How Rohlik Uses an AI Agent to Save EUR 2.8M in Procurement
How Rohlik uses an AI agent in procurement: the European online grocer captured more than EUR 2.8 million in annualized savings within three months, and its annual supplier negotiations finished one month faster with about 80% of the workflow automated. The system was built by the startup Duvo, which runs on Anthropic's Claude.
Want your team to find the work nobody has time for? See our AI Workshop for organizations.
Key numbers
EUR 2.8 million+ in annualized savings captured within three months (Anthropic customer story: Duvo)
EUR 1.45 million in annualized savings in the first week, from continuous price monitoring and automated outreach across 120+ SKUs and 15+ suppliers (Anthropic)
1 month cut from the annual negotiation cycle (Duvo)
About 80% of the negotiation workflow automated, from preparation through contract to ERP write-back (Duvo)
All figures come from the vendor and Anthropic. They have not been independently audited.
The old problem: terms lost in email, costs fell but prices did not
Rohlik Group is an online grocer in five European markets with more than EUR 1.5 billion in revenue. It tracks thousands of commodity-linked SKUs. Before Duvo, those prices were checked quarterly at best, and often not at all (Anthropic).
Rebates and agreed terms were buried across thousands of email threads, then retyped into contracts. Every week the cycle slipped, Rohlik stayed on old terms (Duvo).
How Rohlik uses an AI agent
1. Monitor costs and build the negotiation case
The agent monitors commodity prices continuously instead of waiting for the annual cycle. When costs move, it builds the case and contacts the supplier (Anthropic).
2. Capture terms from email
It reads ongoing threads, extracts agreed terms, and checks them against margin guardrails and business rules such as payment windows (Duvo).
3. Draft the contract and route approvals
It generates the contract from the latest approved template, routes it for sign-off, and keeps an audit trail of who approved what.
4. Write signed terms to the ERP
Once signed, the agent writes final pricing into the ERP so the shelf price matches the contract.
High-risk actions always require human approval, and the system remembers those decisions for future runs (Anthropic).
Results
EUR 1.45 million annualized in the first week, growing to EUR 2.8 million+ annualized within three months.
Other workflows: promotional setup down 65% to 70%, supplier onboarding chasing down 50% to 70%, product availability up from 78% to 93% in two weeks, and inbound delivery confirmations up from 52% to 90% (Anthropic).
Olin Novák, CRO of Rohlik Group, said: "That was one process. We're now running dozens." (Anthropic)
The money did not come from doing old work faster. It came from work nobody had time to do.
What executives can take from this
Look for "abandoned work": tasks your team knows matter but only covers for the top 20 suppliers.
Have people write the rules for acceptable terms first, then let the agent work inside them.
Measure value in two parts: money captured from cost changes, and time saved from not retyping.
More cases in this series: How Amazon uses AI for product listings and Kroger case study: cutting promotions to lower shelf prices.
Sources
This article is based on public information. IM Impower was not involved in this project.


