AI for Facility Management
Facility management projects fail on meters and controls far more often than on models, which is why we assess those first.
Facility management runs on work orders, service level agreements and building systems that were commissioned years ago and drifted since. The realistic AI opportunity sits closer to the work orders than to the plant.
AI for facility management covers work order classification and triage, maintenance planning from asset and failure history, energy and plant optimisation, occupancy and space analytics, service level performance analysis, and contract and compliance document management.
Work orders, the reliable place to start
A facility management operation generates high volumes of requests in free text, classified inconsistently by whoever logged them, and routed by rules written years ago. That is a well-posed problem with immediate measurable effect.
- Classification and routing is directly measurable. Time to correct trade, repeat visits caused by wrong dispatch, and first-time fix rate all move.
- Urgency assessment is harder and more valuable. Reported urgency is unreliable in both directions, and the cost of getting it wrong is asymmetric.
- Duplicate detection saves real visits. Multiple occupants reporting the same fault generate multiple dispatches in most operations.
- The historical record is the training data. What was actually done, by which trade, and whether it recurred is already in the system.
- Reactive-to-planned ratio is the metric that matters. It is what the analysis should be aimed at, and it is what the client contract is usually written around.
Repeat work orders are the signal hiding in plain sight
The same asset generating repeated reactive work orders is telling you something the planned maintenance regime is not catching, and in most CAFM systems nobody analyses recurrence because each order is closed individually. Linking work orders to assets and counting recurrence identifies both the assets that should be replaced rather than repaired and the maintenance tasks that are not working. It needs no model — it needs the asset linkage to be reliable, which is usually the actual project.
Plant and energy optimisation, and the instrumentation reality
The savings are real where the conditions are met
Tuning HVAC and plant against occupancy, weather forecast and tariff produces genuine, measurable savings on buildings where the control system can accept a setpoint and the metering can verify the result.
Most buildings do not currently meet those conditions
Sub-meters that were never commissioned, sensors reading a constant since installation, a building management system whose integration points are undocumented, and controls overridden manually years ago and never reset. This is the ordinary state of a portfolio, not an exception.
The honest first project is often instrumentation
Where the data foundation is not there, the optimisation model has nothing to act on and no way to prove it worked. We assess this in the first fortnight and have recommended clients spend the budget on metering instead, which is a smaller engagement than the one they enquired about.
Regulation is pushing the same direction
EU building automation obligations extend to systems above 70 kW by 2030, down from 290 kW, and building performance reporting requires consumption data that is comparable across a portfolio. The compliance work and the optimisation prerequisite are the same work.
Maintenance planning without the failure data
| Approach | Data needed | Realistic in most portfolios |
|---|---|---|
| Failure prediction per asset | Many recorded failures per asset class | Rarely — the failures are not in the data |
| Condition monitoring against known-good | Sensor or performance history per asset | Where instrumentation exists |
| Recurrence analysis | Work orders linked to assets | Yes, and usually unexploited |
| Deterioration by asset class and age | Asset register plus replacement history | Yes, and it informs capital planning |
| Maintenance regime effectiveness | Planned tasks against subsequent reactive work | Yes, and it frequently changes the regime |
| Spare parts and stock optimisation | Consumption history | Yes, straightforward and measurable |
Ask whether the planned maintenance is working before predicting failures
Comparing planned maintenance tasks against the reactive work orders that followed them tells you which tasks prevent failures and which are being performed because they always have been. That analysis uses data every CAFM system holds, needs no model, and has removed cost from maintenance regimes without reducing reliability. It is a better first project than predictive maintenance in almost every facility management portfolio we have looked at.
How an engagement runs
Instrumentation and data linkage assessed before anything is modelled.
Scope and instrumentation assessment
Meter and sensor condition, BMS integration, and whether work orders link reliably to assets.
Historical analysis
Recurrence, maintenance regime effectiveness, and reactive-to-planned ratio by asset class.
Build
Work order triage and routing, or plant optimisation where the instrumentation supports it.
Trial
On live work orders or a defined building set, with savings verified against metering.
Operation
Retraining as the estate changes, with drift in classification monitored.
What you receive
Work directed correctly, and buildings tuned where they can actually be tuned.
Work order classification and routing
Trained on your own historical dispatch and outcome data.
