AI prepares the work. Your HRIS still has to run it.

A chatbot can analyse an export or draft an offer letter. It cannot run a compensation cycle or give you an audit trail. Here is how I split the work between AI and the system of record.

My recommendation is simple: use AI to prepare the work and let your system of record carry it out. AI does not replace your HRIS. It makes a good one more useful.

I am writing this because of a post I saw this week called "Goodbye, Workday". The offer was to comment a keyword, receive 15 prompts, and replace a compensation analyst, an HR manager and a financial manager. Hundreds of people commented. It makes a great headline and a poor HR strategy.

Where AI really helps

I work with Workday every day and I use AI every day, so for me this is not a choice between the two. AI is good at analysing an export, drafting an offer letter or turning headcount data into a first scenario. It can also summarise what is written about a person and suggest wording for pay criteria. These tasks save real hours, and I would not give them up.

What they have in common is that the result is a draft. A person reads it, corrects it and decides what to do with it. The AI prepares the work, and the responsibility stays with the people who own the process.

What a chatbot does not do

A chatbot does not run a compensation cycle. A cycle means setting budgets, controlling who can see what, routing recommendations to the right managers, collecting approvals and pushing the result into payroll. Each of these steps depends on the one before it, and each needs rules, permissions and a record of who did what.

There is a second limit that is easy to miss. AI can summarise what is written about a person, but it cannot know what that person actually contributed. If a pay or performance decision relies on a summary of existing text, it carries every gap and bias that text already had. The manager still has to bring the judgment.

Why pay transparency makes the gap clearer

The EU pay transparency directive is coming into national law. Employers will need to explain how pay is set, show the criteria behind each decision and report pay gaps on data they can defend.

That requires three things that a set of prompts cannot give you. The first is a clean job architecture, so that roles are compared on the same basis. The second is controlled data, so that you know where each number came from and who changed it. The third is a full audit trail, so that you can show how a decision was made when someone asks. These are properties of a system of record that is well configured and well maintained. A chat window does not hold them.

AI can still help here, for example by suggesting wording for pay criteria. But the criteria need to live in the system, and the system needs to apply them every time.

How I split the work

The slides that go with this note put the split side by side. On one side is what AI is good at: analysing a compensation export, drafting an offer letter, turning headcount data into a first scenario, summarising what is written about a person and suggesting wording for pay criteria. On the other side is what the HRIS must do: run the full compensation cycle, set budgets and control who can see what, route recommendations to the right managers, collect approvals and push results into payroll, and hold a clean job architecture and an audit trail.

If your team is told that a chatbot can replace the system, this is the test I would try. Take a task and ask whether the outcome is a draft that a person will review, or a decision that has to be applied, approved and recorded. Drafts are a good fit for AI. Decisions that must be applied, approved and recorded belong in the system of record.

What to watch out for

Be careful with headlines that promise to replace whole roles with a list of prompts. Ask what happens after the answer appears on the screen. Who approves it, where is it stored, and who can prove it later?

Also be careful in the other direction. If your HRIS is poorly configured or the data is messy, AI will not hide that. It will work on the same weak foundation, only faster. The better the system and the data, the more useful AI becomes.

I would like to hear from you. Where does AI save your team the most time today, and where did it stop being useful?