BMW accelerates test-fleet data analysis twelvefold with multi-agentic AI in Azure

Rapid technical innovation, evolving customer demands, supply chain disruption, and the rise of new manufacturers are challenging many of the world’s largest and most successful automotive OEMs. To retain and grow market share, they must find innovative ways to adapt. 

Every BMW that rolls off the production line reflects millions of test kilometers—real and virtual. But this precision comes with a challenge: data overload. Every single development vehicle in a fleet of several thousand now generates 5 to 10 terabytes of measurement data every week—braking patterns, torque, RPMs, battery voltage, and thousands more. The insights were always there, but unlocking them from that incredible data lake could take hours or days, and only IT specialists had the skills to query it. 

This created bottlenecks and put pressure on development timelines. In an industry where safety, reliability, and uncompromising quality are paramount, delays in accessing critical data can affect the precision and consistency that customers expect from every BMW. 

From modernization to momentum 

Building on BMW’s modernization of its Mobile Data Recorder (MDR) system, the next phase deepens interoperability with Microsoft Azure, creating a unified, intelligent core that powers BMW’s “Neue Klasse” development. This modernization initiative, fueled by Azure Kubernetes Service (AKS), Azure Data Explorer, and KQL integration, established a standardized environment that reduced silos and gave engineers consistent access to telemetry in real time. Access was solved—but access alone doesn’t drive innovation. 

With the foundational modernization complete, BMW shifted its focus to adoption. Although modernization successfully solved access challenges, direct interaction with MDR data remained limited—primarily accessible only to domain experts. Early AI agents marked a turning point, enabling more developers to engage with simple queries. However, their capabilities were narrow, lacking the flexibility to move fluidly between querying telemetry, generating documentation, and correlating anomalies across systems.

“We’d solved access with modernization, but access alone doesn’t drive innovation,” says Christof Gebhart, Manager of Advanced Vehicle Measurement Technology at BMW. “With multi-agent AI, engineers don’t just get data—they get insights they can act on immediately. Ultimately, the steps of data extraction and pattern recognition can be performed directly in a single step, and in natural language.”

The new MDR Copilot is beginning to change how engineers interact with development data, transforming static telemetry into dynamic insights that support faster iteration and more informed design choices.

This approach builds on lessons from earlier agents but exceeds them dramatically. Agents don’t just answer questions; they collaborate to test hypotheses, correlate telemetry with context, and generate structured reports. 

How it works in practice

When an engineer asks a question—”How many braking maneuvers were performed by the development vehicles in the last two days?”— the system responds within minutes, complete with charts and written explanations. 

Behind the scenes in Microsoft Foundry, Foundry Agent Service creates and manages specialized agents that work together through Microsoft Agent Framework. Running on AKS, these agents scale seamlessly to handle complex workloads across BMW’s engineering environment. 

Foundry Agent Service links reasoning and retrieval across Azure services, including Azure AI Search for knowledge retrieval and Azure OpenAI in Foundry Models for reasoning. Meanwhile, Azure Cosmos DB maintains persistent conversational memory, helping ensure continuity across sessions and delivering an experience where engineers can interact with test data as naturally as with a colleague.

This design doesn’t just deliver speed—it delivers continuity. Engineers can move seamlessly from one question to the next, making the process feel less like querying a database and more like having a conversation with their test data. In future iterations, Azure Database for PostgreSQL will enable a built-in feedback loop, allowing engineers to rate responses and refine results through Azure AI Search, continuously improving system accuracy.

Sebastian Heinz, Technical Lead, Advanced Vehicle Measurement Technology at BMW, speaks to the rapid and ongoing value BMW gains from its AI investment. “Azure services form the backbone of our multi-agent system, ensuring every engineer benefits from a persistent conversational history—results that remain intact through scaling, restarts, or updates. That level of reliability gives us the confidence to bring AI to hundreds of daily users and truly democratize innovation,” Heinz explains.

Proof at scale: Faster, broader, smarter

“The vision of an AI-powered agent for engineers is this: in a live dialogue, the engineer interacts with the multi-agent AI to explore signals, test hypotheses, and validate results in real time,” says Andre Thomaser, Engineer, Brake Control Function at BMW. “The multi-agent AI automatically connects to multiple specialized agents to acquire the necessary measurement data from the MDRs in the development cars, pre-analyzes it, and visualizes the results. By handling big data, it improves quality and reduces work time.”

With the new system running at scale, BMW’s R&D teams can now analyze thousands of test scenarios in parallel and retrieve insights in seconds—without relying on specialized data experts. Foundry Agent Service and Microsoft Agent Framework help ensure that every query, whether about brake dynamics or powertrain efficiency, returns consistent, explainable results backed by Azure OpenAI. What began as a single proof of concept has become a repeatable blueprint for data-driven engineering across BMW’s global development programs.

The following diagram illustrates the Azure-based architecture of BMW’s multi-agentic AI system for MDR Copilot. AI agents interpret BMW engineer queries in natural language, instantly retrieve telemetry and documentation, and maintain context across conversations.  

While the multi-agent system is still in its early rollout phase, BMW has already demonstrated its potential to dramatically accelerate engineering analysis. As adoption expands, the team expects to see measurable gains in both speed and efficiency across R&D.

Early internal results suggest: 

• Up to 12× faster analysis potential. Insights delivered before the car is back in the garage. 

• Broader access. The system is set to transform a process once limited to a few experts into a truly democratized data environment. 

• Embedded workflows. Multi-agentic systems aren’t experimental anymore—they’re on the path to daily engineering practice. 

• Ensured compliance. Governed Azure services mean IT doesn’t trade speed for security.

The road ahead: From test cars to fleet intelligence

Looking ahead, BMW plans to deepen its use of Azure AI Search and Microsoft Agent Framework to extend multi-agent orchestration beyond test fleets into powertrain, software-integration, and validation programs—further scaling AI innovation across its engineering landscape.

As BMW’s vision continues to evolve, the same AI-driven collaboration between systems and engineers that currently supports test fleets could, in future stages, be extended to other areas of vehicle development. While the integration into production vehicles and customer-facing systems remains a longer-term perspective, the foundation being built today is paving the way for such possibilities.

This is not an abstract concept—it’s a natural evolution of BMW’s foundation on Azure. By uniting modern apps, telemetry, and enterprise AI in a highly secure and scalable environment, BMW is transforming data into a living asset: one that learns, adapts, and drives innovation across every car it builds.

Discover more about BMW on Facebook, Instagram, LinkedIn, X/Twitter, and YouTube.