Urgency assessment
With the asymmetry between under- and over-calling made explicit.
Duplicate and recurrence detection
Repeat faults surfaced at asset level rather than closed individually.
Maintenance regime analysis
Which planned tasks prevent reactive work and which do not.
Plant optimisation
Where controls and metering support it, with savings verified rather than modelled.
Instrumentation assessment
An honest statement of what the estate can and cannot currently support.
Is this the right starting point?
Worth being direct. There are situations in facility management where custom AI work is the wrong spend, and those are listed rather than buried.
Worth doing if
- Work orders arrive as free text and are classified inconsistently.
- Repeat visits caused by wrong-trade dispatch are a known cost.
- The same assets generate recurring reactive work and nobody analyses recurrence.
- Planned maintenance has never been tested against the reactive work that followed it.
- You have working sub-metering and a building management system that accepts setpoints.
Do something else if
- Sub-metering does not work and there is no budget to fix it first.
- Work orders cannot be linked reliably to assets in the CAFM system.
- You want failure prediction on assets with no recorded failure history.
- Building controls cannot accept an external setpoint and will not be changed.
Frequently asked questions
Marked up with FAQPage schema so these answers can surface directly in search results and inside AI assistant responses.
Why do you assess instrumentation before discussing models?
Because it decides whether the project is possible. Plant optimisation needs a control system that accepts a setpoint and metering that verifies the result, and in a typical portfolio a meaningful share of sub-meters were never commissioned, sensors read a constant, and controls were overridden manually years ago. A model deployed into that has nothing to act on and no way to demonstrate it worked. We assess it in the first fortnight, and we have told clients to spend the budget on metering instead — which is a smaller engagement than the one they enquired about.
Is predictive maintenance realistic for building services?
Usually not in the form it is sold, for the same reason as everywhere else: the failures are not in your data. What works is condition monitoring against a known-good baseline where instrumentation exists, recurrence analysis on work orders linked to assets, and testing whether your planned maintenance regime actually prevents reactive work. That last analysis needs no model, uses data your CAFM system already holds, and has removed cost from maintenance regimes without reducing reliability. It is the better first project.
What is the quickest measurable win in FM?
Work order triage. Requests arrive as free text, get classified inconsistently, and are routed by rules nobody has revisited, which produces wrong-trade dispatch, repeat visits and slow first-time fix. Classification and routing trained on your own dispatch and outcome history moves all three, the effect is measurable within weeks, and the risk is low because a misrouted order is recoverable. Duplicate detection usually comes along with it and saves visits immediately.
Can occupancy data improve building operation?
Yes, substantially, and it is personal data about the people in the building. Tuning ventilation and conditioning to actual occupancy rather than a fixed schedule is one of the more reliable savings available, and occupancy sensing also informs space decisions worth more than the energy. The design requirement is aggregation: collect what the control decision needs at zone level rather than tracking individuals, be explicit about what is retained and for how long, and be able to explain it to the people who work there. Built that way it is unproblematic; built as individual tracking it becomes a workplace dispute.
How do we prove savings to a client?
By agreeing the measurement basis before deployment, which is where most of these engagements go wrong. Weather-normalised baselines, an agreed comparison period, and metering at the level the intervention operates on, all settled in advance. Retrofitting a savings claim after the fact produces a number the client's own analyst can dispute, and in a contracted FM relationship that is worse than a smaller verified saving. If the metering cannot verify the claim, the honest position is that the project cannot demonstrate its own value yet.
Related verticals
Organisations in facility management usually share data, buyers or regulators with these. All fourteen are listed on the Real Estate & Construction page.
AI for Property Management
Maintenance triage, arrears and renewal modelling — with the screening and rent-setting law stated plainly.
Read more →AI for Infrastructure Owners and Operators
Inspection imagery, condition assessment and renewal prioritisation — with the advisory line drawn deliberately.
Read more →AI for Commercial Real Estate
Lease data extraction, portfolio analytics and leasing models — starting with the terms nobody currently holds in a database.
Read more →Tell us what the problem looks like.
Thirty minutes, no charge, no deck. We will tell you whether this is an AI problem, a data problem, or a process problem — and we will say when the honest answer is to buy something rather than build it.