Your Employees Are Already Using AI Tools You Haven’t Approved
If you haven’t built a shadow AI tool intake process yet, you’re not preventing AI tool adoption — you’re just not tracking it. Someone on your marketing team signed up for an AI writing assistant last Tuesday. A developer added a code-completion plugin to their IDE two weeks ago. Finance is experimenting with an AI-powered spreadsheet tool they found on a YouTube ad. None of them asked IT.
on Unsplash
That’s the reality at most SMBs right now. The tools are cheap, the sign-up friction is near zero, and employees genuinely believe they’re being productive. They’re not trying to create a governance problem — they just don’t have a channel that makes asking IT feel like anything other than a dead end.
The answer isn’t a six-month procurement cycle. It’s a lightweight shadow AI tool intake workflow that gives employees a fast, visible path to approval and gives IT a defensible record of what’s running in the environment.
What “Lightweight” Actually Means Here
Lightweight doesn’t mean informal. It means the workflow has as few steps as necessary to accomplish three things: capture the request, assess the risk fast enough to stay relevant, and document the decision. If your intake process takes longer than a week from request to decision for a low-risk tool, employees will stop using it and go back to self-provisioning.
The four components of a functional intake workflow are a request form, a triage rubric, an approval or denial with written rationale, and an inventory update. Each one is straightforward. The friction is usually in connecting them and making the whole process visible enough that employees trust it will actually move.
Step One: The Request Form
The form exists to capture the minimum information you need to make a risk decision. You don’t need a fifteen-field questionnaire. You need to know the tool name and vendor, what the employee plans to use it for, what data they expect to put into it, whether it requires a login tied to company credentials, and whether a free tier or paid account is involved.
That last point matters. Free-tier AI tools almost universally train on submitted data unless you opt out or pay for a privacy-preserving tier. An employee pasting customer contract language into a free AI summarization tool is a data handling issue you want to catch at intake, not after a client asks where their NDA went.
Keep the form short enough that filling it out takes under five minutes. If it’s longer than that, you’ll get incomplete submissions or people skipping it entirely. A Google Form, a Jira ticket type, or a simple intake queue in your existing ITSM platform all work. The channel matters less than whether employees know it exists and believe it gets a response.
Step Two: Risk Triage
Not every AI tool request carries the same risk, and treating a no-code AI image generator the same way you’d treat a third-party AI model with API access to your CRM is a waste of time for everyone. A triage rubric lets you sort requests into tiers quickly so you spend your review time where it actually matters.
A simple three-tier approach works for most SMBs. Low-risk requests involve tools with no company data input, no SSO integration, and no regulatory surface — think AI-powered grammar checkers running entirely in the browser with no file upload. Medium-risk requests involve tools that will touch internal data but operate under a documented privacy policy and don’t require privileged system access. High-risk requests involve tools with API access to company systems, tools that will process regulated data (PII, PHI, financial records), or tools from vendors with no published security documentation.
For low-risk tools, a twenty-four-hour turnaround is reasonable. For medium-risk, aim for three to five business days and do a quick vendor security review — check whether the vendor has a SOC 2 report, review their data processing and retention terms, and confirm what happens to submitted data. For high-risk tools, the review needs to follow the same rigor as any significant vendor evaluation: data processing agreement, security controls, subprocessor list, and a clear answer on whether the tool’s AI training uses your data.
This triage step is also where you catch scope creep. An employee requests a tool for one use case; your form review reveals they’re planning to pipe customer data through it for another. Better to clarify that now than to discover it during an audit or a breach notification.
Step Three: Approval or Denial With Rationale
This is the step most informal processes skip, and it’s the one that creates the most downstream problems. If your answer is just “approved” or “denied” with no context, you’ve missed the opportunity to teach employees what good AI tool use looks like, and you’ve left yourself with no documentation if the decision gets questioned later.
For approvals, the rationale should note what the tool is approved for, any conditions attached (for example: approved for internal draft generation only, not for customer-facing content without human review), and any data handling restrictions that apply. This becomes the usage policy for that tool, and it should go into your AI tool inventory alongside the approval record.
For denials, the rationale should be specific enough that the employee understands what would need to change for the request to be approved. “Vendor does not have a data processing agreement available” is actionable. “Does not meet security requirements” is not. When employees understand why a tool was denied, they’re more likely to bring alternative options back through the intake channel rather than just using the denied tool anyway.
If you’re operating under frameworks like the NIST AI Risk Management Framework or preparing for EU AI Act compliance, documented approval rationale is also evidence of a functioning governance process — not just good practice, but demonstrable due diligence.
Step Four: Inventory Update
Every approved tool goes into a central AI tool inventory. Every denied tool gets logged too, with the denial reason, because employees change roles, vendors update their security posture, and what was a reasonable denial six months ago might be approvable today — or vice versa.
Your inventory entry for each tool should include the tool name, vendor, date approved, approved use case, data classification of information permitted to enter the tool, the name of the person who approved it, and a review date. That last item is not optional. AI tools change their terms of service, their data handling practices, and their pricing tiers on a regular basis. A tool you approved under one set of conditions may be operating under materially different conditions twelve months later. A scheduled annual review for each inventory entry keeps your approvals current without requiring a full re-evaluation every time.
The inventory also gives you a clear answer when your leadership team asks “what AI tools are we using?” Right now, most IT managers at SMBs can’t answer that question with confidence. A maintained intake log changes that.
Do This Week
Build the request form and publish the channel. You don’t need the full workflow operational before you open the intake path — you need employees to know where to send requests. Set up a simple intake form with the five fields described above, send a two-sentence note to your company telling people it exists, and commit to responding to every submission within five business days. You’ll catch requests you didn’t know were in flight, and you’ll start building the inventory from the first submission forward.
Once you have a few submissions in hand, you’ll have a much clearer picture of what triage rubric actually fits your environment — what your employees are asking for tells you where your risk surface really is.
InfoDefenders’ tool intake queue gives IT a single place to review employee AI tool requests, triage risk, document decisions, and maintain the inventory — without building the workflow from scratch in a spreadsheet. If you’re ready to move from ad hoc to operational, see how it fits into the platform.
The Bigger Picture
A shadow AI tool intake process doesn’t solve every AI governance problem you have. It doesn’t address the AI capabilities baked into tools you’ve already procured — your Microsoft 365 Copilot rollout, your CRM’s built-in predictive scoring, the AI features your HR platform quietly enabled last quarter. Those require a different kind of inventory and a different governance conversation.
But shadow AI tool intake solves the specific, immediate problem of employees adopting tools outside your visibility. It gives you a mechanism that respects the reality that people are going to use AI tools whether IT builds a process or not, and that the job is to make the sanctioned path faster and clearer than the self-service path.
The companies that are going to have the hardest time with AI governance over the next few years are the ones that tried to stop adoption rather than channel it. Build the intake process. Make it fast. Make the rationale visible. That’s the foundation everything else sits on.
