You cannot govern AI usage you cannot see, and most AI adoption happens through free tiers and browser extensions that never touch procurement. This inventory is built to surface those, which a record of what you bought will not show.
Discover before you survey. Asking teams what AI tools they use under-reports badly. Triangulate SSO logs, expense and card data, browser extension inventories, and network egress to AI domains, then use the survey to fill gaps.
Record whether the vendor trains on your data. This is the single most consequential column. Free and consumer tiers frequently do; enterprise tiers usually do not but often only if explicitly configured. Record the actual contractual position, not the marketing claim.
Classify each tool as sanctioned, tolerated, or unsanctioned. Three states rather than two. 'Tolerated' is honest about tools in genuine use that have not been through review, and gives you a working queue instead of a fiction.
Here's a preview. Unlock the free download to get all 3 sections (2 more below).
| Tool | Vendor | Used by | Purpose | Data it receives | Trains on our data | Contract type | Status | Owner | Annual cost |
|---|---|---|---|---|---|---|---|---|---|
| Enterprise AI assistant | Vendor A | All staff | Drafting and summarising | Internal, Confidential | No (contractual opt-out) | Enterprise, DPA signed | Sanctioned | IT Director | £28,000 |
| Code completion | Vendor B | Engineering (24) | IDE code suggestions | Source code | No (enterprise tier) | Enterprise, DPA signed | Sanctioned | Eng lead | £7,200 |
More free templates IT teams use alongside this one.
Banning AI tools does not work. It moves usage onto personal accounts where you have no visibility at all. This policy assumes people will use AI and makes the boundary explicit: which tools, which data, and how to get something new approved.
An AI approval process that takes six weeks guarantees people stop asking. This form is deliberately short: enough to make a sound decision, fast enough that requesting approval stays easier than going around it.
AI risk assessments often reduce to a data protection review, which misses the failure modes specific to these systems: confidently wrong output, drift, prompt injection, and decisions affecting individuals. This template covers those explicitly.
Triangulate SSO and OAuth grant logs, expense and card data, browser extension inventories from endpoint management, and network egress to AI domains. Surveys alone under-report significantly because free-tier usage does not feel like adopting a tool.
The tool and vendor, who uses it and why, what data it receives, whether the vendor trains on your data, the contract type and whether a DPA exists, approval status, a named owner, and cost.
AI tools used for company work without going through review: usually free tiers, browser extensions, or personal accounts. It matters more than ordinary shadow IT because the data leaving the organisation is often unstructured and may include customer information.
