Bringing AI tools in is fast. What comes next is where it stalls. When we assess companies, the reason is rarely model quality — it is the lack of safeguards that let the company trust AI with real work. These nine points are the first things we check.
- No record of who used what → log questions, attachments and answers end to end
- Confidential or personal data can leave → inspect and mask inputs before sending
- People use personal accounts → one company gateway to the contracted AI
- Costs are uncontrolled → per-department totals and limits; agree who pays API/LLM fees before signing
- AI touches systems without permission → allow-lists for systems and actions
- Answers that are not true → answer only from company sources, decline without a source, show sources
- Non-compliant wording goes out → check output before sending, route hits to a person
- Locked into one AI → route to another model when prices rise or outages happen
- Nobody uses it after launch → watch real usage, train in the new order of work, leave a guide
Finding where you are stuck is part of the free assessment. "You don’t need development" is a valid conclusion.
All insights