A 100-Year-Old Retailer's AI Answers Work Questions in Seconds

Picture a new employee at a big company trying to find out how to request access to a system. The answer exists somewhere — buried in a wiki, an IT portal, or a colleague's memory. At HEMA, a Dutch retailer with more than 750 stores and a century in business, that hunt used to eat entire afternoons. Engineers called it "portal-hopping." As the company grew, the old "just ask the person next to you" habit stopped working.

So HEMA built an internal AI assistant named HAL. Instead of forcing people to visit yet another website, HAL brings answers to them — inside the chat window or app they already have open. Two technologies make this work. One is MCP, which you can think of as a universal adapter that lets the AI plug into company systems without custom wiring for each one. The other is Amazon Bedrock AgentCore, the engine running behind the scenes. Security runs through Microsoft's Entra ID, so employees log in the way they always do.

The payoff is speed. Questions that once meant navigating three or four portals, sometimes over a full afternoon, now get answered in seconds. What started as a tool for engineers has spread to product owners and business analysts. HEMA says it's already changing how teams work day to day.

The catch: HAL only knows what HEMA deliberately feeds it. Structured records like service catalogs transferred cleanly, but messy, unwritten knowledge is harder to capture. HEMA's next step is bigger — turning HAL from something that answers questions into something that can actually take actions, like provisioning access. That's more useful, but also riskier, because a wrong answer is annoying while a wrong action can cause real damage.