Helping employees resolve customer issues faster
Co-op uses Databricks Mosaic AI to improve colleague support
Queries managed weekly
Co-op, one of the world’s largest consumer cooperatives owned by millions of members, is the UK’s fifth largest food retailer with over 2,500 stores. Additionally, Co-op is the UK’s leading funeral services provider and a major general insurer, with a growing legal services business. In 2023, Co-op identified an opportunity to dramatically improve how colleagues access essential information for the day-to-day running of their stores. The beloved brand developed a GenAI solution, creating a retrieval augmented generation (RAG)–based virtual assistant application designed to streamline policy and process document searches. Leveraging the Databricks Data Intelligence Platform, Co-op’s teams aim to more effectively manage around 50,000 to 60,000 weekly queries, to reduce the load on their support centers and improve both colleague support and customer satisfaction.
Experimenting with GenAI for improved informational access
Co-op, a beloved UK cooperative offering everything from groceries and insurance to funeral services, is known for supporting their members and the communities they live in. The brand is committed to environmental sustainability, healthy living and inclusion — all while rallying behind their food suppliers and providing excellent service to customers. Although Co-op already leverages Databricks for data warehousing, engineering and analytics, the integration of generative AI offered huge opportunities to dramatically enhance other vital business areas such as internal knowledge sharing.
Employees working in Co-op’s UK-wide food stores have access to a comprehensive library of over 1,000 web-based guides covering all store policies and procedures. Navigating these documents using the existing traditional keyword search engine can be time-consuming and cumbersome, requiring precise search terms to find the required information.
Joe Wretham, Senior Data Scientist at Co-op, explained, “While it might not seem like a huge amount of documentation, our store colleagues have to find and navigate the information they need, often while working under pressure in our busy stores.”
Despite being accessible via mobile phones or in-store devices, the existing system’s limitations still impact employees’ ability to retrieve the information needed to run stores effectively. First, the volume of queries to the application is substantial, with 50,000 to 60,000 questions asked weekly. Even if the correct document is found, locating the specific information is time-consuming since documents are lengthy and thorough. Finally, the inefficacy of the discoverability process often leads to reliance on the company’s support centers for assistance — increasing operational costs and reducing efficiency.
Finding the best model to enhance in-store customer service
“To remove the strain on our colleagues, we thought there was an opportunity to do something with GenAI to get our colleagues the information they need quicker,” Wretham said. Co-op’s data science team, longtime users of the Databricks Data Intelligence Platform, embarked on their first GenAI venture, the “How Do I?" project, to create a more efficient system for accessing information in-store. Through this proof of concept, they built a RAG virtual assistant application powered by Mosaic AI Agent Framework, incorporating the same 1,000 documents they were using for informational purposes. Specifically, the team used Databricks Workflows to automate the daily extraction and embedding of documents from Contentful, a popular content management system, ensuring up-to-date information for Co-op teams. Because these documents were stored in Databricks Vector Search, an optimized storage solution that manages and retrieves vector embeddings using semantic recall, they could be quickly retrieved to support user queries.
To manage the machine learning lifecycle, Co-op implemented MLflow, an open source platform that facilitates model swapping and experimentation — a vital feature for fine-tuning their system. Along with Databricks’ serverless computing for scalable and efficient processing power, the brand could now handle larger volumes of data. The development process involved significant experimentation with various AI models, including DBRX and Mistral as well as OpenAI’s GPT models. An evaluation module was built within Databricks Data Intelligence Platform to measure the accuracy of different models by firing hundreds of test questions at the application in many configurations and assessing the accuracy, response times and built-in safeguarding features. Despite all models scoring well, OpenAI’s Chat GPT-3.5 was selected as it provided the best balance of performance, speed, cost and security. The additional use of Databricks Model Serving also helped simplify model deployment, ensuring seamless integration into Co-op’s existing infrastructure. Paired with Databricks Assistant, a context-aware AI assistant, the combination of all these technological components proved to be invaluable for resolving syntax queries and simple issues.
Aside from the model evaluation, the integration and development process was also streamlined, allowing Co-op to focus on innovation rather than technical complexities. Databricks’ seamless integration with external tools from OpenAI and Hugging Face and their commitment to open standards enabled quick setup and iteration of AI models. This flexibility was crucial for experimenting with different strategies, like fine-tuning prompts to optimize response accuracy and relevance, adjusting parameters to control responses and iterating on prompt phrasing to improve the model’s understanding and outputs. Reflecting on the ease of exploration, Wretham noted, “The key benefit Databricks has provided is removing barriers and simplifying processes, allowing us to easily explore and innovate. This has been true both during the exploratory phase and throughout the application development.”
During the entire process, Databricks solution architects provided essential support, helping navigate technical challenges and ensuring data security compliance. As Co-op transitions to Unity Catalog, their team will enhance data governance and access controls for more secure data handling. Overall, the collaboration with Databricks significantly improved Co-op’s ability to manage and retrieve information efficiently.
Paving the way for future successful GenAI projects
Implementing the new GenAI solution at Co-op will give team members faster and more accurate access to information, reducing the workload of their support centers and encouraging more self-service among employees. Although the project is still in the proof-of-concept stage, initial feedback from internal tests is overwhelmingly positive. Employees have found the AI-powered application intuitive and much quicker to retrieve the necessary information than the company’s previous setup.
Since the application routinely receives about 23,000 initial queries and 35,000 follow-up questions weekly, the transition to a new system will help Co-op handle this volume of queries more effectively, improving the productivity, efficiency and time management of their in-store staff. Furthermore, Co-op is considering additional GenAI projects, such as automating legal document processing and personalizing customer offers — both of which highlight the transformative potential of AI within retail operations.
Looking ahead, Co-op plans to conduct a trial of their “How Do I?” project in selected stores later this year, with the potential for full-scale deployment if proven successful. The trial is crucial for collecting user feedback and making any necessary adjustments to optimize the system’s performance before full implementation. Wretham concluded, “Databricks has served as the foundation for all our data and AI work and has removed the barriers to making workflows easier at every stage. We couldn’t be more excited to see what’s in store for the future.”