Skip to main content
BP logo

CUSTOMER
STORY

bp turns weeks of maintenance analysis into 30-second answers with Genie Agents

Accelerating maintenance analysis with AI-ready data

bp operates across offshore production, refining, terminals and pipelines, and every business depends on safe, available equipment. But maintenance data alone spans about 60 source systems, including 47 SAP instances and Maximo, across upstream, midstream, downstream and logistics. 

Each system spoke its own language. There were 11 different ways to spell Gulf of Mexico. Maintenance engineers relied on data engineers and analysts to get answers, and analysts became the bottleneck. Gathering requirements could take weeks, followed by rounds of back and forth, so finding out whether equipment might fail in the next 30 days took weeks.

Creating an AI-driven maintenance foundation for Genie Agents

bp brought its systems of record together on Databricks, atop its cloud storage and governed by Unity Catalog. In the silver layer, the team replaced SAP and Maximo terms with a shared data model using clear names like equipment, work order and notification, with proper primary and foreign keys. The gold layer is modeled with domain experts in the language they use. Before Genie goes on top, data runs through an AI data-readiness scoring model that assesses five areas: data quality, standardization, metadata and lineage, usability and governance. A score of four or five means the data is ready for AI.

Genie Agents adoption

With AI-ready data in place, setting up a Genie Agent takes about 10 minutes. bp’s maintenance Genie Agent is configured with instructions on the engineering standards behind the data model, definitions like mean time between failure, table relationships and rules such as always summing cost from the cost line table. Questions about potential equipment failure that used to take weeks can now be answered in about 30 seconds. bp rolled out an MVP to 400 business users, received good feedback and requests for more and is now ready to expand to 9,000 people. Genie reaches users wherever the business needs it, including Teams, API calls to bp’s own platform and AI and BI solutions.

Scaling Genie Agents with metadata and governance

In a test, bp asked Genie the same question against two identical sets of tables, one with proper keys and column definitions and one without. Genie answered four times faster on the well-modeled data because it didn’t have to scan as much to find the answer. Metadata like data owner, classification, frequency and retention flows automatically into Unity Catalog, where the readiness scoring model reads it. The readiness assessment certifies gold data products for sharing, so the same maintenance data can also serve digital twins, production management and risk management. The team also uses Genie Code to profile existing data and generate a starting silver data model, which gets about 75% of the way there for industry-standard systems before subject matter experts refine it.

FAQ: bp on Databricks Genie

bp built a maintenance Genie Agent on AI-ready data in Unity Catalog. Engineers can ask questions such as whether equipment might fail in the next 30 days and get answers in about 30 seconds, instead of weeks.