What I will cover:
- Why the physical layer of a CMDB drifts out of date
- Why the health score doesn’t flag it
- What that inaccuracy costs a colo
- How to fix the problem
The CMDB should be the single source of truth. It is not. And many departments pull critical information from it — capacity planning, incident and change, service delivery, finance, the commercial team — each taking what they need to do their jobs.
For most of the record, that works well enough. Then you get to the building — the physical estate — and the honest questions start. Which cabinet is that in? Which rack unit? Which PDU outlet, and is it on the A feed or the B? Which patch panel port? Which cross-connect or meet-me room connection leaving which cage?
Then the estate carries on.
A device is installed, a reboot is performed, a cross-connect or meet-me room connection is added, a CRAC unit has a filter changed. Moves, adds, changes, decommissions — MACD — along with servicing, repairs and maintenance: the ordinary daily churn of IT, facilities, network, building fabric and spares/inventory.
Not everything is captured, because almost nothing was ever built for the person doing it. They are in a cold aisle, under time pressure, not at a desk and not about to open a laptop and fill in a form. And in a colo the churn is worse, because some of the people making the changes are not even yours — smart hands, remote hands, and third-party maintainers.
So, the record does not fail. It drifts, quietly, at the rate the estate changes, which is daily. And the health score does not catch it. Completeness only asks whether a field is filled in. Correctness hunts for stale, orphaned and duplicate records against the sources feeding them. A location reading “Rack 14” scores as ‘complete and correct’ for as long as it says something. It does not have to be true. Nothing is checking this information against the floor.
The database being wrong is not the problem. This is.
A slightly inaccurate database is not, by itself, something to lose sleep over. What should keep you up is everything downstream that treats it as true — and each of those things sits on a different desk.
Felt by finance and the billing team.
Start with billing, because it runs both ways. A tenant draws power the business is not charging for, in a space nobody recorded, and every month it stays off the invoice is margin walking out of the building — with nothing to trip an alarm. The other direction is louder: bill a tenant for a circuit they decommissioned last year, and the company will not simply refund it, it will spend the renewal defending its competence. The common case is not that a colo bills nothing. It is that it bills incorrectly and cannot say which way.
Felt by the sales and commercial team.
Then capacity. A customer wants to buy, and the salesperson needs to see what is available — the answer should fall straight out of the record. When it cannot, it takes several people to find it. The capacity is there, but it is hidden: real space, real power, real cross-connect and meet-me room capacity, real cooling, all paid for and earning nothing. Multiply that across a room, a floor, a site, the whole estate, and it is significant revenue missed, and risk increased.
Felt by service delivery and the account team.
Then the customer. A tenant asks a simple question — what have I got, where is it, what is it drawing — and the answer takes three people and two days to assemble. Or a smart hands job overruns because the engineer cannot find the kit. None of that ever shows up as a fault. It shows up as a tenant who trusts you slightly less at renewal.
Felt by procurement and facilities.
Then the contracts. Facilities management and third-party maintenance are priced off the asset record. When that record overstates the asset numbers, or lacks accurate condition data, the business overpays — quietly, on renewal after renewal.
Felt by operations.
And underneath all of it, the people — some of them gap-filling data and mending broken processes, chasing what changed, reconciling one system against another, fixing the same breakages every month. That is not a project cost. It is a permanent tax, and what it buys is a record that is only slightly less wrong.
Felt by the sustainability and risk team.
And now the reporting. Carbon and ESG numbers get built off the same physical record — and when it is incomplete, energy and emissions are estimated from nameplate ratings, the draw printed on the box rather than the power actually pulled on the floor. That gap used to be an efficiency footnote. It is now a figure the business certifies to regulators, customers and its own board, and “we estimated it” is a poor answer to give any of them.
Notice where those costs land. The misbilling is finance’s problem. The stranded capacity is the commercial team’s. The slow answer to a tenant is service delivery’s. The overpaid contract is procurement’s. The ESG figure is the sustainability and risk team’s. The manual true-up is operations’. Not one of them sits on the budget line of whoever keeps the CMDB running — which is exactly why the underlying data problem never gets funded, and exactly why it can be a challenge to fix.
The fix is capture, not cleanup.
Cleanup does not solve this problem. Cleanup is a treadmill — you finish the pass and the estate has already moved. Neither does an audit, which buys you an accurate picture of the day the auditors left. The only thing that holds is capturing change where and when it happens.
And one thing is always true of a physical estate: nothing changes unless a person changes it. Every move, add, change and decommission is done by someone standing in front of the physical thing, at a known moment, knowing exactly what they did. That is the source of truth. It was there the whole time. It has never fed the CMDB only because nothing was ever built to capture it where that person actually is.
That is what we built Spire™ to do. It captures MACD at the moment it happens, holds the physical estate as one record — halls, rooms, cages, rows, cabinets, rack units, power from source to consumer, ports, fibre, cross-connects — and feeds it back into your CMDB. We made the capture take seconds so that it gets used, including by the people who do not work for you: a smart hands operative is already noting the job is done, and the only question is whether that note is a line in a ticket or your record updating itself. Your service management platform does not move. Nothing gets replaced. The physical layer underneath it simply becomes true — which, as it happens, is the only basis on which the AI agents everyone is now asking about can be allowed anywhere near a data hall. This is how our customers can rely on their CMDB data.
What do I do next?
Do not start with the record. Start with the work.
Pull the last ten physical jobs done in your estate — a device installed, a cross-connect added, a card swapped, a filter changed. They will be sitting in tickets or in a smart hands log. For each one, check whether the CMDB moved. Not whether the data looks plausible. Whether that specific piece of work left a trace. It takes an afternoon.
Checking a single cabinet only tells you something if that cabinet happened to change. If nothing has been touched since it was recorded, the record is correct and you have learned nothing. Starting from the work removes that luck. Ten jobs completed and only three recorded is a capture rate — and your bills, your capacity plans, your contracts and your carbon numbers are all running on it.
If those ten jobs all left a trace, you do not have this problem, and I would be glad to hear that too. If they did not, you have a number instead of a suspicion.
And if nobody can answer your questions, tell me what you found. That is a better first conversation than a demo — for you and for me.
Martin Docherty (DCIS®) works on data centre asset management at Assetspire. Find him on LinkedIn.