The “No-Regrets” Move: Fix Your Knowledge Before You Build Anything Clever

By John Dicken, Senior Director, AI-Enabled People Solutions, Kantar

Article one was about separating process from human wisdom before touching any technology. So what do you do first? Here is the one move from our own journey I would call genuinely no-regrets, whatever your AI plans turn out to be.

Two years ago we were asked a deceptively simple question: “Can you review our 4,000 HR policy documents and bring them up to a single global standard aligned, compliant, on-brand?” Multiple languages, inconsistent formats, scattered across the business. The job no one wants.

Why we started here, not with the exciting stuff

It would have been easy to skip straight to building an advisory agent for colleagues to talk to. But point a clever agent at 4,000 inconsistent, outdated documents and you get a clever way of surfacing bad answers, faster. A regret waiting to happen.

So instead of building the front door first, we built small, task-based agents in Copilot Studio, each doing one narrow, unglamorous job: translation, policy checking, metadata tagging, consolidating duplicates. None clever alone. Together, transformative.

What actually happened

In six weeks, those agents helped turn 4,000 scattered artefacts into 400 structured, validated documents – a clean, trusted knowledge base now supporting colleagues in 60 countries. It was significant enough that Microsoft featured it as one of their own customer stories, not because the technology was exotic, but because the outcome was real and repeatable.

The lesson was not “buy an agent platform.” It was: fixing your knowledge to a genuinely AI-aligned standard is a no-regrets investment. Whatever you build next is only ever as trustworthy as the content underneath it.

What this means for you, whatever stage you are at

You do not need 4,000 documents or six weeks. It scales down: find the knowledge base your team trusts least and ask whether small agents could bring it up to standard before you build anything colleague-facing on top. Low-risk, and it compounds every future agent gets better because the foundation is clean.

Next in this series

Getting the knowledge right is necessary, not sufficient. In article three, I turn to the harder challenge: building an AI enabled People team that is capability-led, not technology-led so people choose to lead this change rather than have it imposed.

If you would like to explore this in person, Jordan Mori from Argon and I will be hosting a small, invite-only round table for CPOs and HR leaders on AI in HR on 15 October, where we will be going deeper into moving from isolated experiments to a genuinely redesigned way of working.

You can find out more about the event here.