Case study · Insurance

An AI assistant for a national HR function

An AI assistant that answers employee questions from company policy, estimated to reduce HR workload by 25% and evaluated for global rollout.

25%estimated reduction in HR workload
Policy-groundedanswers drawn from the company's own documents
Globalevaluated for rollout across the group

The context

In a large organisation, much of an HR team’s day goes on answering the same questions: how much leave is left, how a benefit works, which form applies, what the policy says about a particular situation. The answers exist, spread across policy documents, intranet pages and handbooks, but finding them is easier for HR than for employees, so the questions keep coming.

The challenge

The HR function of a global insurer’s national business wanted to free its team from that repetitive load without lowering the quality of answers. Accuracy mattered more than speed: in HR, a confident wrong answer is worse than no answer at all.

What we did

We designed the assistant together with the regional head of HR, starting from the questions employees actually ask and the documents that answer them.

The assistant answers employee questions using the company’s own policy documents, and shows where each answer comes from. When a question is personal, sensitive or not covered by policy, it does not improvise: it hands the case to the HR team. That design kept the team in control of anything requiring judgement, and let routine questions be resolved immediately.

The result

The system was estimated to reduce HR workload by 25%, and on the strength of that result it was evaluated for rollout across the group.

Why it worked

  • It was grounded in the company’s own documents. Answers came from policy, with sources shown, not from general knowledge.
  • It knew when to hand over. Clear escalation for sensitive or uncovered cases protected trust in the system.
  • It was designed with the people who own the process. Working with HR leadership from the start shaped both the scope and the adoption.

Start with a diagnostic.

Share one report. Within seven days, we show you in money what AI would have delivered on your own past numbers. If it does not deliver, you do not pay.

Book a diagnostic →