What Happens When It's -3 Degrees and Everyone's Furnace Breaks

When Geoff Farinha got in his car the morning we talked, it was negative 3 degrees. February weather before Christmas. Single digits across Massachusetts.

"Great for business," he told me. "Tough for everyone's mental health."

When the temperature drops like that, everyone's HVAC system gets pushed to its limits. The marginal equipment fails. The calls flood in. And companies like USI HVAC have to figure out how to respond.

This is the demand planning problem that every service business faces in some form. How do you prepare for spikes you can't fully predict? How do you staff appropriately without overspending on labor you don't need?

The Manual Approach

USI always has an on-call technician. But during extreme weather, one isn't enough.

"We have a service manager. He'll work directly with the on-call schedule that he makes a year prior, but also his key technicians. We'll see if anyone can be on standby or on backup. Sometimes we'll have two or three. This past summer we had situations where we had four on-call technicians. I've never seen that."

The schedule gets built a year in advance using historical weather data. They check Weather Underground, look at what happened this time last year, and make educated guesses about when demand will spike.

Then they layer in technician skill levels. Some of their people are strictly construction—they can handle service, but only in a pinch. You don't want to put them on call during the coldest week of the year. So maybe you schedule them for milder weeks and save your most skilled service techs for when things get brutal.

"You do the best you can to be completely honest with you. You can definitely overprepare and you can definitely underprepare. I think you're never perfect. It's always 'I wish I had one more' or 'maybe we have too many.' But you just do the best you can."

The Data Opportunity

Here's where my brain started spinning. I spent 18 years in signals intelligence, and our entire job was turning data into predictions. When Geoff described this manual process—checking historical weather, building schedules a year out, adjusting on the fly—I immediately saw the data problem underneath.

What if you could automate the correlation?

Geoff's on the same wavelength. When I asked about using AI for demand planning, he didn't hesitate:

"I would love that. Right now it's a very manual process for us."

But then he went somewhere I hadn't thought of:

"If something could plug into Service Titan and say these customers last year you had this many after-hours calls with. But on top of that, to take equipment—this customer has equipment with an average age of 25 years, well past its lifespan—be prepared for potentially more calls because the equipment's older. Something like that would be fantastic."

Weather plus service history plus equipment age. Three data streams that together could predict demand far better than any one of them alone.

Why This Matters

The difference between being prepared and being caught off guard is the difference between profit and chaos.

If you're understaffed during a cold snap:

Customers wait too long and get frustrated
You burn out your available technicians
You might miss calls entirely and lose the customer
Your reputation takes a hit

If you're overstaffed during a mild week:

You're paying people to sit around
Your margins suffer
You're less competitive on pricing

The companies that can predict demand accurately can staff appropriately. They don't over-hire and they don't get caught short. That's a competitive advantage that compounds over time.

What the Data Could Do

Imagine a system that combines:

Weather forecasts - Not just temperature, but the rate of change. A sudden drop from 40 to 10 stresses systems more than steady cold.

Historical service calls - Which buildings called last winter? Which equipment types failed? What were the conditions?

Equipment profiles - Age, maintenance history, manufacturer reliability data. A 25-year-old unit is more likely to fail than a 5-year-old unit.

Customer segments - Some customers call at the first sign of trouble. Others wait until it's catastrophic. Understanding behavior patterns helps predict volume.

Now imagine that system flagging: "Next week's forecast shows a 40-degree temperature swing. Based on historical patterns and current equipment profiles, expect 3x normal service call volume. Recommend adding two on-call technicians Wednesday through Friday."

That's not science fiction. That's just connecting data that already exists in Service Titan, weather APIs, and basic equipment records.

The Human Element

But here's what Geoff understands that pure data doesn't capture: the human element of on-call work.

You can't just schedule anyone. You need people who are skilled enough to handle emergencies. You need people who are willing to be on call. You need to balance the burden so the same people aren't always getting the 2 AM calls.

"There's only a limited labor pool to pull from," Geoff noted. And that labor pool has preferences, families, and limits.

The data can tell you how many people you need. It can't tell you who should be asked. That's still a leadership judgment call—one that requires knowing your people as individuals, not just as headcount.

The Bigger Picture

Every service business has some version of this problem. Maybe it's not weather—maybe it's seasonal demand, or event-driven spikes, or economic cycles. But the pattern is the same: demand is variable and staffing is fixed, and the gap between them is where you either make money or lose it.

The companies that figure out how to predict and prepare will outperform the ones that react and scramble. The data exists. The tools to process it are getting cheaper and more accessible. The question is who's going to connect the dots first.

Geoff's doing it manually today because the automated solution doesn't exist yet—at least not in a form that plugs cleanly into his operation. But he knows exactly what he'd want it to do.

That's usually how it works. The people doing the job understand the problem better than anyone. They're just waiting for the tools to catch up.