One assistant, multiple channels: how Carrefour automated customer service at scale
45%
ogólny wskaźnik automatyzacji we wszystkich kanałach
84%
of chat conversations handled entirely by the bot
43%
of inbound calls resolved directly by the virtual assistant
Context
Large-scale retail has its own logic. Operations spanning hundreds of stores and millions of transactions generate a proportional volume of customer contacts — and those contacts tend to cluster around the same recurring questions: where is my order, what about this promotion, where is the nearest store. For customer service teams, this means a reality where routine consumes the bulk of available resources, and consultants spend their time on issues that — with the right tool in place — could effectively handle themselves.
This is not purely an operational challenge. Customers of large retail chains are increasingly unwilling to wait in a phone queue, and increasingly expect an answer immediately — at any hour, through whichever channel they happen to reach for. It is this collision between rising expectations and finite resources that has made customer service automation a baseline requirement for smooth operations, not just a competitive advantage.
Client
Carrefour Polska is a nationwide retail network operating over 700 stores across five formats: hypermarkets, supermarkets, neighbourhood stores, specialist outlets, and an online shop. The company also owns a network of 20 shopping centres and 40 petrol stations. It is part of the Carrefour Group — one of the world’s leading food retailers, present in nearly 40 countries and employing over 300,000 people.
The partnership with KODA began in 2024. Karol — Carrefour’s virtual assistant — went live in March and April 2025.
Challenge
Carrefour’s Customer Service department was handling requests coming in simultaneously across multiple channels: phone, email, and contact forms. Consultants’ day-to-day work was dominated by routine, repetitive queries — order status, product availability, store locations, promotion details. This generated significant operational load and tied up human resources in precisely the areas where human judgement and empathy were not what was needed.
The starting point for change was a precise audit: mapping the most frequently recurring contact types and identifying which ones could be handled automatically without any loss of quality. A key requirement from the outset was that the solution had to be a single, coherent one — working equally well in both text and voice.
As the client describes it:
The key driver behind our search for new solutions was the need for operational optimisation. We wanted to automate repetitive processes in a way that would, on one hand, relieve our consultants of the burden, and on the other, provide customers with immediate, round-the-clock support — without compromising the quality of communication.
Worth noting: KODA was not on Carrefour’s original shortlist of vendors. The team came into the process as a partner of one of the initial bidders. The approach to the project, the team’s expertise, and the vision for automation they brought to the table were enough to convince Carrefour to choose them — a decision they now describe as exactly the right call,
Solution
The project delivered a single assistant operating across two channels: a chatbot on the website and a voicebot on the customer service line. Karol — the name Carrefour gave their virtual assistant — handles customer contacts regardless of whether they prefer to type or talk, drawing on a shared logic and knowledge base throughout.
Rather than building separate tools for separate touchpoints, the result was one solution that scales across channels simultaneously. In retail, this matters: a customer might start looking for an answer on the website and end up calling the helpline, and they expect the same quality in both places.
Karol handles both structured process flows and open-ended customer questions. Users can check order status, verify product availability at a specific store, or find their nearest location. At the same time, thanks to integration with large language models, the assistant generates its own responses to frequently asked questions about promotions, terms and conditions, and Carrefour’s services.
The integrations with Carrefour’s internal systems are what make the more complex scenarios possible — giving Karol access to real-time data when it matters. Where a conversation genuinely requires a human, the assistant transfers it to a consultant — in both chat and voice — ensuring a smooth handover without losing the context of the exchange.
Key functionalities:
- smooth handover to a human consultant with full conversation context retained
- simultaneous customer support across two channels: website chat and inbound voicebot
- AI-powered automatic responses to repetitive questions, driven by large language models
- structured process flows: order status, product availability, store locations
- integration with Carrefour’s internal systems
Implementation
The project kicked off in September 2024 with an intensive analytical phase: workshops with the Carrefour team, mapping the most common customer contact scenarios, and establishing the data sources the assistant would draw on. Project documentation and the first conversation flows were developed in parallel.
In October 2024, the technical infrastructure was set up on Google Cloud Platform and integration work began — including with Salesforce. This stage also saw the configuration of Karol’s tone of voice: the way the assistant was to communicate with customers.
Karol reached Carrefour’s customers in March and April 2025. Going live did not mark the end of active development, however. In the weeks that followed, the logistics integration architecture was overhauled — moving from BigQuery to a direct connection with the ShopSpace system. This allowed Karol to provide customers with more precise delivery information: estimated arrival time, any delays, and a link to track their shipment.
From kick-off to final optimisations, the entire project came in at eight months.
Results
Karol reached Carrefour’s customers in March and April 2025. The results came quickly, and they speak for themselves.
The overall automation rate across all channels reached 45%. That means nearly every other request the company receives is now handled by the virtual assistant, with no consultant involvement required. The number of cases that genuinely need a human has dropped by 24%.
In the chat channel, the numbers are even more striking. The chatbot achieved an automation rate of 84% — the vast majority of conversations on the website reach a resolution without any escalation to a person. On the helpline, the voicebot takes 43% of inbound calls from start to finish, entirely on its own.
In practice, this translates into meaningful relief for the customer service team, which can now direct its attention where it is actually needed: complex cases that call for empathy, judgement, and human oversight. For customers, it means access to support around the clock — no queues, no waiting.
Looking ahead
Karol’s deployment was not a project with an end date — it was the beginning of an ongoing collaboration. The assistant is continuously developed, both in terms of its knowledge base and the range of scenarios and functionalities it can handle. Questions and situations it cannot yet resolve feed directly into the next iteration: every unhandled case is a signal pointing to where the system is worth expanding.
The experience of the first deployment has since opened the door to a broader scope of work. Together with Carrefour, further areas are being explored where automation can deliver tangible benefits — both in external customer-facing interactions and in internal processes.
From the outset, our goal was to build a solution that would genuinely meet the needs of Carrefour’s customers while relieving our team of the most repetitive contacts. Together with KODA and the project team, we mapped the most common service scenarios, designed the processes that could be automated, and then continued to develop them based on real data and customer experience. Regular performance reviews, ongoing optimisation of existing functionalities, and a gradual expansion of the cases Karol could handle were all central to how the collaboration worked. This is what allowed the solution to evolve alongside the business and the expectations of our customers.
Our perspective
Anna Mirecka, Business Partnerships Manager at KODA: The Carrefour partnership is one of those projects built on equal parts trust and commitment. It is regular optimisations, fast responses to issues, and above all a genuinely collaborative approach on both sides. In our recurring meetings we review results together, share recommendations drawn from experience, and decide on next steps — whether that means rolling out new features, new conversation flows, or migrating to the next generation of language models.
The Carrefour team sets ambitious challenges, and we answer them with concrete solutions. The result is real, measurable relief for the contact centre: a tool that grows and evolves alongside the business — and more projects ahead of us together.