Back office ops · Production

Co-op builds a RAG virtual assistant on Databricks to streamline 50,000–60,000 weekly in-store policy queries

The problem

Co-op store colleagues had to navigate over 1,000 policy and procedure documents under time pressure using a traditional keyword search that was slow, required precise terms, and drove 50,000–60,000 weekly queries to support centers, raising operational costs and reducing efficiency.

Workflow diagram · grounded in source
1
Colleague submits a query
trigger
“the "How Do I?" project, to create a more efficient system for accessing information in-store”
2
Daily document ingestion
integration
“the team used Databricks Lakeflow Jobs 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”
3
Semantic document retrieval
ai_action
“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”
4
LLM response generation
ai_action
“OpenAI's Chat GPT-3.5 was selected as it provided the best balance of performance, speed, cost and security”
5
Answer delivered to colleague
output
“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”
Reported outcome

Initial internal test feedback is overwhelmingly positive—employees found the application intuitive and much quicker than the previous system.
Co-op plans a store trial with potential for full-scale deployment.

Reported metrics
Weekly queries to manage50,000 to 60,000
Weekly initial queries23,000
Weekly follow-up queries35,000
Policy documents in knowledge baseover 1,000
Show all 6 reported metrics
weekly queries to manage50,000 to 60,000
weekly initial queries23,000
weekly follow-up queries35,000
policy documents in knowledge baseover 1,000
initial test feedback qualityoverwhelmingly positive
information retrieval speedmuch quicker to retrieve the necessary information
Reported stack
Databricks Data Intelligence PlatformDatabricks Lakeflow JobsDatabricks Vector SearchMLflowDBRXMistralOpenAI's Chat GPT-3.5Databricks Model ServingDatabricks AssistantAgent Bricks Custom AgentsContentful
Source
https://www.databricks.com/customers/co-op
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Initial internal test feedback is overwhelmingly positive—employees found the application intuitive and much quicker than the previous system.

What tools did this team use?

Databricks Data Intelligence Platform, Databricks Lakeflow Jobs, Databricks Vector Search, MLflow, DBRX, Mistral, OpenAI's Chat GPT-3.5, Databricks Model Serving, Databricks Assistant, Agent Bricks Custom Agents.

What results were reported?

Weekly queries to manage: 50,000 to 60,000; Weekly initial queries: 23,000; Weekly follow-up queries: 35,000; Policy documents in knowledge base: over 1,000 (source-reported, not independently verified).

How is this back office ops AI workflow structured?

Colleague submits a query → Daily document ingestion → Semantic document retrieval → LLM response generation → Answer delivered to colleague.

Related back office ops cases