How to keep your CMDB accurate — and finally trust it
A configuration management database is only as useful as it is believed. The moment people stop trusting it, they go back to their own spreadsheets — and the CMDB becomes another stale system to maintain. Here's why that happens, and a practical way to stop it.
Why CMDBs drift
Almost every inaccurate CMDB got that way for the same handful of reasons:
- It was populated by hand. Manual entry is wrong the day it’s typed and worse every day after. People move, devices get rebuilt, software is installed and removed — and nobody updates a database in lockstep with reality.
- The truth is spread across systems. Your endpoint manager, directory, ITSM tool and procurement system each hold a partial, slightly different version of each asset. None of them is complete, and they rarely agree.
- Nothing owns each record. When a CI has no clear owner and no authoritative source, there’s no mechanism to keep it current — so it rots.
- Imports are one-off. A bulk import feels like progress, but it’s a snapshot. Within weeks it’s drifting, and there’s no process to refresh it.
What “accurate” actually means
It helps to be precise about the goal. A trustworthy CMDB has four properties:
- Coverage — it contains the things that actually exist. No ghost records for retired kit; no blind spots for devices nobody registered.
- Correctness — the attributes (owner, location, model, status, warranty) match reality.
- Currency — it reflects the world as it is now, not as it was at the last audit.
- Singularity — one record per real-world asset, not five conflicting entries that each tell part of the story.
Miss any one of these and confidence collapses. A CMDB that’s 95% correct but six months out of date is still untrustworthy, because nobody knows which 5% is wrong.
A practical approach
1. Connect authoritative sources, don’t re-key them
For every attribute, decide which system is the source of truth — the endpoint manager for installed software, the directory for identity, procurement for warranty — and pull from it directly. If a human is re-typing data that already exists in another system, that’s a defect to remove, not a process to optimise.
2. Reconcile and de-duplicate into one record
Match records across sources on stable keys (serial number, asset tag, UPN) and merge them into a single canonical asset. This is where most of the value is: a device seen by Intune, Entra ID and your CMDB should be one entry that carries the best of all three, not three rows that argue with each other.
3. Give every record an owner and a lifecycle state
An asset with no owner and no status (active, in repair, retired, in stock) can’t be governed. Lifecycle state is what lets you separate “missing because it’s gone” from “missing because we lost track of it”.
4. Automate the refresh
Accuracy is not a project you finish; it’s a property you maintain. Reconciliation has to run on a schedule — continuously, in the background — so the record corrects itself as the source systems change. Anything that depends on someone remembering to update it will drift.
5. Measure it
You can’t manage what you don’t measure. Sample a slice of the CMDB against physical or authoritative reality each month and track a simple accuracy rate over time. If it’s trending up, your process works; if it’s flat or falling, something upstream is broken.
A CMDB that’s wrong is worse than no CMDB at all — because people make confident decisions on bad data. The goal isn’t a perfect database; it’s one that corrects itself faster than reality changes it.
How ClearVisibility helps
This is exactly what ClearVisibility is built to do. It connects to the systems you already run — Microsoft 365, Intune, Azure, Entra ID, Active Directory, Secure Score and ServiceNow — reconciles their records into one canonical asset per device, user and licence, and keeps it current automatically on a schedule you control. Coverage, correctness, currency and singularity stop being a manual chore and become a property of the platform.
See it on an estate like yours
A 30-minute demo shows ClearVisibility doing this on data that looks like yours.
Book a demo