AI assistants are good at the parts of HR work that are mostly lookup and paperwork: finding a balance, checking who is away, drafting a reply, filling in a request. They are less suited to the part that matters most: deciding things about people. The safest way to use AI in HR is to draw a clear line between the two.
What AI is good at in HR
Much of an HR team’s day is answering questions whose answers already sit in the system:
- “How many days of leave do I have left?”
- “Who on my team is off next week?”
- “Which requests are waiting for me?”
An assistant that reads from your own HR data can answer these in a sentence, which saves the back and forth of a message to HR. It can also prepare the next step, such as filling in a leave request from “I’d like the 2nd to the 6th of November off”.
Where a person has to stay in charge
Approving leave, changing a salary, recording a performance outcome: these decisions affect someone’s job and pay. They need a person who is accountable for them, and the employee needs to know a person made them.
The practical rule is simple: the assistant may prepare a change, but only a person can make it. The assistant fills in the details and shows them on a card; nothing happens until someone presses Confirm. If they press Dismiss, nothing changes.
Keep the assistant inside the same permissions
An assistant must see no more than the person using it would see. An employee’s assistant should not be able to read a colleague’s records just because the question was phrased cleverly. The simplest way to guarantee this is for the assistant to work with exactly the same access as the person asking, rather than a powerful system account.
Be careful with sensitive details
Some HR data is especially sensitive. The type of a colleague’s leave can reveal health or family circumstances. A careful assistant can tell a manager that a request is waiting, and for which dates, without passing the reason or the leave type to the AI model at all.
Questions to ask any HR AI tool
- Can the assistant change data on its own, or only prepare a change for a person to confirm?
- Does it act with the permissions of the person asking, or with broader access?
- Which data does it send to the AI model, and is sensitive data held back?
- Is every change it prepared, confirmed or dismissed recorded?
How astraHR handles this
astraHR’s assistant answers from your organisation’s own data, scoped to what the person asking is allowed to see, because it acts with their own access. For leave it can look up balances, requests and who is out, and it can prepare a request, a cancellation or a decision. A prepared change appears as a card and takes effect only when the person presses Confirm. Other people’s leave types and reasons are never sent to the AI model. The assistant is part of the Complete plan and switches on after the first payment.