Key Takeaways
- Albert Heijn, the Netherlands’ largest supermarket chain, operates approximately 1,200 stores across the Netherlands and Belgium.
- Its Rasa-powered AI agent reduces 50% of customer service contacts.
- After moving to Rasa CALM in 2025, customer satisfaction increased 0.5 points on a 5-point scale.
Albert Heijn, the Netherlands’ grocery market leader and a subsidiary of Ahold Delhaize, set out to improve customer satisfaction while reducing pressure on human support teams. The company needed a digital agent that could handle complex customer requests in Dutch, operate securely at scale, and integrate seamlessly into existing systems without compromising control over data or operations.
To support this objective, Albert Heijn selected Rasa as its agent platform. This enabled more flexible and reliable handling of complex, real-world customer interactions, thereby reducing customer service wait times.
Challenges
An ongoing collaboration designed to scale
Albert Heijn began working with Rasa in 2019. The initial deployment focused on customer service, with 20 - 25 structured conversation flows covering the most common questions. This first implementation immediately reduced contact volume by resolving a meaningful share of customer inquiries without involving human agents.
The collaboration expanded to WhatsApp, enabling 24/7 support across channels. In 2024, Albert Heijn migrated to Rasa CALM (Conversational AI with Language Models) to support more complex, multi-turn interactions with greater predictability and control.
“We needed a solution that could run on-premises, support Dutch language complexity, and scale with our ambitions. Our experience with cloud-only solutions showed us how critical control and reliability are for our customers,” Stijn Verhoeven, Product Owner Self-Service and Generative AI, Albert Heijn.
Albert Heijn measures success using two core metrics introduced at launch in 2019:
- Prevented contact rate, which tracks how often the agent resolves an issue without human escalation
- Quality score, based on customers confirming their question was answered or continuing their journey via agent-provided links
Today, the agent achieves a 50% prevented contact rate and a 42% quality score across operations in the Netherlands and Belgium, delivering substantial operational cost savings.
The CALM-based agent runs across web and mobile channels and supports authenticated, personalized experiences. Albert Heijn also transitioned from button-heavy interfaces to free-text conversations, allowing customers to switch topics naturally while the agent maintains context using CALM’s built-in conversation patterns.
The Results
High-impact use cases in production
The AI agent delivers significant value in high-frequency scenarios, such as subscription cancellations and order changes made before cutoff times.
One notable improvement involves digital stamp campaigns, where customers collect loyalty stamps at checkout and redeem rewards through the Albert Heijn app. Previously, missing-stamp issues generated high support volumes. With CALM, the AI agent checks the customer's eligibility and automatically adds missing stamps to their accounts, eliminating manual effort from customer care teams and enhancing the overall customer experience.
Handling digital stamp issues through the app led to an 80% reduction in contact volume for that specific request.
Albert Heijn currently maintains around 30 deterministic flows for common questions and actions. For long-tail inquiries, the company uses a RAG-based (Retrieval-Augmented Generation) setup that applies Large Language Models to generate responses grounded in existing FAQ content published across its website and app.
Orchestrated
Eliminated
Grounded
Looking ahead
Albert Heijn plans to expand its AI capabilities across several areas:
- Voice channel support, enabling customers to interact with the agent by phone
- Additional languages, including English, French, German, Polish, and Turkish, to better serve diverse audiences
- Deeper personalization, using authenticated customer data to deliver more contextual support
- Dedicated flows for business customers, with clearer separation between B2C and B2B needs
- In the U.S., Ahold Delhaize is consolidating six existing assistants into a single multi-brand agent powered by the Rasa Platform
By continuing to deploy Rasa on-premises, Albert Heijn retains full control over data, logic, and execution while steadily expanding automation across channels and use cases. The result is a production- grade AI agent that scales with business complexity, delivering measurable gains in customer satisfaction and operational efficiency.
