Auchan runs a self-hosted Llama assistant to answer employee HR questions

The problem

Employees at Auchan, the French retail group, have questions for HR such as holiday rules or maternity and paternity leave. Until the assistant, they had to contact the HR team or use third-party software to get answers.

What they built

Auchan built a chatbot that answers staff questions on HR topics. It is limited to HR for now. Employees ask the assistant instead of contacting the HR team. The assistant uses Llama, an open-weight model from Meta. Open-weight means the model files can be downloaded and run on your own servers.

How they did it

The data in this tool is personal and Auchan treats it as confidential, so the company runs Llama on its own infrastructure rather than through an outside service. It uses retrieval-augmented generation (RAG), a method where the system first looks up passages in company documents and then writes an answer from them. Auchan currently runs the model on graphics processors (GPUs). The IT chief says specialised chips for running models could lower the cost later.

Results

Auchan reports that HR managers save one and a half to two days per month. The IT and data chief says the company considered using Llama for every business function, but found it relatively expensive to run, and he calls the return on investment very debatable in Auchan's case. He still sees a real gain in ease of use.

What to copy

Pick self-hosting only where data sensitivity justifies the cost, and check the running costs before you extend it to other teams.