looking at before drawing a conclusion.
One more caveat before moving on: the system runs daily validation checks covering the past three days and fills in any gaps it finds. That means a number you pulled yesterday can nudge slightly as those checks catch up—don't read a small overnight shift in a recent day's figure as a data problem.
The separate Agent usage report: broader, faster, still in preview
Now the second half of the puzzle's resolution. The Agent usage report isn't the "Active agent users" metric you just read inside the general usage report—it's an entirely separate report, currently in preview. It tracks Declarative, SharePoint, and Custom engine agents built by Relecloud, by Microsoft, and by Microsoft partners, which is broader agent-source coverage than the general report's org-built-only scope. It also covers both licensed and unlicensed Microsoft 365 Copilot users, and its data appears within about an hour of interaction—a fraction of the general report's 72-hour window. Its key metrics include total active users and agents, a summary alongside a daily time series, and active usage broken down per user, per agent, and per agent-user pair.
Put the two reports side by side and the mismatch from the Introduction stops being a mystery: the general usage report's agent metric is a fast-glance headline nested in a broader 72-hour report, covering only org-built agents. The dedicated Agent usage report is a deeper, faster, preview surface covering a wider set of agent sources. Someone reading one and someone reading the other, for the same agent population overlap, can legitimately land on different counts—not an error, a difference in what's being measured and how fast it refreshes.
| Dimension | Copilot usage report (its agent metric) | Agent usage report (preview) |
|---|---|---|
| What it is | The "Active agent users" headline inside the broader usage report | A separate, dedicated report |
| Agent scope | Org-built agents only | Declarative, SharePoint, and Custom engine agents built by the org, Microsoft, and partners |
| Users covered | Licensed users | Licensed and unlicensed |
| Latency | ~72 hours | ~1 hour |
| Data available since | Nov 1, 2024 | — |
Adoption is measured now. But leadership's other question was cost—where do you even look for that, and is it the same place you just looked for usage?