Scaling AI in utilities with a federated data lakehouse

Utilities manage data across generation, transmission, distribution, field operations, customer systems and enterprise functions. Yet fragmented data, platforms and ownership models can make it difficult to turn that information into trusted, reusable data for analytics and AI.

A federated data model gives business domains greater ownership of the data they understand best while maintaining shared governance, security and standards. This can accelerate access to trusted data, enable reusable data products and create the foundation to scale AI.

CGI and Databricks help utilities put this model into practice, combining federated data architecture with the governance, engineering and AI capabilities needed to scale.

Why do utilities need a federated data strategy?

Utilities operate across business domains with distinct data, operational and regulatory requirements:

  • Generation: nuclear, renewable and hydro
  • Transmission and grid operations
  • Distribution and field operations: outages, reliability, asset health and vegetation management
  • Customer and meter-to-cash: advanced metering infrastructure (AMI), meter reads, billing and payments
  • Gas operations
  • Enterprise functions: finance, workforce, procurement and ESG

Managing data centrally across these domains can create bottlenecks as priorities and specialized requirements compete.

A federated data strategy distributes ownership while maintaining enterprise control through four principles:

  • Domain ownership: Each business domain owns the data it knows best.
  • Data as a product: Data is curated, documented and made available for reuse.
  • Self-service: Domains work with their data on a shared platform with less reliance on central IT.
  • Federated governance: Security, cataloging and standards remain central while domains operate within shared guardrails.

CGI applied this approach in a large U.S. utility environment, where federated lakehouses evolved across areas including gas, power delivery, renewables, customer and nuclear. The broader data-platform program increased self-service and collaboration while reducing data duplication and improving consistency across subsidiaries.

CGI's Federated Data Strategy and Data Modernization Workshop helps utility leaders prioritize high-value data products, identify governance and security gaps and sequence modernization around business priorities.

How does Databricks support a federated utility data model?

The Databricks Data Intelligence Platform provides a shared foundation for distributed data ownership without creating new silos. Domains can own and publish data products aligned to their business needs, while Databricks Unity Catalog provides common governance across them.

Distribution teams, for example, can create reusable products around meter, outage, asset and reliability data while generation teams manage plant performance and maintenance data and customer teams manage account, usage and billing data. Unity Catalog provides shared access control, auditing, lineage, metadata and discovery across these domains.

Reusable, metadata-driven ingestion helps utilities scale without building a custom pipeline for every new source. CGI uses metadata-driven frameworks to standardize ingestion and data-quality checks, reducing custom development and ongoing maintenance.

The result is a faster path from raw utility data to trusted, reusable information for analytics and AI.

Modernization can also be sequenced around business value, without requiring the entire data estate to move before benefits are realized.

CGI's Databricks Lakehouse Architecture and Migration Assessment helps utilities define their Unity Catalog governance model, data-product architecture, metadata-driven ingestion approach, quality controls and migration waves based on business value and risk.

How does governed data help utilities scale AI?

Governed, reusable data products give AI applications trusted business context. For utilities, that can support use cases spanning load forecasting, asset health, outage and reliability management, generation optimization, field operations and customer service.

AMI demonstrates the connection.

CGI developed and demonstrated an AI Meter Hub pilot solution using Azure and Databricks to show how unified meter data can overcome fragmented visibility and slow operational decision-making. The solution brought meter data together to support analysis, anomaly detection and operational forecasting.

In the pilot, the solution:

  • Detected, mapped and attributed more than 1,200 brownouts to specific meters
  • Improved visibility into meter health, transformer loads and power quality
  • Used anomaly detection to identify voltage dips and consumption spikes to support more proactive grid operations
  • Enabled field operations teams and analysts to ask questions about meter data using natural language
  • Applied predictive models to consumption forecasting and anomaly detection

Moving AI from pilots to production

Moving AI into production requires repeatable lifecycle controls for how models and generative AI applications are deployed, monitored and governed.

MLOps provides discipline for machine learning, including versioning, deployment, monitoring and retraining as data and operating conditions change. Generative AI adds requirements around grounding, evaluation, access, safety, auditability and cost, which LLMOps extends across applications and agents.

Databricks capabilities such as Unity Catalog and AI Gateway provide controls across data, models and AI interactions. CGI's Databricks Brickbuilder LLMOps accelerator adds reusable patterns for evaluating and governing LLM applications across the lifecycle.

For utilities, governance also needs to reflect the operating domain, jurisdiction, data and intended use. Applicable requirements can differ across bulk electric system, nuclear and gas operations. The common principle is to establish ownership, approved data, access controls and lineage from the beginning, then monitor performance, quality, usage and business outcomes in production.

How can utilities build a foundation for AI at scale?

Federate data around the business. Establish a governed foundation. Use that foundation to scale analytics and AI into operational value.

CGI's Governed AI, MLOps and LLMOps Readiness Session helps utilities assess AI use cases, determine whether their data and governance are ready for production and establish the lifecycle controls needed to scale. It can also apply CGI's Databricks Brickbuilder LLMOps accelerator to establish repeatable evaluation patterns for generative AI.