Most facilities have data but little AI intelligence

Across industries, facility leaders are investing in emerging technology: robotics, sensors, IoT, and AI. One estimate found that 28% of organizations have already embedded AI solutions in their FM operations, rising to 46% for large organizations.

Yet, despite this investment, the conversation about artificial intelligence in facility management is stuck on adoption. Most of the organizations within that 28% still have not solved the harder problem: creating simplified, automated workflows to turn reactive and predictive insights into corrective actions that result in real outcomes. Even with more data at their disposal, facility leaders are still making operational decisions the same way they did five years ago.

Collecting data is straightforward. Sensors, platforms, and integrations are available at a number of price points and levels of complexity. Operationalizing this technology, so the right person gets the right alert and knows what to do with it, is still hard.

In this two-part series, we look at what it actually takes to close that gap. Part one lays out a framework for how facility leaders can move from siloed data to decision-ready intelligence, and how to put that framework into practice. Part two [LINK] shifts from the administrative side to the physical side, examining which facility tasks are ready for automation and robotics today and which still need a human in the loop.

The gap between procurement and performance

Ask facility leaders why they want AI, and the honest answer is often not operational. “Even facility leaders who are running an incredibly well-oiled machine say they need AI," says Anjali Bivek, Senior Manager of Innovation. "Every leader out there, for the most part, does not want to be left behind. There’s an underlying fear that unless a facility ‘adopts AI,’ it will be stuck with a horse and buggy while everyone else has a really fast car."

That fear often translates into a procurement exercise. The instinct is to buy another sensor, another dashboard, another integration, sometimes before anyone has defined what problem it needs to solve.

But a sensor only reports a condition. Turning that report into a decision, and that decision into a completed work order, depends on things no purchase order covers: who is authorized to act on the alert, how quickly they are expected to respond, and how the outcome feeds back into the system so the next recommendation is sharper than the last.

More than 60% of organizations are unsure if they have the data management practices in place to support AI. Gartner predicts that, through 2026, organizations will abandon 60% of AI projects because the data behind them was never ready to support AI in the first place. The right-sized budget buys sensors, platforms, and integrations. It does not buy the operational discipline to act on the data these technologies generate.

"Technology for technology's sake is not going to help," Bivek says.

A better approach to operationalizing AI starts with a different question. "Clients are usually describing an outcome they want, not the technology," says Rachel Njiru, Director, Enterprise AI Enablement. "You will hear the word AI, but if you get into the details of what they actually want, the goal is fewer failures on their equipment, better management of energy consumption, or more consistent service."

Reaching those goals means pinpointing which operational decision needs improvement and implementing the simplest, most responsive solution that delivers the outcome.

Reaching AI maturity

Once an organization has identified an outcome AI can help achieve, the next step is making sure the underlying data is ready to support it. Not all data is useful data. Visibility often gets mistaken for readiness, and a dashboard often gets mistaken for a decision. A more useful way to assess where an organization stands is to think in three stages, since the fix required at each one is different.

Stage one: Siloed data

A building management system tracks equipment alerts. A CMMS tracks work orders. A separate system tracks who is clocked in and where. Financials, timekeeping, and supply ordering live somewhere else entirely.

The systems exist, but they do not talk to one another. Answering a basic operational question, such as whether a wing's restrooms were serviced on schedule, means checking one system to confirm who was clocked in, another to see what work was assigned, a third to verify it was completed correctly, and a paper log to check whether anyone was actually in that space at the right time. Most organizations, including sophisticated ones, operate here.

Stage two: Connected data

The systems are unified into a single view. This stage is where many organizations mistake visibility for transformation. A dashboard can look impressive without being reliable, and a beautiful interface is not the same as a working operation, says Njiru.

The relevant questions at this stage are about ownership: who is responsible for acting on what the dashboard shows, and does the insight actually translate into work getting done, or does it just sit there as data people look at? Without clear answers to those questions, connected data is still just data. It has not become decision-ready.

Stage three: Decision-ready data

In this stage, the system does not just surface information. It surfaces the right alert, to the right person, at the right time, and then closes the loop on what happened as a result.

For example, if quality scores show a conference room is being over-serviced, the system does not just report that status. It recommends dropping the cleaning frequency from five times a day to three and reallocating that labor to a space with lower quality scores and heavier use. A site supervisor approves the change, and it pushes down to the whole team. The alert does not just inform. It tells someone what to do next, and with which resources.

Very few organizations operate consistently at this stage. It requires redesigning how work gets assigned and executed, not simply adding another dashboard on top of an unchanged workflow.

Operational constraints also need to be accounted for. Alert fatigue is a genuine risk. A system that generates more signals than a team can reasonably act on trains people to ignore it. Staffing models built around scheduled rounds, not real-time response, cannot always accommodate an alert the moment it fires.

These constraints are solvable with deliberate design. The harder work, and the reason most organizations stall before reaching this stage, is getting the underlying data connected and trustworthy enough to act on in the first place.

What AI-driven intelligence actually looks like

ABM Connect^TM shows what it looks like for a facility to reach stage three. It connects frontline data, centralized systems, and proactive alerting into a single loop, with clear ownership over who acts on what.

Before the platform existed, an operator trying to answer a basic operational question had to work across systems that were never designed to talk to each other, describes Manas Malik, Senior Product Manager, ABM Connect.

One system showed who was clocked in for the day. Another showed what work had been assigned. A third showed whether that work had actually been completed correctly. Paper logs, kept behind restroom doors and on clipboards, showed whether team members had been in the right spaces at the right times. Those logs often did not match what the other systems reported.

ABM Connect pulls frontline task data, building system alerts, and space utilization data into one place, pairing it with a mobile app that lets team members document issues as they find them. A janitorial team member walking a floor can flag a stain outside their scope, attach a photo, and generate a work order on the spot.

Engineering technicians get a different version of the same discipline. A rooftop HVAC unit showing a subtle pressure variance, one that would not have been significant enough to trigger a work order under the old model, gets flagged by the predictive maintenance platform before it becomes a failure.

The results show up in both quality and cost. At one client site, Malik notes, janitorial tickets filed by occupants dropped 22% over an eight-month period, a meaningful improvement in service quality. On the engineering side, a fault detection and diagnostics layer reads live data directly from building systems to catch inefficiencies. At one site, that combination of predictive maintenance and smarter capital planning produced a net savings of $180,000 in the first four months alone.

Crucially, AI never replaces the facility workforce. It replaces the hours spent digging across disconnected systems just to understand what is happening in a building.

Next steps

Closing the gap between adoption and operationalization does not start with a new purchase. It starts with understanding how the facility is actually using the data and systems it already has. A few questions are worth asking honestly before adding anything new.

• Who currently receives your maintenance alerts, and what authority do they have to act on them without further approval?
  An alert that requires three layers of sign-off before anyone can respond is not decision-ready, no matter how sophisticated the system generating it.
• When was the last time your data actually changed an operations decision?
  If the honest answer is rarely or never, your facility likely has visibility, not intelligence.
• If your building management systems disagree about the condition of a space, which one wins, and why?
  If there is no clear answer, the systems are still siloed in practice, even if they technically share a dashboard.

The answers to these questions reveal whether your facility has the operational discipline to act on what it already knows, not just the tools to collect more of it.

Getting data connected and decision-ready is the first part of the equation. The second part is physical. Once an organization knows what needs to happen, the next question is whether a person, a machine, or some combination of the two can do the work. That is where part two of this series picks up: which facility tasks are ready for automation and robotics today, and which still need a human in the loop.

Read the next article in the series, “What LaGuardia's robotic dog says about automation readiness.”