An AI policy starts with a business question: do you know which tools your people use and what data they enter? You may know which apps you have purchased. That does not set whether staff use other tools, personal accounts or browser-based services for everyday tasks. The useful next step is evidence about your setup, followed by decisions about acceptable use.
Shadow AI means AI activity outside the business's approved arrangements. It does not automatically set malicious intent or a data incident. A staff member may be trying to finish a legitimate task without a clear approved route. Treat that possibility as a reason to improve visibility and guidance, while taking the data limit seriously. A useful policy should tell people what to do when their task does not fit the approved process.
Separate the Tool from the Work
Ask staff what they are trying to accomplish before starting a discussion about app names. A proposed internal outline, a summary of a client document and a draft public announcement involve different data and different review needs. The business purpose matters because it explains why the staff member chose the tool and what an approved alternative would have to support.
Then find the app, account arrangement and data involved. ChatGPT, Microsoft Copilot and Gemini are tool names, not complete descriptions of an approved workflow. Approval should connect the product and account with the intended task. Record which data may be used and who checks the result. A decision about one workflow should not quietly become permission for every possible use of the same app.
For example, you could approve an outline based on public data while requiring a separate decision for private company records. That is an illustrative policy choice, not a report about a DataSolved client. The value of the example is its clarity: staff can distinguish the task and data rather than guessing what a general instruction means.
Find the Data Limit
Make a list of the company data that needs explicit treatment. Include client records, financial details, credentials, private matters and business material subject to specific limits. Describe the limit in the language staff use. If a team handles a private proposal, use that example in training rather than relying only on an abstract category called private data.
The limit should cover uploads as well as typed prompts. A staff member might remove a name from a prompt but attach a document containing finding data. Have staff consider the whole interaction: the text, attachment, account and intended result. Review should happen before the data is submitted, while a clear reporting process should explain what to do if someone realizes a mistake afterward.
Avoid treating a single label as the entire decision. A private document may have different permitted uses depending on the client agreement, business purpose and approved service arrangement. Let the people who understand the data and its restrictions decide how it may be used. IT can support the resulting policy and controls without assuming authority over every business limit.
Use Discovery to Answer Defined Questions
Discovery should begin with questions leadership needs answered. Which AI tools are in use? Which are approved? Who uses them and how often? What categories of company data enter prompts or uploads? Define the users and devices included before monitoring begins. Explain the purpose to staff and decide who may review the resulting data.
A discovery report should distinguish observation from interpretation. An observed tool name is one finding. A advice about whether to approve it is a business and IT assessment. A risk ranking should help rank discussion, while the next-step plan should find the decisions and owners. Do not turn a report into a claim that every possible AI activity has been found.
DataSolved's paid 30-day AI Usage Discovery includes monitoring, a report for leadership and a review meeting. It is a fixed-fee starting engagement. Pricing is confirmed after a short call. The purpose is to give leadership evidence before choosing the next phase of policy, controls or approved adoption.
Govern the Use You Intend to Keep
An approved-tool list should state the permitted accounts and tasks, along with data limits. Explain who may approve a new use, what staff should submit for that decision and where the current policy is available. Keep the process workable. Staff who cannot find an approved route need a way to ask for one rather than a reason to continue outside the process.
IT controls can support those rules through warnings or blocks on private data in supported AI prompts and uploads, limits on unapproved apps and reporting. Agree the intended behavior and test it against sample approved tasks. A control that interrupts legitimate work needs review. A control that allows prohibited data needs correction. Give each issue an owner and a follow-up decision.
The NIST Guidance for AI Risk provides voluntary guidance for managing AI-related risk. It does not certify a company's AI use. For a small business, a practical starting point is to connect risk decisions with people in charge, clear data limits and review. The advice here is operating guidance, not an assertion that a specific process is already in place.
Support Useful Work with Human Review
Approved adoption needs more than access to a tool. Choose a department-level task, describe a useful result and find the person responsible for checking it. Practice with material that fits the approved data rules. Ask the reviewer to check meaning, accuracy and suitability for the intended audience. The person using the result should be able to explain the decision to use it.
For Microsoft Copilot readiness, include the existing Microsoft 365 permission structure in planning. Microsoft's privacy records states that Copilot surfaces organizational data within the user's existing permissions. Review broad access and roles rather than assuming a new license resolves earlier sharing decisions.
Use the process consistently: See It. Govern It. Use It Well. Discovery sets the relevant activity, governance defines the limits and enablement supports approved work. Regular review connects all three as staff, tools and business needs change. Leadership gets a process it can run, with records showing who approved each decision and why.
Choose the Next Step
Start by listing what you know and what remains uncertain about AI use. Include approved apps, data concerns and any planned Copilot rollout. Bring the people responsible for IT and company data into the conversation. They can help decide what to look at and what the report should answer.
Schedule an AI Discovery Scoping Call to discuss the paid AI Usage Discovery. The AI governance hub explains how the findings can lead into policy, controls, training and regular review.
Connect This Guidance to Your Business
See how these steps fit your business. Start with AI governance and security and AI governance and adoption. Our guide for DFW law firms covers access, staff tasks and client work.
Put Your AI Policy in Writing
Use the AI acceptable-use policy template for small businesses to document approved use, data limits and staff tasks. Adapt the draft to your business and review it with counsel before adoption.
Free Checklist for Business Owners
AI Acceptable Use Policy TemplateA starting policy for how your team uses ChatGPT, Claude, Copilot and Gemini.
