Optimise Copilot licences by checking current assignment, measured use and whether the employee had suitable work during the review period. Investigate low activity before reclaiming a seat. For active users, inspect the resulting work before treating usage as a reason to renew.

This matters when a consultancy or professional-services firm starts reviewing its AI spend. Removing idle access is a defensible objective. A league table of prompt counts, however, rewards activity without establishing which licences help deliver client work.

Before the renewal meeting, ask each workflow owner to identify the task behind the allocation. A seat without a named use case has a weaker claim on the next budget than one waiting for a known piece of work.

An active user has tried a feature

In the Microsoft Copilot usage report, an active user is an enabled user who tried a user-initiated Copilot feature in one or more Microsoft 365 apps during the selected period. Submitting a prompt counts; simply opening the Word Copilot chat pane does not.

That definition supports an adoption measure. It contains no test of whether the generated work was accepted or whether the employee completed the task faster after corrections.

The same distinction applies in reverse. An inactive user has no qualifying activity in that window. The report alone cannot tell you whether they lacked access, had no suitable task, were away, or tried the tool previously and abandoned it after poor results.

Microsoft’s measurement and reporting guidance separates operational reporting from analysis of productivity and business outcomes. Keep that separation in the renewal discussion: use telemetry to identify where to investigate, then inspect the work that justifies the allocation.

Reconcile the reporting population before changing licences

Start with current assignments rather than assuming every row represents a seat you can reclaim. Microsoft’s user-level table includes people licensed at any point in the past 180 days, including users whose licences have since been removed.

Record the selected reporting window and its cutoff. The report offers 7-, 28-, 90- and 180-day views; Microsoft says a day’s activity typically becomes available within 48 hours of that day’s end in UTC. A review using incomplete recent days risks classifying use that has not yet arrived as inactivity.

A blank last-activity field also deserves care. Microsoft documents rare missing dates when someone used Copilot shortly after assignment and never used it again, as well as delayed client events affecting several apps. Treat those limitations as checks to resolve, rather than a blanket reason to disregard the report.

Have an authorised administrator reconcile the inventory and reporting settings. User-specific details are hidden by default in usage reports. Keep the review within the organisation’s approved access and employee-data practices; there is no need to expose individual prompt content to establish whether a suitable task occurred.

A quiet month can hide the next valuable task

Consider a hypothetical proposal lead at a consultancy. Their intended use is preparing a first draft from approved bid material. No suitable proposal arrived during the review window.

An adoption reminder will not produce the missing bid. Removing the licence solely because of that quiet period also leaves the original use case untested. The sponsor should decide whether the next expected opportunity justifies retaining the allocation, taking the actual subscription terms into account.

Now consider a colleague who had suitable proposals to prepare but avoided Copilot because assembling the inputs took longer than drafting manually. The next action is different: inspect one real proposal and resolve the input bottleneck before paying for another general training session.

An employee with repeated use presents another question. If the drafts require extensive correction, a high activity count can coexist with a poor business case. Compare the whole task, including preparation and review, with the existing route. Faster generation alone does not settle the purchase.

These are proposed review scenarios, not findings from a Model Operator deployment. Their purpose is to make the reason for each allocation visible.

Use one decision sheet for each proposed change

Keep the worksheet short enough for the workflow owner to complete. Use task counts where they are available, and label missing evidence instead of converting it into a zero.

FieldEvidence to recordHow it changes the decision
Current assignmentAdministrator-confirmed licence statusExcludes historical holders from the reclaim queue
Review periodWindow, UTC cutoff and report freshnessPrevents a partial period being treated as complete
Suitable workNamed task, opportunities during the window and expected recurrenceSeparates an untested use case from avoided use
Observed activityRelevant app usage for the same periodShows whether the intended route was attempted
Delivered workAccepted output and preparation, correction or review burdenTests whether activity helped the task
ExplanationAccess problem, absence, poor fit or other confirmed reasonDirects the intervention to the actual obstacle
Decision and ownerRetain, enable, reassign or review later; named approver and next eligible taskMakes the spending decision accountable

Do not turn this into an individual performance score. The unit of judgement is the licence allocation for a named use case. A manager owns whether that work occurs and what acceptable output looks like, which is why AI change management has to define the manager’s role.

For infrequent work, record the cost of keeping capability available through quiet periods. For recurring work, examine repeat use and task outcomes. Both decisions deserve evidence, but they need different observation windows.

Choose an intervention that can alter the result

Microsoft provides organisational messages aimed at inactive Copilot users. A reminder is useful when someone has a suitable task and has forgotten or missed the available capability. It does not resolve a missing source document or a workflow that is faster without the tool.

Where access is the obstacle, resolve it and review the next eligible task. If the use case has failed on output quality, inspect a representative example before expanding the rollout. Reassign a seat when its present holder has no defensible use case and another approved user does.

Keep reassignment separate from cash savings. Moving a licence to another employee changes who can use the purchased capacity. Reducing the bill depends on the subscription agreement, renewal timing and purchasing terms. Check those before presenting reclaimed seats as realised savings.

The same discipline should apply to buying implementation support. First establish what the existing platform can do for the chosen task. The consultancy-versus-in-house decision follows the remaining delivery gap; a low adoption chart does not establish a need for a custom agent.

Model Operator is a Dubai-based AI product studio, personally led by Alexander Chrisostomou. Its AI Initiative Consulting covers workflow analysis and adoption planning. Bring one licence-review sheet and the task it describes to a Model Operator build conversation. Use that evidence to decide what to fix before committing to more seats or implementation work.