Your BMS Says Green. Your Occupancy Says 31%. Which One Is Wrong?
A floor is running at 31% occupancy.
The BMS is heating it as scheduled.
The booking system says several meeting rooms are full.
The energy dashboard shows no fault.
Which system is wrong?
Possibly none of them.
And that is the problem.
Most commercial buildings do not suffer from a shortage of data. They suffer from data that becomes meaningful only when one system is read against another.
Occupancy without energy tells you where people were.
Energy without occupancy tells you what the building consumed.
Bookings without actual presence tell you what people intended to do.
A BMS tells you whether plant followed its instructions — not necessarily whether those instructions still made commercial sense.
The expensive answer often sits between those systems.
That is why agentic AI could matter enormously to Facilities Management.
It is also why simply adding an AI assistant to another dashboard will not be enough.
FM is moving from “what happened?” to “what should we do?”
For years, smart-building technology has largely competed on visibility.
More sensors. More dashboards. More alerts. More data.
The next competition is different:
Who decides what happens next?
Johnson Controls’ 2026 research found that 67% of surveyed facilities teams already use AI and 61% intend to expand its use. More importantly, data quality and integration emerged as the most frequently cited barriers to further AI adoption.
That matters because the industry is already moving towards AI that does more than describe a problem.
New agentic AI capabilities are being designed to surface insights, recommend actions and increasingly support approved operational workflows while keeping facilities teams in control.
The direction is becoming clear.
The valuable FM co-pilot will not simply tell you:
“Energy consumption increased 11%.”
It will need to answer:
“Why did it increase, was the increase justified, what should change, what is the expected value, and who needs to approve it?”
That requires something much harder than a chatbot.
It requires context.
The hidden problem: individually correct systems can produce a collectively wrong building
Consider one office floor.
The booking platform says 78% of meeting-room capacity is reserved.
Occupancy sensing shows that only 36% is actually being used.
The BMS sees no fault because HVAC started exactly when programmed.
The energy system sees normal weekday consumption.
Facilities therefore receives four apparently reasonable signals.
But read together, they reveal a different problem:
the building is conditioning demand that does not exist.
No individual dashboard necessarily identifies that.
This is the ceiling of single-domain optimisation.
Energy software can optimise energy.
Space software can optimise space.
A BMS can optimise plant against its configured control strategy.
A booking platform can optimise reservations.
But many of the commercially important questions are cross-domain questions:
- Which underused floors are still carrying near-normal HVAC cost?
- Which highly booked rooms are consistently empty?
- Which buildings have enough unused capacity to absorb teams from another site?
- Which comfort complaints correlate with occupancy, temperature, CO₂ and plant behaviour?
- Which operating schedules stopped making sense after hybrid working changed attendance?
The answer does not live in one system.
It lives in the relationship between them.
That is the principle behind DIREK’s approach to reading building information across space, energy, environment and operational systems.
See how D-XPERT reads building domains together
Occupancy has returned. Uniform demand has not.
This becomes more important as offices get busier.
CBRE’s 2026 Global Workplace & Occupancy Insights found average global office utilisation had risen to 53%, from 38% in 2024.
Peak utilisation reached 80%.
That sounds like a return-to-office story.
Operationally, however, it creates a more interesting problem.
An 80% Tuesday and a 30% Friday can exist in the same building.
Peak demand may justify the estate’s capacity.
It does not justify treating every hour of every weekday as peak demand.
The question therefore changes from:
“How occupied is our building?”
to:
“When occupancy changes, what else should change with it?”
Heating.
Cooling.
Ventilation.
Lighting.
Cleaning.
Security.
Catering.
Space allocation.
Eventually, perhaps even the estate itself.
This is why occupancy monitoring becomes substantially more valuable when it terminates in an operational decision rather than a utilisation report.
Read why occupancy data becomes more valuable when connected to the BMS
AI does not solve fragmented building data. It exposes it.
This may be the most important point for FM leaders considering agentic AI.
A more capable model cannot reason from information it cannot access.
If occupancy lives in one platform, energy in another, work orders somewhere else, environmental data in standalone sensors and operating schedules inside the BMS, an AI co-pilot inherits the fragmentation.
Verdantix’s 2026 Global Real Estate Survey makes the consequence unusually clear: 74% of organisations said AI had pushed them to centralise data from multiple systems and solutions.
That is not the glamorous part of agentic AI.
It may be the most important part.
The prerequisite for useful AI in FM is therefore not necessarily another AI model.
It is a reliable operational context layer that can connect the information already available, identify what is missing and preserve enough provenance to explain the resulting recommendation.
In practical terms:
Connect what you have.
Sense what you don’t.
Then reason across the whole picture.
Where a building lacks occupancy, environmental or granular energy information, that does not necessarily mean instrumenting everything permanently.
Targeted sensing can fill the specific information gap that prevents a decision.
DIREK, for example, supports portable occupancy, environmental and energy sensing packages that can be deployed without replacing existing systems.
