A supervisor discovers that staff use a personal AI account to shorten customer notes. The immediate problem is understanding the task, the data, and the account. A warning about “shadow AI” alone will not tell the supervisor what to support or change.
Invite employees to describe the work they are trying to finish. Explain how the information will guide support and decisions, and follow through on that explanation. Staff have little reason to disclose a workaround if the only visible response is punishment.
Ask for the task before the tool name
Use a short inventory that an employee can complete without exporting private content. Ask what they do, how often they use the tool, which account owns access, and what categories of information they provide. Ask whether any output reaches a customer or changes a business record.
The following entries describe a hypothetical company. They show how the same inventory can lead to different actions.
| Task and tool use | Account and data | Disposition |
|---|---|---|
| Draft internal meeting agendas | Company account; general project topics without private client details | Support the use within the approved scope and give the team shared instructions. |
| Summarize customer notes | Personal account; identifiable customer information | Pause this data use while the designated owner reviews the arrangement and a suitable alternative. |
| Draft a technical estimate | Company account; approved requirements; engineer reviews the result | Investigate accuracy and review effort before widening use. |
| Rewrite public recruiting copy | Several personal accounts; no private applicant data | Consider a supported shared process if the task warrants it; avoid buying seats merely because use exists. |
Do not collect raw customer prompts just to prove that use occurred. Gather the minimum information the decision needs. If a particular concern requires more detail, give the relevant owner a controlled way to review it.
Give staff a disposition they can understand
Use four plain outcomes: support the current use, replace the arrangement, restrict a specific use, or investigate an unresolved question. State the reason and the next owner. “Under review” without a date or an alternative leaves people to invent another workaround.
For the customer-note example, explain which data use must pause and how staff should complete the task during review. Preserve a safe existing process. A tool restriction that makes the job impossible creates a conflict between the supervisor's instruction and the team's workload.
NIST's Generative AI Profile offers voluntary risk-management guidance for AI use. It provides a reference for considering risk; it does not certify that a chosen tool is safe or establish the value of buying it.
Look for the reason the workaround appeared
Ask what the employee could not do with the approved process. They may need faster access to information, help with a draft, or a simpler way to format a report. The response may involve instructions or a process change rather than a company-wide software purchase.
Check whether a supported tool already covers the task. Compare its practical fit with the workaround, including permissions and review burden. Do not assume that moving use into a company account resolves every question about the data or output.
Return the inventory results to the team at an appropriate level: which uses you will support, which need changes, and what you still need to learn. Let staff correct misunderstandings about the work. Keep account-access details and sensitive findings with the people who need them.
Explain the disclosure boundary
Tell employees who will see their responses and what information they should omit. Do not promise anonymity if the inventory asks for account ownership or a task that identifies one person. In a small company, the manager may need to discuss a use directly with its owner. Explain that process before collecting responses and give staff a route to ask questions. The aim is to learn enough to support the work and address specific concerns, while avoiding a broad collection of private prompts or customer records.
Our AI guardrails guide covers wider use rules. This inventory establishes what people do before you decide how those rules apply. For leadership that connects actual use with business decisions, explore fractional CAIO support.

