How energy teams turn theft detection into governed action with Genie and AI business processes

Energy theft is the deliberate use of gas or electricity without paying for it, typically by tampering with a meter or supply so consumption goes unrecorded. Unlike a billing error or an unpaid bill, it involves physically modifying the connection, making it difficult to detect and potentially dangerous.

For revenue protection teams, theft is both a financial and a safety issue. Tampered meters and wiring can cause fires and gas leaks. Energy theft costs energy consumers in Great Britain alone over £1.4 billion a year, with only 40% of target cases being detected.

The challenge is no longer simply detecting more theft, as most companies already have ML models that can flag suspicious accounts. Teams need to unify revenue protection operations around a single governed workflow that accelerates the loop from flagged meter to recovered revenue to a safer household. Teams need to be able to interpret signals, prioritize investigations, prepare field teams, recover losses, and give leaders trusted metrics in a timely manner.

An ML insight is not yet a business outcome

Most organizations can train models that generate useful predictions, such as a churn probability, fraud score, failure forecast, or theft risk score. But the score itself is not the outcome. Value comes from what the business does next, and that is where many ML programs stall. The insight lands in a dashboard, while someone still has to translate it into action.

Energy theft is a powerful example: leaving an insight unactioned costs both money and safety. A flagged account does not resolve on its own. Someone must assess whether it warrants investigation, dispatch a field visit safely, recover the loss, record the outcome, and feed the results back into the model. When manual handoffs connect these steps, the process slows where speed matters. Every day a genuine case waits means more stolen energy and a prolonged safety risk.

The real challenge, then, is not detection but operationalization: connecting detection, investigation, recovery, and reporting into a single governed loop. On Databricks, that loop runs in a Databricks App, with live case state held in Lakebase and business logic running in the same lakehouse as the underlying data, governed end to end by Unity Catalog. The insight becomes action without leaving the platform.

Watch Daniel Zoccali, an Energy Solutions Architect at Databricks, demonstrate how Databricks can help detect and operationalize energy theft investigations.

From flagged account to investigation-ready action

Operationalizing an insight starts with interpretation. The analyst triaging a flagged account is not a data scientist, and even an accurate composite risk score may not explain what to do next. A bare number can send crews toward false positives and consume scarce field capacity. That is why the workflow pairs the score with an AI-generated case summary, served through Databricks Model Serving and governed by Unity Gateway, that explains in plain language why the account was flagged. A lengthy manual review becomes a quick, consistent assessment.

The analyst can then move from interpretation to execution. Instead of spending days gathering evidence and preparing compliance details, they generate a dispatch-ready report grounded in lakehouse knowledge. It includes recommended next steps, an evidence checklist, and safety notes so the engineer arrives prepared. Because the workflow runs as a Databricks App, it carries the case beyond the document and into the enterprise systems the team already uses.

Lakebase keeps the process responsive in real time. As the platform’s Postgres transactional layer, it stores scored outputs and live case state, allowing the app to read and write updates with low latency. When a recovery is confirmed, the running total updates immediately. And because the model runs through Unity Gateway, teams can choose or change the underlying model through configuration rather than re-engineering the workflow.

Trusted answers matter just as much as fast actions

Operational speed also needs executive visibility. Revenue-protection leaders need to track recovery against their targets and answer CFO questions without waiting for a report.

Genie One lets leaders ask questions in plain English and receive answers grounded in metric definitions in Unity Catalog. That means measures such “revenue recovered” and “precision rate” have a consistent meaning, while teams can inspect how each number was produced. The same governed semantics allow an Agent Bricks Multi-Agent Supervisor to automate the month-end board pack by orchestrating Genie queries and returning a traceable, board-ready output. Analysts can spend their time improving models rather than assembling slides.

Governance is at the heart of the platform, not an afterthought

Governance is not optional when investigating energy theft. In the United Kingdom, Ofgem’s Data Access and Privacy Framework and ELEXON’s Balancing and Settlement Code govern how energy data is accessed and used. With Ofgem actively investigating suppliers’ compliance, governance is not a theoretical concern, it is an operational and regulatory requirement.

When models, serving endpoints, and datasets are registered in Unity Catalog with fields labeled as PII, teams can trace data lineage through to the models that consume it, audit usage, and apply model-level guardrails and access controls. Routing calls through Unity Gateway helps keep data, PII, and prompts within the organization’s security perimeter while providing visibility into AI costs for each investigation. Trusted, explainable, access-controlled answers are what make AI adoption possible in a regulated operation.

The same pattern operationalizes any ML insight

This pattern extends well beyond energy theft. Under the hood, it is a repeatable way to turn a model output into action: Lakeflow transforms raw data into features, models are trained and served on the platform, and outputs are written to Lakebase for an operational app. Genie One and Databricks SQL answer leadership questions, while agents automate reporting. All within a governed Databricks App running on serverless compute.

Change the model and the signals, and the same closed loop can support insurance claim fraud, predictive maintenance, payments fraud, or churn intervention. The opportunity is not simply a better score; it is acting on that score faster, within one governed workflow where detection, investigation, analytics, and reporting reinforce one another. The value of AI lies not only in the model, but in the business process built around it.

Ready to learn more? Watch the full demo and expert deep-dive from our Energy Virtual Industry Forum: Power the energy transition with unified data, real-time insights, and AI at scale.

Frequently asked questions

What is energy theft?

Energy theft is the deliberate use of gas or electricity without paying, often through meter tampering or illegal connections. It is distinct from a billing error or unpaid bill and can create serious safety risks.

How can Databricks help detect energy theft?

Databricks connects ML scoring to a governed Databricks App for investigation and recovery. The workflow can generate case summaries, prioritize investigations, prepare dispatch-ready reports, track recoveries, and provide executive analytics. The lakehouse supplies the data foundation, while Lakebase stores live case state for responsive operational applications.

How does AI support energy theft investigations?

AI can explain flagged accounts, surface evidence, recommend next steps, and prepare field reports. Analysts and engineers remain responsible for safety, compliance, and execution.

Can leaders ask questions about recovery in natural language?

Yes. Genie One provides answers grounded in governed data and shared definitions, including metrics such as revenue recovered, investigation volumes, recovery rates, and precision.

Does AI replace revenue protection analysts or field engineers?

No. AI accelerates triage and reporting, while people make decisions requiring judgment, regulatory interpretation, customer handling, and field execution.