Explore DIREK smart-building sensing packages
The technology is not the point.
Closing the information gap is.
A finding that nobody acts on is worth very little
There is another problem AI alone does not solve.
FM teams already have alerts.
Lots of them.
The harder problem is converting a finding into a decision before the conditions that created it change.
Suppose an analytics system identifies an underused floor.
What happens next?
A useful operational recommendation should answer at least four things:
- Evidence: What happened, and which data proves it?
- Value: What does continuing the current operation cost?
- Action: What specifically should change?
- Ownership: Who has authority to make that change?
And then one more question matters:
When should the recommendation be checked again?
Because buildings move.
Occupancy changes.
Teams relocate.
Weather changes.
Tariffs change.
Seasons change.
A recommendation that was correct in February may be wrong in July.
That is the difference between reporting and an operational decision loop.
Real buildings show why this matters
Finding the cost hidden inside operating schedules
At a Northern England secondary school, DIREK connected EnergyLens to existing meter infrastructure rather than replacing the BMS.
The analysis found plant starting 1 hour 45 minutes before it was needed, systems continuing 3 hours 15 minutes after the target shutdown, and weekend operation above the expected baseline.
Across nine months, £10,239 of out-of-hours electricity cost was identified, representing 18.1% of electricity spend during the analysed period.
The first corrective plan did not require a major capital project.
It required schedule corrections, timer checks and a shutdown protocol.
That distinction matters.
The value was not:
“We collected energy data.”
The value was:
Here is what is happening → here is what it costs → here is what to change.
Read the DIREK out-of-hours energy waste case study
Turning occupancy into an estate decision
The same pattern appears in space.
At a major UK construction and built-environment organisation, occupancy analysis across desks and meeting rooms supported a 46–50% desk-footprint reduction with peak capacity maintained, with £98,000 of annual savings identified.
Again, occupancy itself was not the decision.
It was evidence for the decision.
Read the office footprint optimisation case study
ESG has the same underlying problem: can you reproduce the evidence?
The cross-domain issue extends beyond daily operations.
It reaches ESG reporting.
GRESB’s current methodology reinforces the importance of actual asset-level performance data and constrains how missing information may be estimated.
The broader direction is clear: sustainability evidence is moving towards measured, traceable performance.
So an uncomfortable but useful test for an estates team is:
If an auditor asked you tomorrow to reconstruct a specific reporting month from the underlying building data, how much manual work would it take?
If the answer involves finding old spreadsheets, requesting exports from multiple suppliers and reconciling inconsistent timestamps, the problem is not primarily ESG reporting.
It is evidence architecture.
That is why continuous operational data can serve two purposes.
It can help run the building today.
And it can create the evidence needed to explain what happened later.
Explore DIREK’s approach to ESG evidence and reporting
Five questions to ask your estate before buying an AI co-pilot
You do not need an AI strategy workshop to find out whether your estate is ready.
Ask it five questions.
1. Which floors were below 40% occupancy last month but continued running normal heating or cooling schedules?
This tests whether occupancy and building operation can actually be compared.
2. Which meeting rooms have high booking rates but low measured utilisation?
This exposes the difference between demand on paper and demand in reality.
3. Which out-of-hours energy loads cannot be explained by actual building use?
This tests whether energy is being interpreted in operational context.
4. If we had to reproduce one month’s ESG evidence tomorrow, which data would require manual reconstruction?
This tests provenance and data continuity.
5. When a problem is identified and costed, who receives the action — and can we verify that it worked?
This tests whether your analytics stack ends with an alert or with a decision.
If these questions require several dashboards, spreadsheet exports and three weeks of consultant analysis, buying a more sophisticated AI model will not fix the underlying problem.
It may simply explain the fragmentation more eloquently.
What should agentic AI in Facilities Management actually do?
A useful FM agent should be able to:
Observe across domains → establish context → identify a problem → quantify it → recommend an intervention → route it to an owner → verify the result.
That is a much higher bar than conversational access to a dashboard.
It is also where agentic AI becomes commercially interesting.
Not because the building can talk.
Because the building can finally connect what it knows to what somebody should do.
The DIREK view
DIREK’s thesis is simple:
Every part of a building can be managed while the whole remains unoptimised.
That is why D-XPERT is designed as a cross-domain operational intelligence layer rather than another replacement building system.
It reads existing BMS, meters, booking systems and sensors first.
Where the necessary information does not exist, sensing fills the gap.
Space, energy and environmental information can then be read together so recommendations are based on the operating context rather than one isolated metric.
Explore D-XPERT cross-domain operational intelligence
The opportunity for agentic AI in FM is real.
But the winning question for 2026 is probably not:
“Which AI model should we put on our building?”
It is:
“What can our building answer when the question crosses more than one system?”
That answer tells you whether you have an AI-ready estate — or simply a collection of intelligent systems that still cannot see one another.
Want to see what your estate can actually answer?
DIREK connects existing building systems, adds sensing where information is missing, and turns occupancy, energy, environmental and operational data into cross-domain decisions.