Jim Barnett is the CEO and Co-Founder of Wisq, an Agentic AI platform for HR.
HR can be messy work, full of context that doesn’t fit neatly into a form and judgment calls without clean, documented answers.
Take, for example, an employee who has been late three times this month. Her manager wants to know if that’s grounds for a written warning. Before anyone can answer that, HR has to find out whether tardiness has come up before and whether any circumstances explain it. Only after asking a few questions does anyone know if the situation amounts to a performance issue or something else. Whoever signs off on the decision owns it, including if it gets challenged later.
Simple Requests Don’t Need Judgment. They Need A Rule
HR teams can look to IT as an example of a department that has been there, done that with routine question-answering. “Reset my password.” Done, here’s the link. “Send me a new laptop.” Done, it’s on the way. “Update my dependent count from two to three.” Done. None of those needed much context. They mostly needed rules.
HR has requests like that, too. “Where can I find our travel and expense policies?” “When is the next company holiday for my team?” Both questions are pretty easily answered. Neither one needs investigation. Both just need a lookup, and many AI solutions built for HR today can already handle that part well.
But an attendance pattern is much more complicated than a password reset. So is a leave request that depends on a state law and a set of personal circumstances that aren’t documented. Those requests don’t have a rule to match against; they have context to weigh and a decision someone has to own, which is a different kind of problem than IT ever had to solve. These complex HR inquiries can also be multiparty, multi-day and long-horizon, or require a handoff to the appropriate person on the HR team.
Ambiguous Policy Breaks Consistent AI Answers
Leave policies make the gap obvious. A bereavement policy might read: “discretionary leave for the death of a close family member, pending HR approval.” It’s one sentence, but it isn’t sufficient on its own. Questions branch off of the policy. Who counts as close family? How many days is reasonable, and what happens if someone needs more time? The knowledge usually lives with whoever has handled the last twenty requests, and it leaves the building when they do.
AI’s consistency is often dependent on the policy behind it. If five different versions of an employee handbook are floating around, or a policy leans on jargon that means something specific to HR and nothing to anyone else, some AI solutions might actually guess. Sometimes the guess will be reasonable, and sometimes it won’t. There’s no way to know in advance.
The fix is cleaner policy and a system trained to understand how things actually get done and not to guess. This means fewer versions floating around, plain language instead of jargon and unwritten judgment calls, like how many bereavement days is reasonable or what counts as a close family member, written down somewhere both AI and humans can find them.
Most HR AI Still Runs On IT’s Ticketing Model
Take a closer look at how most AI marketed for HR actually works, and it’s the same logic IT support has used for years. Take the request and match it to a rule. Return an answer and close the ticket. That model might work for the password reset and the dependent count change, but it is clearly inadequate for the complex, messy inquiries that consume HR teams.
Some vendors have gotten good at the simple question answering. The problem is that the easy requests and the messy ones still run through that same ticket-shaped logic, because that’s the model most of these systems started from.
The stakes make this worth getting right. A wrong answer on a vacation balance gets corrected in a minute. A wrong or inconsistent answer on a leave-of-absence question can turn into a compliance problem or a discrimination claim.
There are two ways to build AI for HR: Extend the generic Q&A and IT ticketing model and either hope it stretches to cover judgment calls too (hint: it won’t), limiting HR automation to a small percentage of inquiries and situations, or build something meant for HR reasoning and judgment from the start. As HR teams know well, most of what HR actually handles looks more like a complicated attendance or corrective action question with a story behind it than a password reset.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?