EY · Tech Consulting · Medium · 25-35 min

European Retailer Deploying GenAI Under the AI Act

European Retailer Deploying GenAI Under the AI Act is a medium EY tech consulting case interview that runs 25-35 min. A European grocery and general-merchandise retailer — €14B revenue, 1,100 stores across Germany, France, and the Netherlands — wants three GenAI uses live this year: a customer-service assistant, a store-labor scheduler, and personalized offers. A strong answer works through 5 phases: Classify each use under the AI Act; Design the assistant: retrieval over policy and product data, disclosure that it is a machine, human handoff; Treat workforce scheduling as the dangerous one; Separate personalization from prohibited manipulation; keep a purpose limitation and a way to opt out; Recommend a launch order and the evidence file (data, human oversight, logging) the general counsel can defend.

Last updated 2026-09-05

The brief

A European grocery and general-merchandise retailer — €14B revenue, 1,100 stores across Germany, France, and the Netherlands — wants three GenAI uses live this year: a customer-service assistant, a store-labor scheduler, and personalized offers. The general counsel has circulated the EU AI Act and GDPR together and asked which of the three is high-risk. Works councils in Germany will be consulted on anything that touches scheduling. Customer-service volume is 42M contacts a year. You are advising the CDO on what can launch, what needs a conformity path, and what should wait.

How to approach it

  1. Classify each use under the AI Act — limited-risk transparency vs high-risk employment — and under GDPR lawful basis, before picking a model
  2. Design the assistant: retrieval over policy and product data, disclosure that it is a machine, human handoff
  3. Treat workforce scheduling as the dangerous one — works council, high-risk duties, human review of the roster
  4. Separate personalization from prohibited manipulation; keep a purpose limitation and a way to opt out
  5. Recommend a launch order and the evidence file (data, human oversight, logging) the general counsel can defend

What a strong answer does

  • Classifies the three uses differently instead of applying one AI policy to all of them
  • Flags employment scheduling for works-council consultation and a heavier AI Act path
  • Grounds the assistant in retailer content and discloses it, rather than a free-form model on the public site
  • Names GDPR roles — controller, processor, retention — next to the AI Act classification
  • Stages launch: assistant first, offers second, scheduler only with a human on the loop

Red flags interviewers score down

  • Calls all three 'just a copilot' and skips classification
  • Lets the scheduler auto-assign hours with no human review and no works council
  • Trains on customer chats with no lawful basis
  • No logging or oversight story for the general counsel

Cases are written in each firm's style, written and reviewed by working consultants; they are not the firms' own published cases.

More EY cases