Rimi Baltic: AI-supported pricing across three countries

Context

Rimi Baltic, part of Sweden's ICA Group, operates roughly 280 hypermarkets, supermarkets and convenience stores across Estonia, Latvia and Lithuania. Grocery pricing across three countries with different competitive situations is complex: thousands of products, frequent promotions, and price rules that were previously managed largely by hand. This is one of the few publicly documented AI deployments by a retailer headquartered in the Baltics — which is exactly why it belongs in this library.

What was deployed

In January 2019, Rimi Baltic decided to implement price optimisation software from Revionics (later acquired by Aptos), which uses machine learning to model price elasticity — how demand for each product responds to price changes — from historical sales data. The rollout was deliberately staged: in a first phase of about four months, Rimi migrated its existing rule-based pricing onto the platform; only then did it switch on elasticity-based optimisation, trained on two years of historical sales. The first product category went live in March 2019, and the rollout completed in under a year (Retail Optimiser, November 2020).

Reported results

Rimi Baltic's promo and pricing manager, Mārtiņš Ἲezberis, stated publicly that the company "were definitely able to increase sales" and that "the increase was higher than we had projected in our business case" (Retail Optimiser). Notably, he declined to publish specific figures, because COVID-19 lockdowns made year-on-year comparisons unreliable. No quantified uplift exists in the public record, so none is given here.

What to learn

• Rule migration before optimisation. Rimi did not jump straight to machine learning. It first moved existing pricing logic onto the new platform, then layered AI on top.
• Two years of clean historical data was the entry ticket. Elasticity models are only as good as the sales history behind them.
• Respect honest non-disclosure. A practitioner refusing to quote numbers because the comparison baseline was distorted is a sign of rigour, not failure.

Source: Retail Optimiser, November 2020.