University survey separates staff AI experimentation from weekly use

The customer problem is a rollout dashboard that treats initial experimentation as sustained service use. Ask HR, institutional research and the administrative sponsor how they define an active user, which staff are missing from feedback, and whether the targeted task occurs often enough to justify a license. A bounded engagement could redesign one pilot's measurement plan. The value hypothesis is a more credible renewal decision, not increased use for its own sake. The small administrative sample supports careful questions about denominators; it cannot forecast another campus's adoption rate, quantify addressable demand or justify savings and staffing claims.

Instrument the selected administrative workflow with separate measures for access, attempted tasks, repeated use, completed tasks and reviewed correctness. Prerequisites include an agreed eligible population, privacy approval and a baseline observation period. Keep telemetry minimized and separate operational evaluation from individual employee surveillance. Proposed validation should reconcile dashboard counts to a manually checked sample and show how missing responses alter conclusions. Integrate only the logs needed for the decision, regardless of hosting model. The preprint offers survey evidence rather than behavioral traces; do not convert its percentages into expected production load or assume they validate an agent architecture.

Assign institutional research to measurement design and the administrative service owner to service outcomes, with HR reviewing staff participation and communications. Document the invitation population, response coverage and denominator for every reported rate. Train pilot participants and include a route for nonusers to report barriers without penalty. Review privacy and accessibility before collecting data and before publishing small-group results. Proposed acceptance requires a reconciled participant count, documented missingness and an agreed quality measure alongside frequency. Compare over a defined follow-up period; proposed monitoring is not an observed result. Risks include selection bias, surveillance concerns and pressuring staff to use unsuitable tools.