Rules-Governed Retrieval-Augmented Generation for Enterprise Quality Assurance: Production Deployment and Measured Outcomes in Automated SAP Test-Case Mapping

Abstract

Manual Object-to-Test-Case Mapping (OTM) in large-scale SAP enterprise change-management environments imposes a recurring cost equivalent to one to two analyst positions per year, with quality outcomes that vary by individual rather than converging on a repeatable, auditable standard. Three structural failure modes-knowledge loss, record incompleteness, and systematic regression-scope gaps-are endemic to manual OTM production and resistant to training-based remediation.
Objective: This article presents AIMO (AI Impact Mapping for OTM Objects), a rules-governed Retrieval-Augmented Generation (RAG) system designed, built, and deployed in production within a Quality Assurance Centre of Excellence to automate OTM production while satisfying the governance and audit-traceability requirements of compliance-governed SAP change management.
System: AIMO pairs an eight-step RAG pipeline with eight deterministic organizational business rules that take precedence over model output at every processing stage, converting retrieval-grounded generation from an efficiency intervention into a compliance instrument. The system has been deployed in production, processing live Change Requests across multiple SAP modules including SD, MM, FICO, TMS, and SCM.

Results: Production deployment measurements demonstrate an 84.6% reduction in analyst effort per Change Request (from approximately 156 to 24 analyst-minutes), greater than 90% historical test-case reuse from verifiable organizational records, 100% regression-scope coverage of applicable objects on every processed Change Request, and an external API cost below USD $0.03 per Change Request. Across the full deployment period, AIMO has delivered documented cost savings exceeding USD $1.38 million annually in recovered analyst capacity and defect-leakage prevention, with an operational cost of approximately USD $24,300 per year-a return of USD 56 per dollar invested.
Conclusions: AIMO demonstrates that a deterministic, version-controlled rules layer-not model capability-is the decisive architectural element for compliance-governed enterprise AI deployment. The governance properties are model-agnostic, independently auditable, and measurable as quality outcomes. The production deployment provides replicable evidence that the rules-governed RAG pattern is instantiable in live regulated enterprise environments at commercially viable cost.