From CMDB Chaos to Service Assurance

Almost every organisation has a CMDB. Far fewer trust it. It exists, it's populated, and yet when an incident hits, people still ping each other on chat to find out what actually depends on what. A CMDB you don't trust is just an expensive spreadsheet. Here's the playbook I use to turn one around.

Start with the questions, not the data

The classic mistake is to import everything and hope value emerges. It doesn't. Begin with the decisions the CMDB needs to support — impact analysis for changes, service mapping for major incidents, audit readiness for compliance — and work backwards to the data those decisions require. Scope follows purpose, not the other way around.

Measure data quality like you mean it

Completeness, correctness, and freshness are not vibes — they're metrics. Build KPI dashboards that show the percentage of CIs with required attributes, the age of last discovery, and the coverage of critical relationships. When data quality is visible on a dashboard senior management actually looks at, it stops being an afterthought and starts being a commitment.

Model relationships, because incidents live there

A list of configuration items is inventory. A graph of relationships is service assurance. The value shows up the moment you can answer 'if this database degrades, which business services feel it?' Validate that model in workshops with the people who own the services — they will tell you, quickly, where your data is lying.

Govern it so it stays true

Cleaning a CMDB once is easy to underestimate and easy to waste. Without configuration management practices — clear ownership, audit workflows, lifecycle tracking from onboarding to retirement — entropy wins again within a quarter. The goal isn't a clean CMDB; it's a CMDB that stays clean under change.

Do this work and something else becomes possible: every AI feature you layer on top — impact prediction, automated triage, Now Assist — inherits a foundation it can trust. Service assurance isn't the reward for finishing the data work. It's what the data work is for.