Requirements Processing Optimized at BMW

Initial Situation

The development of modern driver assistance systems at BMW requires processing thousands of requirements from diverse sources. For the Emergency Steering Support System, engineers had to consolidate specifications from internal requirement documents, UN regulations, ISO standards, and country-specific homologation requirements. The manual analysis of these documents consumed significant capacity from the Requirements Engineering team and was error-prone.

Consistency checking between different document versions proved particularly challenging. With every regulatory update, experts had to manually verify which system requirements were affected. This process often took several weeks and still occasionally resulted in overlooked dependencies that only surfaced in later development phases.

KIMBA Solution

As part of the KIMBA project, an AI-powered system for automated requirements analysis was developed and piloted at BMW. The system is based on a language model specifically trained for technical documentation that can process structurally complex documents with cross-references.

The solution encompasses several core functions: A semantic search function enables engineers to ask natural language questions against the entire document base. The automated quality check validates new requirements against formal criteria and identifies incompleteness. The comparison function detects contradictions and redundancies between different specifications.

Results

The pilot demonstrated significant improvements in the development process. The time required for initial requirements capture was reduced by 65 percent. The system identified 27 potential inconsistencies within hours that would likely have been missed during manual review.

Early error detection prevents costly rework in later project phases. BMW estimates the savings potential from avoided design adaptations at a seven-figure amount per vehicle project. Documentation quality also improved measurably, as the system automatically flags missing links and incomplete traceability.

The success of the pilot led to the decision to extend the system to additional Customer Functions. By 2026, the entire BMW system model is expected to be accessible through the KIMBA solution.