Having an AI tool doesn’t mean HR has a working process. Here are seven common assumptions, and one practical check for each.
1. A licence isn’t a working process.
Access lets you use a tool. An HR workflow still needs approved sources, permissions, testing and someone who owns it.
Practical check: Define one task, the information it uses and what the tool is allowed to do.
2. Better setup isn’t a guarantee.
A bad answer could come from a missing source, an access or retrieval problem, or the model’s own limits. Sometimes the task just isn’t a good fit.
Practical check: Investigate the case that failed, fix it and test again. Don’t assume every error is a configuration problem.
3. An approved tool isn’t blanket permission.
Decide which information it may use, who can retrieve it, what it keeps and what actions it’s allowed to take.
Practical check: Test it as the employee who will actually use it, not only as the person who built it.
4. Polished isn’t the same as checked.
Clear wording and tidy formatting don’t prove an answer is right. Check the facts, the links, the meaning and whether it applies to the situation.
Practical check: Whoever sends it should understand it and stand behind it.
5. Useful doesn’t mean autonomous.
AI can organise notes, draft a process map, prepare a report or flag records for review. Choose work where someone can check the result.
Practical check: Compare the whole job, including checking, corrections and upkeep, not just how fast the first draft appears.
6. Scope decides the project.
Answering policy questions needs selected guidance. Answering personal questions needs controlled access to records. Taking action in a system needs permissions and approvals. These are three different projects.
Practical check: Ask for an estimate for the work you actually need, not an enterprise price applied to every use.
7. A chatbot isn’t an HR department.
A chatbot can be one way to get information. It shouldn’t be the only way to get help, and people stay accountable for decisions.
Practical check: Name who handles exceptions and sensitive situations, and make that route easy to find.
Where to start.
Choose one task and test the whole setup:
- Define the job: the information it uses, the sources that are approved and what it’s allowed to do.
- Check the result: accuracy, access, exceptions and the total effort involved.
- Assign an owner: for updates, follow-up and a fallback when it doesn’t work.
Getting access is the start. Trust comes from evidence.
See how we build this: HR Knowledge Agent · HR Agent Readiness Review