SAP HCSM — 90% event-noise reduction across a 16-billion-transaction platform

Seven SAP lines of business. Twenty-seven cloud-operations sub-units. One platform. Zero domain separation.

The challenge

SAP’s Hosted Customer Success Management organisation ran cloud operations across seven top-level Lines of Business, each with its own monitoring tooling, alert taxonomy, and incident workflow. The result was operational fragmentation at scale: 60-plus event sources generating millions of raw signals per month, manual FTE time consumed by alert triage, and service availability capped by the slowest-moving LoB. A 99.5% contractual SLA required platform-level coordination the fragmented model could not deliver. The business case for consolidation was clear; the architecture to accomplish it without domain separation — preserving a unified CMDB while respecting each LoB’s operational autonomy — was not.

The Blu Wingu approach

Blu Wingu was engaged as Principal Consultant and AIOps Architecture lead across the HCSM ITOM programme from 2019 to 2024. The engagement began with a CMDB-first foundation: a single instance ingesting approximately 22 million configuration items continuously synchronised from 25-plus source systems and 300-plus vCenters. Without an authoritative topology, correlation algorithms have nothing to correlate against.

The Two-Flavour Event Management Architecture was the signature design decision. ServiceNow native Event Management — driven by CMDB-hierarchy correlation — handled sources that mapped cleanly to the CI graph. The MoM Alerting Framework (GitOps/Python, OctoBus-native) handled sources with different data-shape requirements. Both routes shared OctoBus as the central data spine, delivering 90% noise reduction without forcing every LoB into a single integration model. The OctoBus Golden-Standard Integration Pattern codified zero-ServiceNow-customisation routing, enabling the estate to scale to 60-plus sources without proportional Centre of Excellence overhead.

The AI layer ran in parallel. Blu Wingu led delivery of AskWIPS — SAP’s first vector-RAG enterprise copilot serving 53,000 service agents via HANA Vector Search and SAP AI Core. HCSM CoPilot followed: 3,000-plus uses in four weeks, 82% accuracy, approximately nine minutes saved per use, path to 30,000 users. Approximately 56 GenAI use cases shipped; 75 in pipeline. The LLM-as-a-Judge evaluation framework was productionised in Heidelberg, November 2024. Blu Wingu directed the HCSM Architecture Board as formal approval authority for all plugin activations, Store-app installations, and integration scopes.

The 2023–2024 PagerDuty migration was the financial inflection point: 27 Cloud Operations units transitioned to ServiceNow, ML correlation reduced manual effort by 85%, and €3.2M in annual savings came from automated management of 1.3 million incidents per year.

The outcome

Over four years, the HCSM platform moved from a fragmented multi-tool estate to a unified AIOps architecture: 106,000 end users, approximately 10,000 daily active users, 28 TB instance, 16 billion transactions per year. Service availability rose from 99.2% in 2020 to 99.9% in 2023 against a 99.5% contractual SLA. SLA-impacting incidents fell from 16% to 0.2% in year two, then 0.1% in year three. The 10-times alert reduction and 51% TCO reduction were achieved without rearchitecting the underlying LoB operating models.

What’s defensible

Every metric traces to architecture artefacts authored by Blu Wingu and signed off by SAP’s Architecture Board — the €3.2M saving, the 90% noise reduction, the SLA trajectory, and the GenAI usage figures are programme-of-record outcomes.

Cross-links: CSDM v5 Service Modelling