There is a gap in the fixed asset lifecycle that almost nobody reconciles in real time. An asset exists twice: once as a physical thing in a building, and once as a record in a register. Those two versions drift apart from the moment the asset arrives, and in most organizations nothing forces them back together until an audit, a disposal that will not tie, or an insurance renewal that asks an uncomfortable question.
This is not a software gap. Every system involved is perfectly capable of holding the right answer. The gap is that the physical event and the financial record are updated by different people, on different triggers, with no obligation to tell each other anything.
I spent about thirty-six hours recently inside a live experiment that pointed a digital twin, a governance layer and a group of genuinely brilliant practitioners — from building information modeling, AI governance and facilities technology — at exactly that gap on one real building. It is the most useful thing I have watched in a while, and not for the reason I expected.
What a digital twin actually is, for finance people
If the term has only ever reached you as a conference slide: a digital twin is a continuously updated model of a physical building — its spaces, its systems, its equipment — that changes when the building changes. A live building model, in other words, rather than a set of drawings that were accurate on the day they were signed.
The reason this matters to a fixed asset register is simple. A register is a claim about the physical world: this asset exists, it is here, it is this old, it is worth this much. A live model is the only version of the physical world that can answer back.
Data exchange is solved. Coordination is not.
The instinct in this space is to ask how the systems talk to each other. That question is largely answered. APIs exist, standards exist, integrations exist, and vendors will happily sell you more of all three.
The unsolved problem is coordination: getting the right piece of information to the right person at the right moment, across departments that do not share a calendar, a vocabulary, or a manager. Facilities knows the chiller was replaced. Accounting knows what capitalizing it requires. Nobody owns the sentence between those two facts.
My own assessment framework is built around that failure, and it is why it is shaped the way it is. It walks every stage of the lifecycle — from capital planning and procurement through capitalization, depreciation, transfers, physical verification, retirement, reporting and audit — against every dimension that can break a stage: governance, policy, process, people, data, systems, controls, documentation and the rest. Cross those two lists and you get well over two hundred points where a handoff can fail silently. Most of them are currently held together by a person remembering, or by tribal knowledge that leaves the building when they do.
That is the number worth sitting with. Not two hundred things that are broken — two hundred places where nothing would visibly break if they were.
What we actually tested
The setup was deliberately unglamorous. A live model of a real building acted as the shared factual reference, so no department was working from its own privately drifting copy of the truth. On top of that sat an AI governance layer enforcing exactly one rule: a physical event could not become a financial or operational decision without the evidence to support it.
Credit where it is due, because none of this was mine. Greggory Don Butler designed and ran the governance framework every decision had to clear. Kimon Onuma supplied the live building model. Thomas Dalbert provided the API access that made it testable rather than theoretical. Mike Bordenaro kept pushing the group out of the sandbox and toward a real building, which is the part most experiments never survive.
The warranty that stopped the line
At one point the warranty status on a piece of mechanical equipment could not be verified. The process did not guess, and it did not quietly proceed on an assumption that would have been invisible six months later. It halted, named the specific missing item, and routed a human to go resolve it — then picked back up from verified fact.
If you have ever tried to work out after the fact whether a repair was a warranty claim, an expense or a capital improvement, you already know what that pause is worth. That determination is nearly free at the moment of the event and nearly impossible eighteen months later.
The work order that stayed wrong in public
A work order was routed to the wrong room. What made it interesting is what did not happen: it was not silently corrected. The error stayed visible, stayed attributed, and was fixed openly once the model exposed it.
That is the discipline most asset environments lack. Errors get absorbed rather than recorded, so the same failure repeats indefinitely because no one can prove it has a pattern.
The eight things that all had to be true at once
The experiment looked like a technology demo and behaved like a governance exercise. Eight separate layers had to hold simultaneously, and every one of them is a question a fixed asset environment should already be able to answer:
- A shared, accurate picture of the physical building. One version, not one per department.
- A record that cannot be quietly rewritten later. If history is editable, the audit trail is decoration.
- A clear line on who — or what — has authority to approve or execute. Named, not assumed.
- A working definition of "enough evidence" before a decision is made. Written down in advance, not negotiated under deadline.
- Explicit roles for people versus automated systems. The gap between "someone reviews this" and "something reviews this" is where most control failures live.
- Hard boundaries on what a system may do rather than merely record. Recommending and acting are different permissions.
- Confidence that the physical facts being relied on are still current. Data has a shelf life, and nobody prints the date on it.
- A requirement to re-check a decision when the underlying facts change. Instead of letting an old answer ride forever, which is the default everywhere.
Note how few of those are technology problems. Seven of the eight are policy, governance and ownership questions that an organization can answer today, on the systems it already owns.
Why this gets worse at portfolio scale
We were looking at one building, and it produced a steady stream of pauses and exposed gaps. Now multiply that by a portfolio: thousands of assets, dozens of locations, incomplete acquisition history, three generations of naming conventions, and a register nobody has physically verified in years.
Those conditions already exist. The difference is only that nothing is currently stopping to point at them. In one building, an unverifiable warranty halts a process. In a portfolio, it surfaces two years later as an audit finding — or it never surfaces at all, which is the outcome that should worry you more. A finding is expensive. A wrong number that nobody ever questions gets used to make capital decisions.
What to do before any of this reaches you
The temptation with a story like this is to go shopping. Resist it. Everything above is an amplifier, and an amplifier stacked on a broken process is just a faster wrong answer — which is the whole argument in Don't Automate a Broken Fixed Asset Process, and the reason dirty data in still means dirty data out no matter how sophisticated the layer on top gets.
The preparation work is unromantic and entirely within reach:
- Name the trigger for every physical event with a financial consequence. Installation, replacement, relocation, retirement. Who has the obligation to say so, and to whom.
- Write down what counts as sufficient evidence for capitalization, for a placed-in-service date, for a disposal. If it is not written, it is being decided differently by every person who touches it.
- Establish where the physical truth lives. A model, a CMMS, a verification cycle — something other than a spreadsheet that agreed with reality in 2019.
- Make errors visible instead of absorbing them. A logged correction is data. A silent one is a repeat.
- Find the handoffs nobody owns before a system inherits them. That is precisely what a Fixed Asset Health Assessment is for, and why it looks at governance and ownership rather than just at the register.
An organization that has done those five things can adopt a live building model and get compounding value out of it. An organization that has not will get a very expensive, very well-instrumented mirror held up to its handoff problem. Worth remembering that the fixed asset problem is usually not in fixed assets.
Where the work is going
This is an open effort and it publishes as it goes, mistakes included, which is rarer than it should be. The more technical write-up of the test is "When the Building Said No" at Automated Buildings, and the broader project lives at BIMStorm.com/AI, including an ongoing live session series building toward Greenbuild in New York.
I will keep writing about what it means for fixed asset and capital planning work as it develops. My read so far: the technology arriving in this space is going to be genuinely good, and it is going to be merciless about coordination. The organizations that benefit will be the ones that fixed the handoffs first.