EDF UK taps Snowflake for AI market forecasts, customer service

EDF UK centralized its enterprise data on Snowflake’s AI Data Cloud to run near-real-time energy market forecasts and to deploy AI agents in Slack that answer customer billing and smart meter queries.

The platform serves more than 1,000 users across retail, business, wholesale markets, finance and Zero Carbon Homes. It operates alongside dbt, Matillion, Amazon Bedrock and Amazon SageMaker. EDF uses dynamic tables and event-driven architectures to ingest hundreds of data sources onto Snowflake.

Alex Read, senior enterprise product manager for data at EDF UK, leads the team responsible for tooling, architecture and common services while federated business units run product teams that build data and machine-learning products.

In wholesale markets, EDF is rebuilding a volume forecasting platform used for purchasing and hedging in an energy market the company values at more than $13.53 billion. The platform combines industry flows, weather data and internal consumption patterns to train AI models that forecast demand. Small improvements in forecast accuracy can affect financial results when applied to large volumes.

Regulatory changes from Ofgem’s Market-Wide Half-Hourly Settlement prompted more granular pricing and billing. FreePhase uses three pricing bands to reward customers who shift usage to lower-cost periods. The Sunday Saver Challenge uses smart meters to give customers up to 16 hours of free electricity on Sundays when weekday usage is moved away from peak times. These products require fine-grained consumption data and near-real-time processing.

Read’s team is building an AI agent that will sit in staff Slack channels and answer customer queries about programs such as the Sunday Saver Challenge. The agent pulls information from Snowflake and is being developed on Snowflake Cortex; the project is nearing production. Read said, “A customer query will be automated against Snowflake. They’ll be talking to the data on the platform to answer questions.”

EDF is cataloging and defining data assets so AI can use them. The company is using Snowflake features such as Semantic Tables and Horizon Data Catalogs and adding Cortex Analyst and Cortex Code to make insights and code available to end users.

Governance follows a hub-and-spoke model: central IT provides common services and enablement while federated teams of data engineers, data scientists and MLOps engineers build products close to business needs. Read noted Snowflake’s integrations reduce the number of vendors required to bring products into production.

Read expects Snowflake to support more near-real-time use cases with quality-assured, cataloged data feeding AI models across the business. Read added, “People of all skill sets and personas will have AI tools in their hands, fed by Snowflake.”