Where are we relative to the market?
Using P25, median, P75, and compa-ratio as reference points.
Case Page HOW HR CAN ACTUALLY USE AI
This case asks a simple compensation question: how can AI help HR move from benchmarking inputs to clearer, more consistent decisions?
Case Board | Compensation Thinking | AI in HR
How HR Can Actually Use AI
From benchmarking to better compensation decisions: a structured exploration of where data stops being enough, and where decision logic starts to matter.
Reasoning Flow
Problem
Benchmarking gives inputs, not decisions.
Benchmarking gives inputs, not decisions.
Thinking
Market, internal context, and decision type.
Market, internal context, and decision type.
Output
A more explainable compensation decision.
A more explainable compensation decision.
Opening Thought
Everyone is talking about AI in HR lately.
Chatbots, automation, dashboards — new tools keep emerging. But I kept wondering what this actually means for the kind of work I do in compensation.
Because in practice, most compensation decisions do not look like obvious AI use cases.
They look more like this: a hiring manager asking why two similar roles have very different offers, or trying to explain why someone is below market — but still within range.
And that is where I started noticing something.
The challenge is not that we do not have data. It is that we do not always have a consistent way to turn that data into decisions.
Rethinking The Role Of Benchmarking
In most organizations, benchmarking is already part of the process. We have market data, salary ranges, and external reports.
But those inputs do not automatically translate into clear decisions.
In many cases, the same data can lead to different conclusions — depending on who is making the decision and how they interpret it.
That is when I started to think: maybe the real gap is not data, but structure.
And that is also where AI starts to become interesting. Not because it can replace compensation judgment, but because it may help make that judgment more consistent, more explainable, and easier to govern.
A More Structured Way To Think
I began sketching a simple way to approach these situations — not as a perfect model, but as a way to make the thinking more consistent and explainable.
At a basic level, most decisions seem to come down to three questions:
What’s happening internally?
Looking at internal ranges and how the case compares to peers.
What kind of decision are we making?
An offer, an adjustment, or a correction — each requires a different lens.
Once these elements are clear, the decision itself becomes easier to justify and communicate. That, to me, is where AI could be useful: not as a final decision-maker, but as a layer that helps structure the case before HR or compensation makes the call.
A Simple Example
One situation that made this particularly clear to me was an offer for a mid-level Software Engineer. On paper, it looked reasonable — slightly above the market median. Internally, it was already higher than several existing team members in similar roles.
That was the point where benchmarking stopped looking like a pricing exercise and started looking like a decision problem.
It is not just about finding the right number — it is about making decisions that are consistent, explainable, and sustainable. That is also why I think AI in compensation should be framed less as prediction, and more as decision support.
Child Case
Where AI Might Actually Help
This is where I see potential for AI — not as a replacement for HR judgment, but as a way to support more structured decision-making.
After benchmarking, before final sign-off
AI is most useful in the middle layer: once market and internal inputs are available, but before HR or compensation makes the final call.
Structure the case
Standardize
Normalize case inputs so similar decisions start from the same structure.
Read Consistently
Apply the same market and internal lens across cases.
Surface Tension
Flag issues like internal inequity or explanation risk.
Draft Explanation
Turn the reasoning into clearer manager- or employee-ready language.
Replace compensation judgment
It should not decide pay on its own, override governance, or treat market data as sufficient without internal context.
But that also means AI needs boundaries. It should help organize inputs, frame the trade-offs, and support explanation — not make final pay decisions on its own.
In that sense, AI becomes less of a chatbot, and more of a decision layer.
Closing Thought
Takeaway
Structure before speed.
I’m still exploring this space.
But one thing feels increasingly clear:
AI in compensation is probably not most useful when it tries to find the right number faster.
It becomes more useful when it helps make decisions more consistent, more explainable, and easier to govern.
And in compensation, that shift could make a real difference.