The FM Industry Does Not Have an AI Problem. It Has an Information Problem.

There is no shortage of conversation about AI in facilities management at the minute.

Predictive maintenance. Automated workflows. Digital twins. Intelligent buildings. Systems that can identify risk before somebody even knows there is a problem.

And I think much of that technology will genuinely change the way we manage buildings.

But there is one fairly basic question that I keep coming back to:

What exactly are we asking the technology to make decisions from?

Because the most sophisticated AI platform in the world cannot compensate for an asset register that does not reflect the estate, a PPM structure that has developed inconsistently over ten years, or fifteen different versions of the same building information sitting in different spreadsheets.

That is why, for me, one of the most important conversations in FM right now is not really about AI at all.

It is about data readiness.

MRI Software published UK research earlier this year which found that 52% of FM professionals were not confident their data was accurate enough to support AI or digital decision-making. More than eight in ten respondents expected smart technology and automation to help define the next five years of FM, yet one in five organisations said they were not currently measuring FM performance at all.

That is quite a gap.

And I suspect anyone who has spent much time inside operational CAFM systems will recognise it.

The industry is talking about predicting failures while some organisations are still trying to establish exactly which assets they own.

We are discussing automated decision-making while engineers are selecting the wrong asset because there are three slightly different versions of it in the system.

We are talking about digital twins while building names, room references and asset codes still differ between estates, finance, compliance and maintenance teams.

That is not a criticism of those organisations.

It is simply the reality of how estates information develops.

Most CAFM databases were not created in a laboratory.

They have grown over years through mobilisations, acquisitions, contract changes, surveys, imports, new buildings, different suppliers and different operational teams.

Every individual decision may have made sense at the time.

Eventually, however, those decisions accumulate.

And that is where organisations can end up with a very capable piece of software sitting on top of information they do not completely trust.

The technology is moving quickly

MRI’s latest EMEA FM commentary talks about the move from reactive to predictive facilities management.

The direction makes sense.

Connected systems, IoT, analytics and AI can potentially identify patterns that a person looking at individual work orders would never see.

MRI’s newer Agora Intelligence and Orchestrator products take that further by combining recommendations with automated workflows, while retaining governance, traceability and human oversight.

IFMA is describing a similar direction internationally.

Its September feature on the “intelligent facility” describes AI, digital twins, BIM, predictive analytics and intelligent orchestration converging to move facilities away from purely reactive operation towards proactive, data-led management.

So the technology is not theoretical anymore.

The interesting question is becoming less:

Can the system do it?

and more:

Is the organisation ready for the system to do it?

Those are two very different things.

A predictive maintenance example

Take something relatively simple.

An air handling unit fails repeatedly.

In theory, a modern CAFM and analytics environment could look at historic jobs, sensor readings, failure patterns, engineer comments, downtime, cost, age and condition and start identifying that deterioration before the next breakdown.

That is genuinely useful.

But only if the data connects properly.

If engineers have been raising jobs against AHU 4, AHU-04, AHU/04, Air Handling Unit 4, and occasionally just the building rather than the asset, then the maintenance history is already fragmented.

If replacement components were never reflected in the asset register, the age may be wrong.

If engineer notes are inconsistent, the failure mode may be impossible to identify.

If the PPM schedule is attached to a duplicate asset record, you may not even be comparing the right maintenance history.

The AI has not failed. The information architecture has.

That distinction matters.

This is becoming an asset-management issue, not just a CAFM issue

IWFM has launched its Strategic Asset Management Roadshow this autumn.

The programme specifically links AI and digitisation with full lifecycle value, while also discussing the implications of poorly managed assets, the Golden Thread and the move away from short-term “race to the bottom” approaches.

That feels like the right conversation.

A CAFM system should not simply be somewhere we store maintenance tasks.

It should help the organisation understand its physical estate.

What assets do we have?

Where are they?

What service do they support?

How critical are they?

What condition are they in?

What maintenance do they require?

What has happened to them historically?

What will they cost us over the next five, ten or twenty years?

The Institute of Asset Management has increasingly emphasised the same outcomes-led approach: connecting assets, data, risk, organisational objectives and lifecycle decision-making rather than simply measuring activity. Its updated asset-management maturity methodology is aligned with ISO 55001:2024, and its 2026 guidance highlights data quality and transparent reporting as important enablers of long-term value.

That is a much broader concept than “keeping the asset register tidy”.

Good asset information is becoming part of business decision-making.

Regulation is pushing in the same direction

There is another reason this matters.

Building safety regulation is increasingly formalising the requirement for accurate, usable information.

For higher-risk buildings in England, the Building Safety Regulator describes the Golden Thread as a digital record that must be secure, available, understandable and capable of acting as a single source of truth about the building.

Crucially, this is not just information created during construction and forgotten after handover.

It has to remain relevant and usable through occupation and ongoing management.

That principle is worth thinking about even for buildings that do not fall within the specific regulatory regime.

Does the information we hold actually describe the estate we are responsible for today?

That sounds obvious.

In practice, it is a surprisingly high standard.

So where should organisations start?

I do not think the answer is a two-year data-cleansing programme before anyone is allowed to implement anything new.

That creates a different problem.

You can spend forever trying to perfect a dataset.

The better approach, in my experience, is to work backwards from the operational requirement.

What decisions do we need to make?

What workflows do we need to support?

What compliance evidence do we need?

What assets genuinely matter?

What information would make those processes work properly?

Then define the minimum trusted dataset needed to support them.

That may mean resolving the building hierarchy first.

It may mean agreeing mandatory asset fields.

It may mean rationalising PPM structures.

It may mean identifying the authoritative source for each dataset rather than copying everything into another spreadsheet.

It may mean changing what engineers capture through MRI Go.

And in some cases, it may simply mean accepting that information we have collected historically serves no useful operational purpose and does not need to migrate into the new environment.

That is where I think independent operational consultancy has an important role.

The software provider should know its technology better than anybody.

But somebody still needs to translate what the business is trying to achieve into what the technology and the underlying data actually need to look like.

Those are related disciplines, but they are not exactly the same thing.

AI will make good data more valuable — and bad data more expensive

There is a temptation to think AI will somehow solve poor data.

In some areas it will absolutely help identify anomalies, duplicates and patterns.

But as organisations give technology more influence over operational decisions, the quality of the underlying information becomes more important, not less.

If a human sees an obviously incorrect asset description, they may instinctively question it.

An automated workflow may simply act on it.

That means governance matters.

Structure matters.

Ownership matters.

And understanding why a particular field exists matters.

The businesses that get the most from AI in FM will not necessarily be the ones that buy the newest technology first.

I suspect they will be the organisations that understand their estate, understand their processes and have enough confidence in their information to allow technology to do something genuinely useful with it.

So while the industry quite rightly talks about predictive FM, intelligent buildings and agentic workflows, I think there is a much less glamorous competitive advantage available right now:

Know your assets. Know your data. Know which information you trust.

Then automate. Then predict.

— Rob Davison Managing Director Quality Evolution Services