Where AI Works in HR (And Where It Doesn't)

The risk in not using AI in HR isn't job loss. It's that your competitors are already using it — and every month you wait, the gap gets wider.

---

Most HR leaders I talk to are waiting. Waiting until they understand AI better. Waiting until the regulations settle. Waiting until someone they trust uses it first.

That's the real risk.

The risk in not using AI for HR isn't that it's going to take your job. It's that your competitors are already using it, and the gap between the orgs experimenting and the orgs waiting compounds every month. Not in the technology. In how strategic the people function gets to be.

This is for HR Directors and Executive Directors at small-to-mid-size organizations who keep hearing about AI in HR but don't know what it actually means for their team. It covers where AI is working right now, what one client did with it that surprised even me, and — just as important — the places AI absolutely should not be making the decision.

The reframe: AI isn't replacing HR, it's freeing it

The panic I hear about AI in HR is almost always framed wrong. The fear isn't "my organization will fail." It's "I'll be replaced."

Successful organizations aren't replacing HR people with AI. They're using AI to offload the work that was never the real value of HR in the first place.

A typical HR generalist spends 30 to 45 minutes writing a single job description. With AI assistance, that drops to a few minutes — seconds if you're not editing carefully. Same for first drafts of training presentations, policy summaries, internal newsletters, FAQs, meeting notes. The work still gets done. It just stops being the work that fills the calendar.

The orgs getting this right are taking the six hours a week HR gets back and reinvesting it in the things employees actually feel: engagement programs, career development, retention conversations, wellness initiatives. Those are the things that move the needle on culture. Not the job description nobody re-reads after week one.

That reframe is the whole point. AI isn't the threat to HR. It's the thing that finally lets HR do what HR was supposed to do.

A case study: using AI to build human connection

The most surprising AI project I've worked on wasn't about efficiency. It was about loneliness.

A national nonprofit client had employees across roughly 30 states. Fully remote. Their onboarding was loose, people didn't know who to go to for what, and underneath the operational problem was a quieter one: people weren't connecting with each other on a human level.

We built something we called the Compass Project — a chatbot trained on every job description at the organization, plus voluntary demographic information employees opted to share. Interests, hobbies, side projects. All opt-in. None of it stored as sensitive data.

Two things happened.

A new hire working on a grant project didn't know who to ask. They pinged the chatbot, described the work, and got pointed to the right person in another department in seconds. Workflow unblocked.

Then someone asked the chatbot who else at the org gardens. It pointed them to a colleague in Finance. They connected over something completely unrelated to work — and the connection stuck.

That second piece is what most organizations miss when they think about AI. The chatbot didn't replace human connection. It created the conditions for it across a workforce that would never have met in person.

That's what good HR tech looks like. It removes the friction so the human part can happen.

Three more places AI is earning its keep in HR right now

The Compass Project is the example I get the most questions about, but it's not the only place AI is doing real work. A few others worth knowing about:

One-on-one career coaching.

Picture an employee at level two who wants to be promoted into a manager role. They don't always have a clear path. An AI coaching tool — trained on the org's own role levels and competencies — can help that employee identify the skills they need to develop, suggest stretch projects, and give them a clearer self-assessment than most managers have time to provide. It doesn't replace the manager conversation. It gives the employee something concrete to bring to it.

Recruitment, done carefully.

Tools like Lattice and similar ATS platforms can anonymize resumes to reduce bias in the early screen, summarize hiring panel feedback, and surface sentiment patterns across multiple interviewers. The keyword is carefully. These tools should inform decisions, not make them. We'll come back to that.

Learning and development.

Most company-wide training is one-size-fits-all because that's the only thing that scales. AI changes that math. You can build personalized learning paths for each employee, tied to their actual performance review and role expectations. Skills gaps close faster, and the training feels relevant to the employee — because it is.

Three places AI should never make the decision

This is the part I want HR leaders to be loud about inside their organizations, because it's where the legal and ethical risk lives.

AI should not be making hiring decisions.

Several states have laws regulating the use of AI in employment decisions, and the regulatory landscape is moving fast. Beyond the legal exposure, AI models have documented bias issues when trained on historical hiring data — they tend to replicate the patterns that produced the problem in the first place. Use AI to screen for skill-keyword matches, anonymize resumes, summarize feedback. Don't use it to score candidates or recommend final hires.

AI should not be making separation or firing decisions.

This is a place where the human context — the back story, the manager dynamic, the documentation history, the legal record — matters more than the data. AI cannot weigh nuance. It can flag attendance patterns. It cannot understand why those patterns exist. Anything touching a separation, a PIP, or a constructive feedback conversation should be human-led.

AI should not be handling sensitive employee data, complaints, or crises.

This one is the most overlooked and the most dangerous. A well-meaning HR practitioner pasting an employee census into a public AI chatbot is creating a data privacy incident. Names, dates of birth, Social Security information, medical history, mental health disclosures — none of that should ever be input into a general-purpose AI model. The same goes for active employee complaints or anyone in crisis. AI cannot read the room. It cannot recognize when a question is the symptom of a much larger problem. That work stays with people.

There are gray-area uses worth being thoughtful about. Compensation analysis is a good example. Using AI to crunch benchmarking data, model scenarios, and surface anomalies is appropriate. Letting AI tell you what to pay each employee at year-end is not. The pattern is: let AI inform the decision. Never let it own the decision.

What to do this week

Most leaders don't get stuck on whether to use AI. They get stuck on understanding the whole space before they touch any of it. Don't.

Find the one task your HR team does every single week that takes the most administrative time. Job descriptions. Policy updates. Onboarding paperwork. Meeting summaries. Whatever the single biggest drag is.

Spend one hour this week researching whether a tool exists to handle it. There almost certainly is. You don't need to understand the full scope of AI. You need to know whether your team's most repetitive task can be done in a tenth of the time.

That's the start. The competence builds from one task to the next.

You're not going to fall behind because you didn't read every AI white paper. You're going to fall behind because you waited for permission to start.

---

If you want to see where your organization stands across the full HR function — not just AI — take the HR Health Report. 40 questions, 10 minutes, real diagnostic of where your HR posture is strong and where the gaps are.

If you'd rather talk through where AI fits inside your specific org, let's set up a conversation.