Contractors, part-time staff and people on leave belong in AI adoption measures according to the work and period you are measuring. Establish who was in scope, when approved access was available and whether the relevant task occurred. Keep missing access, absence and unknown coverage visible before interpreting a low activity count.
Count people as people, and keep the dates that explain their opportunity. A part-time employee is not a fraction of a unique user. A contractor missing from an employee report is not necessarily a non-user. Someone away for the reporting period has not demonstrated abandonment.
The wider people and change guide covers participation and support. This guide focuses on reconciling workforce changes with an adoption report, before applying the activation and retention measures.
Define the workforce behind the rate
Start with the question the rate should answer. “How many people in the booking-summary trial used the approved tool?” needs a different population from “How many licensed accounts were active?” Write down the roles, task, observation period and account coverage before comparing teams or months.
For example, Microsoft's Copilot usage report defines enabled users and active users for its selected timeframe, then calculates the active-user rate against enabled users. That product definition does not establish that every person doing the work appears in the report, or that recorded activity produced useful work.
Reconcile the report with a dated workforce record:
- Population. Which employees and contractors were intended to participate in this task, and when did they join or leave it?
- Access. Which approved identity, tool and source permissions were available, and from what date?
- Opportunity. Did the relevant work occur during their scheduled time, including any absence?
- Coverage. Can the available report observe that person's approved activity, or is coverage unknown?
Keep these conditions separate. “Not in the report” is an observation about data coverage until someone establishes why. The AI adoption metric dictionary explains how to label the resulting numerator and denominator.
Count part-time people once and examine the work
For a rate based on unique people, count each eligible part-time person once. Do not divide unique active users by full-time-equivalent staffing. FTE describes contracted capacity; a unique-user count describes people. Combining those units makes the percentage difficult to interpret.
Contracted hours still matter to the opportunity being observed. A coordinator working two days a week might handle several relevant bookings in a fortnight. Another may handle the relevant task only at month end. Check the task's actual timing rather than assuming that fewer working days mean either no opportunity or lower commitment.
A New York Fed analysis published in April 2026 found reported workplace AI use differed between full-time and part-time workers, 42.7% and 24.7% respectively. It used supplemental November 2025 survey questions, including use over the preceding twelve months among employed respondents. Those self-reports do not show that contracted hours caused the difference, and they are not targets for your team.
For comparisons, agree an observation period that can include the task's normal cycle and report differences in exposure. If that period has not elapsed, say so. Do not demand extra AI use merely to make a part-time colleague resemble someone with a different workload.
Establish which contractor accounts you can observe
Include contractors when they are part of the task population you intend to assess. Then establish the approved working arrangement with the responsible manager and access owner. Employment label alone does not tell you which organization supplies an account, what information it can reach or whether your report covers it.
In a July 1 public X post, Θωμᾶς del Vasto described being a contractor at a US headquarters without Copilot access while other contractors and interns reportedly had it. A reply asked whether the company or an agency supplied those accounts; the author was unsure. The account is unverified, but that unresolved question is useful: identify the provisioning arrangement before comparing usage.
Ask the access owner to confirm the account, permitted task and reporting boundary. If a supplier-managed identity is outside the available telemetry, record that limitation. Do not enter zero use, assume an entitlement or encourage a personal account as a workaround.
Keep in-scope contractors with unresolved access visible as an access gap. Keep people genuinely outside the approved task scope separately identified, with a reason. Otherwise, excluding blocked workers can make an adoption rate improve while participation gets worse. Use the unequal-access guide to resolve the access decision.
Preserve leave and return dates in the observation
Leave changes what you could reasonably observe. Record the relevant absence interval and its effect on task opportunity. Someone absent for the whole period differs from someone who returned near its end, and both differ from someone who worked throughout without encountering the task.
Preserve the original cohort when reporting retention. Show absence, changed access and incomplete observation alongside it. If you also calculate a rate among people with an observed opportunity, label that population and its exclusions explicitly; do not silently replace the original rate. The retention guide's cohort method covers that calculation.
A return date does not establish immediate access or a complete follow-up period. Confirm access again and wait for the agreed task cycle before interpreting repeat use. Avoid repeatedly restarting the clock whenever someone does not use the tool; that would hide the very pattern you meant to examine.
Keep the reporting record limited to what explains eligibility and opportunity. It does not need a diagnosis or private reason for leave. Microsoft's privacy guidance on limiting sensitive organizational data supports minimizing data and avoiding sensitive health information. Apply your organization's access and aggregation arrangements, especially where a small group could identify an individual.
Reconcile one reporting period together
Suppose Sam Taylor, a workforce analyst, and Claire Bennett, a community-venue bookings manager, review an October 1 to 14 trial. Coordinators use approved booking requests and current venue rules to draft booking summaries. A coordinator still checks the room, date and stated access needs against the source before confirming a booking.
Sam brings the activity extract and dated trial roster. Claire brings the work schedule and booking assignments. They resolve three different records before describing participation.
| Record in the period | Reporting treatment | Next action |
|---|---|---|
Jamie Lee works Tuesdays and Thursdays and handled two relevant bookings with approved access. | Count Jamie once in the people population. Check activity against those actual opportunities. | Review the attempts and summary checks, without expecting daily use. |
Alex Morgan joined as a contractor on October 13 and is awaiting source-folder access. | Keep the late start and access gap visible. Do not label the missing activity abandonment. | Assign the access decision, then observe a complete agreed task cycle. |
A coordinator was on leave throughout October 1 to 14. | Retain the roster record and absence interval. Report no opportunity in this period. | Recheck access and task opportunity after return. |
These records cannot establish an adoption percentage by themselves. Sam still needs the complete population and the defined activity evidence. What they do establish is why three superficially similar activity records require different interpretations.
Keep the dated version rather than replacing it with today's roster. Microsoft's organizational-data preparation guidance explains how effective dates connect organizational attributes with activity. The same discipline matters in your reconciliation: a later join date or restored permission should not rewrite what was available earlier.
Questions about workforce adoption measures
Should part-time employees count less in the adoption denominator?
Not when the numerator counts unique people. Count each eligible part-time person once, then explain the task opportunities available during their scheduled work. Use a separately defined capacity measure if you need an FTE view; do not combine unique users with an FTE denominator. Check the task cycle before comparing activity.
What if contractors do not appear in the tool report?
First establish whether those contractors are in the approved task population and which organization supplies their accounts. Confirm whether the report can observe those identities. Missing coverage is not zero use. Record unknown coverage or missing access separately, assign its owner and verify the working arrangement before including it in a comparison.
Should people on leave disappear from a retention cohort?
Keep the original retention cohort visible and explain the absence interval and opportunity affected. A separately labelled opportunity-based view may exclude periods without the task, but it must show its population and exclusions. Do not silently delete absent people or classify them as having abandoned AI use. Use the original-cohort method consistently.
How should we treat someone returning near the period's end?
Record the return date, check approved access and identify when the relevant task next occurs. Describe the observation as incomplete until the agreed follow-up period or task cycle has elapsed. Keep the person visible in the appropriate roster, and avoid interpreting a few days without activity as evidence about their willingness or ability to use AI.
