Your organization may already have AI tools that IT never selected: a spreadsheet connected to a chatbot, a form that sends data to an AI service, or an agent that can take a multi-step action. The first move is not a ban. It is a small, repeatable inventory.
In its Q3 2026 SMB AI Pulse, Pax8 says 67% of AI-engaged businesses surveyed had built at least one AI solution in regular use, while 35% had discovered unapproved AI in the preceding six months. The denominator matters: these are not percentages of all small businesses, schools, or nonprofits. The online survey covered 1,150 technology-involved leaders at firms with 5–499 employees in the U.S., U.K., Germany, and Sweden. It is a useful signal, not a census of your environment.
A 30-minute discovery pass
- Ask where AI is already used. Have department leads list tools, automations, API connections, and personal accounts used for work. Make disclosure a way to get help, not a trap.
- Trace the data. For each use, record the input types, the destination/vendor, and whether the workflow includes confidential or regulated information. Do not paste sensitive examples into an inventory.
- Find the action boundary. Does the tool only draft, or can it send, buy, update records, grant access, or trigger another system? Mark every external or irreversible action.
- Check ownership. Name a business owner and a technical owner. Note who can pause the workflow, review errors, and change its instructions.
- Check accounts and cost. Is there an organization-managed account, contractual terms, retention policy, usage limit, and a way to see spending?
- Triage rather than merely count. Prioritize workflows that combine sensitive data, broad access, and autonomous actions. A low-risk writing helper should not consume the same review effort as an agent with finance access.
What to do with the list
Put each use into one of three buckets: approved with documented boundaries, needs a short review, or pause pending a safer design. Give staff an approved path to propose tools; a policy with no usable alternative simply pushes use out of sight. Revisit the inventory as integrations and agent permissions change.
From the field
I would start by sharing how I use AI in my own work, then ask, “Where, if anywhere, is AI making your day easier?” I do not want it to feel like an interrogation or a policy quiz. I keep the conversation open and free of jargon, and let people tell me which tools they use, what they use them for, and what information goes into them. That gives me a better picture of where AI saves time, where staff may need help, and whether the way they are using it is safe for the work they do.
The takeaway
Shadow AI is a visibility problem before it is a discipline problem. The useful metric is not “zero unauthorized tools” on day one. It is whether you can identify the workflows that touch important data or act on behalf of the organization, assign owners, and close the biggest gaps.