Dynamic pricing for the supermarket
Food Waste Challenge 2.0
Kickstart AI x Albert Heijn
Reinforcement learning
Supermarkets significantly contribute to global food waste. Albert Heijn, the leading Dutch supermarket, aims to cut food waste by 50% in 2030. We collaborated on dynamic pricing with reinforcement learning.
Using reinforcement learning for perishable products
Perishable items like fresh vegetables, sushi, and baked goods are particularly prone to spoilage. Driven by its commitment to sustainability, Albert Heijn reached out to leverage cutting-edge AI techniques. The goal? Less food waste and discounts of up to 70%.
Goals
Minimize food waste
Keep availability the same or better for fresh, affordable produce.
Save costs for the consumer by providing discounts of up to 70%.
Approach
Kickstart AI assembled a cross-industry team of AI experts from Ahold Delhaize, ING, KLM, NS, and bol.com. With a broad perspective, we went for a reinforcement learning approach, allowing AI to learn and adapt to demand signals in real-time.
This enabled dynamic discounts based on:
Time of day: larger discounts during less busy periods.
Product type: varying discounts based on perishability and popularity.
Shelf life: increasing when products approach their expiration date.
Inventory levels: to prevent overstocking and spoilage.
Key features
Enriched product and store data
We leveraged weather and demographic data to extract deeper insights, engineered advanced features, and developed a reinforcement learning (RL) model trained within our custom-built sales simulator.
Dynamic markdown model
Reinforcement learning enabled the AI to experiment with various discounting strategies. We refined the reward function to balance food waste and costs, and created an RL model capable of outperforming the existing rule-based approach.
Data-driven insights
By analyzing customer demand, shopping patterns, and historical sales, the team could model and predict when certain products would likely be purchased or left unsold.
Continuous optimization
The model continuously learned from its results, refining future discount recommendations to improve performance.
Reducing waste while optimizing revenue
This demonstrates the power of AI for a more sustainable and efficient supermarket. By leveraging reinforcement learning for dynamic pricing, we have developed a solution that effectively reduces food waste while optimizing revenue and customer satisfaction. This successful pilot paves the way for broader use and new innovations in reducing food waste.
"The experience with Kickstart AI was fantastic! I had tons of interaction with the founding companies (NS, KLM, AH, bol and ING) and the project was super well organized by Albert Heijn. It was a well-rounded experience where I felt I had an impact. Thank you for the organization and work!"
Alejandro (Challenge participant, ING)