What Should HR Actually Build With AI?

HR already has more AI ideas than it can use. SHRM recently mapped 138 AI use cases across HR.¹ The hard part is choosing which ideas are actually worth building.

To make that decision, I use the classic Impact–Effort Matrix.

Impact is the value an idea could create. Effort is the time, cost and complexity required to make it work.

Plot each idea in one of four quadrants: Quick Wins, Major Projects, Fill-Ins and Thankless Tasks.

HR work can affect people’s jobs, pay, benefits and legal rights. The higher the consequence of an error, the more careful the testing and oversight need to be. A weak email draft is easy to fix. A bad employment decision is not.

Start with the easiest useful work.

I would start with Quick Wins and a few low-risk Fill-Ins. They are easier to test, easier to improve, and give the organization experience before it takes on larger projects.

A morning brief is a simple Fill-In. With an approved AI tool connected to your calendar and email, you can summarize your day, prepare for meetings and surface messages that need attention. The value is useful, and the effort is low.

Quick Wins matter more because they combine low effort with meaningful value.

AI changes what is worth building.

I recently built an interactive ROI calculator for Wellhana. I use it with HR leaders to make the potential value of employee financial wellbeing visible. We can change the assumptions together and immediately see how the result moves.

A few years ago, I might have paid thousands to design and build something like it. With AI, I built the first useful version myself.

The same logic applies across HR. You could build a recruitment ROI calculator before asking for more recruiting budget, or turn workforce data and source material into a leadership briefing before making the case for a new initiative.

Once the smaller builds prove useful, move up the matrix. A natural-language HR analytics tool could create much more value, but it also requires clean data, permissions, system access and more careful testing.