SSE Airtricity × Databricks — Customer Service Automation

Solution

SSE Airtricity built the in-house Enhanced Smart Insights (ESI) Energy Advisor on the Databricks Data Intelligence Platform — already on Azure after migrating from Oracle, with Unity Catalog providing centralized governance for both data and AI models and Delta Lake as the storage foundation. Lakeflow Spark Declarative Pipelines ingest billions of rows of smart-meter data nightly (48 readings per meter per day across 65,000+ customers) and feed personalization models that call Anthropic Claude Sonnet 4.5 through Databricks AI Model Serving. An evaluation-first MLflow GenAI framework with custom AI judges defined alongside human feedback automatically rejects and regenerates any insights that fail on length, language, tone, or accuracy before they reach customers. Custom Databricks Apps surface real-time monitoring of cost, model performance, and judge failure rates with automated alerts. The team went from concept to production in four months.

Data flow

Billions of rows of smart-meter readings (48 per meter per day across 65,000+ customers) and customer-survey responses are ingested nightly through Lakeflow Spark Declarative Pipelines into Delta Lake on Databricks (migrated from Oracle), governed by Unity Catalog for both data and AI models. The ESI Energy Advisor calls Anthropic Claude Sonnet 4.5 through Databricks AI Model Serving to generate personalized energy-saving advice. MLflow Traces and an evaluation-first framework — AI judges defined alongside human feedback — automatically reject and regenerate insights that fail on length, language, tone, or accuracy. Databricks Apps host custom monitoring dashboards tracking cost, model performance, and judge failure rates in real time.

Solution architecture

7 components · 4 layers

Compute
- Lakeflow Spark Declarative Pipelines Ingests and transforms billions of rows of smart-meter data nightly (48 readings per meter per day) to feed the AI models
- Agent Bricks / Mosaic AI Generative AI tooling backing the ESI Energy Advisor
Storage
- Delta Lake Reliable Lakehouse storage forming the foundational data architecture for the AI strategy
Serving
- Databricks AI Model Serving Serves Anthropic Claude Sonnet 4.5 to the ESI Energy Advisor without managing infrastructure
- Databricks Apps Custom monitoring apps tracking cost, model performance, and AI-judge failure rates in real time with automated alerts
Governance
- Unity Catalog Centralized governance for both data and AI models inherited from the Oracle→Databricks migration
- MLflow Tracing, GenAI evaluation framework, and AI judges that auto-reject and regenerate failing insights for complete observability and traceability

Architecture clues

Anthropic Claude Sonnet 4.5 via Databricks AI Model Serving
Databricks Apps for real-time cost and quality monitoring
Lakeflow Spark Declarative Pipelines for nightly ingestion
MLflow GenAI evaluation framework with AI judges + human feedback
Runs on Azure
Unity Catalog governance across data and AI models

Evidence from the source

90% of logged-in customers now navigate to the insights section to engage with the recommendations

We went from proof of concept to production in about four months

Within the first month, SSE Airtricity saw a 12% increase in web visits to the insights section

Business outcome

Soft-launched in mid-December 2025, the ESI Energy Advisor reached production in four months, drove a 12% increase in unique visitors to the insights section in the first month, and now engages 90% of logged-in customers with personalized recommendations — delivering bill-day insights vs. competitors' 120-day wait and laying the foundation for tariff recommendations, bill analyzer, peer comparisons, and energy disaggregation.

SSE Airtricity

Solution

Data flow

Similar use cases