Start with the AI services people actually use, including features inside existing software. For each use case, record the owner, provider, purpose, users, data involved, connected systems, and whether the output leads to an action or decision. Note the approved configuration and where review or monitoring records are kept. Ask business teams about everyday use rather than limiting the inventory to engineering projects. Keep the list current as products and settings change. A simple inventory gives the team somewhere to begin when deciding what is allowed, what needs closer review, and what belongs in an audit discussion.
Related: SOC 2 audit and compliance hub.
Further reading: NIST AI Risk Management Framework.