AI in FM, Part I: You Can't Shove AI at Chaos and Expect It to Make Things Better
Everybody and their mother wants to put AI in facility management right now. I get it. I use AI every day. But there's a conversation happening in our industry that's skipping about four critical steps, and it's going to cost people real money.
Here's the thing about AI in facility management: it's only as good as the data underneath it. And I say this as someone who's walked through more facility rooms than I can count across my career. Nearly a decade at Salesforce, where I founded and scaled the RETECH organization from one person to over thirty. Before that, twenty years as an IT guy. I've seen every generation of technology get thrown at buildings, and the failure pattern is always the same.
The technology works. The data doesn't.
Predictive Maintenance Is Real. The Path to Get There Is What's Fake.
The way connected devices and predictive maintenance gets sold to facility management teams is, frankly, dishonest. Here's the pitch: "Install sensors. Connect to AI. Predict failures before they happen. Save 25-30% on maintenance costs."
That’s only half of the story. The uncomfortable truth about AI in facility management is that it doesn't replace the need for good fundamentals — it amplifies whatever's already there. Clean data gets cleaner. Bad data gets more confidently wrong. AI slop is the new garbage in, garbage out.
The Leak Sensor Problem
Let me give you a real example. At a company I worked at previously, we installed leak detection sensors in the bathrooms and kitchens. State of the art. Connected to the building management system. AI-driven alerting.
They triggered constantly.
The response team would drop everything, run to the alert location, and find... the janitor. Mopping. Every single time. The cleaning crew would go through their rounds, water would hit the floor, and the sensor would scream "LEAK!" at anyone who would listen.
Now, any facilities manager in the building could have told you that Jose mops the third floor bathrooms at 6:30 AM every Tuesday and Thursday. That knowledge isn't in any training dataset. It isn't in the BMS. It's in somebody's head — and if you don't capture it before you layer AI on top, you're going to spend a lot of money chasing a guy with a mop bucket.
That's what I mean by shoving AI at chaos. The sensor worked perfectly. The AI worked perfectly. The process failed.
Data Governance Before Intelligence
At sonpito, we have a phrase we keep coming back to: data governance before intelligence. It's borrowed from IT service management, which figured this out twenty years ago. Track everything first. Make it effortless to track. Then measure. Then optimize. Then — and only then — layer on intelligence.
The IT world went through the exact same hype cycle. Remember when every help desk was going to be replaced by chatbots? Some of them were. Most of them weren't. The ones that worked had clean ticket data, documented processes, and a knowledge base that was actually maintained. The ones that failed had none of that, and spent six months training a chatbot on garbage.
Facility management is at that same crossroads right now. And the vendors selling you AI-powered everything are not going to tell you that you need to spend six months cleaning up your data first. Because that's not a fun demo.
Where to Actually Start
If you're an FM professional reading this and feeling like you're behind on AI — don't be afraid. You're not behind. You're actually in a good position if you're willing to do the unglamorous work first.
Get your asset inventory right. Every piece of equipment, documented. Where it is, what it does, when it was last serviced, who's responsible for it. If that information currently lives in a spreadsheet, a whiteboard, and three people's heads, that's your first problem to solve.
Then get your reactive maintenance process clean. When something breaks, there should be a case created in under thirty seconds — on a phone, while walking, because that's how FMs actually work. The fix should be documented. The vendor interaction should be tracked. When you've got that humming, you'll have data worth analyzing.
That's when AI gets reliable. Not before.
So before you sign that contract for that AI-powered predictive maintenance platform, ask yourself: can I see the complete maintenance history of any asset in my building right now, in under sixty seconds? If the answer is no, you've got more foundational work to do. And that's not a failure. That's where almost everyone is. The question is whether you're honest about it.
Fawn and I built sonpito specifically for that foundational layer. Not because it's flashy, but because it's what actually needs to happen before anything else works. Simple tools that your team will actually use.
AI is coming to FM. It should. But let's build the foundation first.
