How to Use AI Forecasting for Field Operations to Stop Reacting and Start Planning

Weather forecasts are pretty accurate these days. The hard part for a company like TR Miller (a residential HVAC, plumbing, and electrical company in Chicago) isn't predicting whether it'll hit 90 degrees next Tuesday. The hard part is predicting what that means for the business. How many calls will come in? How many techs do will be needed? What should the dispatch board actually look like?

That's exactly what Mark Mullane, director of marketing at TR Miller, tackled in his session AI Forecasting to Automate Dispatch and Staffing in Field Operations at Domo BUILD 2026.

Mark walked through how his team built an AI forecasting app that combines eight years of service history with weather data to predict demand day by day. With it, they get proactive staffing, better technician retention, fewer cancellations, and real revenue impact.

What follows is a practical framework you can adapt to your own field operations.

Start with the decision, not the data

Build your forecasting model backward from the operational choice you're trying to improve

Most forecasting projects fail because they start with "what can the data tell us?" instead of "what decision are we trying to make better?" Mark's team didn't set out to build a weather model. Rather, they set out to fix dispatch planning.

The operational problems they wanted to solve looked like this:

• Last-minute staffing calls that hurt technician morale and work-life balance
• Pre-booking spikes and crashes that made maintenance scheduling unpredictable
• Marketing playing catch-up instead of getting ahead of demand
• Cancellation rates climbing when they couldn't handle incoming calls

Once you're clear on the decision you're improving, you can work backward to figure out what inputs actually matter. For TR Miller, that meant connecting weather patterns to call volume at the calendar-day level.

Build your data recipe for demand forecasting

Join multi-year service history with weather APIs to model the relationship between conditions and calls

Mark described the core data setup this way: "We have eight years of data of how many repair calls we ran on each calendar day as well as tying it into then weather APIs to pull in what the weather was on that day." That historical depth lets the model learn patterns that aren't obvious at first glance.

Here's what makes this approach work:

• Historical call counts by calendar day (not just monthly or weekly aggregates)
• Weather API data matched to those same days
• Enough years of history to capture seasonal variation and edge cases

One modeling nuance Mark highlighted is that context changes everything. As he put it, "80 degrees isn't just 80 degrees." An 80-degree day in May triggers first-time AC startups and a spike in breakdowns. The same temperature in August barely moves the needle. Your model needs to account for when in the season a weather event hits, not just what the temperature is. Weekends and holidays also shift behavior, so factor those in too.

Turn forecasts into dispatch board actions

Shift work types based on predicted demand to keep technicians productive in any conditions

A forecast that sits in a dashboard doesn't change anything. The value shows up when you actually adjust what's on the dispatch board based on what's coming.

Mark shared a specific example from this past June. Chicago had an unusually cold stretch for that time of year, with temperatures in the 60s and 70s. ACs weren't running, which meant they weren't breaking. Instead of leaving technicians idle or sending them home, TR Miller loaded up the board with inspection calls (preventative maintenance visits). That kept techs busy, generated revenue, and got customers' equipment checked before the inevitable heat arrived.

The key actions that flow from a good forecast include:

• Shifting the mix between repairs and inspections based on expected demand
• Pre-booking maintenance calls during predicted slow periods
• Pulling back on marketing spend when you're already at capacity
• Pushing marketing when you see a slow week coming

This kind of flexibility requires buy-in from dispatch, marketing, and operations. The forecast becomes a shared planning tool, not just a report.

Generate staffing recommendations that protect your team

Use predicted call volume to have scheduling conversations early, not the morning of

The staffing piece is where forecasting really pays off for your people. Mark explained that TR Miller uses the forecast to generate staffing requirements for each day, then has those conversations with technicians at the start of the week instead of the morning of.

The difference matters. When you call a technician on Saturday morning asking them to work, you're creating stress and resentment. When you tell them on Monday that Saturday looks busy and ask if they can work, you're giving them time to plan. That respect for work-life balance compounds over time.

Mark tied this directly to retention: "For our HVAC service team, we are at over an 80 percent retention rate and a huge portion of that goes out to work-life balance." In an industry where experienced technicians are hard to find and harder to keep, that number speaks for itself.

Measure ROI with ops-native KPIs

Track cancellation rates, retention, and revenue capture instead of model accuracy alone

It's tempting to evaluate a forecasting model by how accurate its predictions are. But accuracy doesn't pay the bills. What matters is whether the forecast changes behavior in ways that improve outcomes.

TR Miller tracks two KPIs that directly connect to forecasting quality:

• Cancellation rate: When you can't handle incoming demand, customers cancel. Last June, TR Miller's cancellation rate sat around 18 percent. This year, with better forecasting and preparedness, it dropped to 14 percent. Mark noted that four-point improvement translates to "dozens of more calls that we're able to run and tens if not hundreds of thousands of more revenue."
• Technician retention: Stable scheduling and predictable workloads keep people around. That 80-plus percent retention rate isn't an accident.

These metrics tell you whether your forecasting investment is working in ways that a mean absolute error calculation never will.

Putting it all together

The framework Mark shared at Domo BUILD 2026 gives you a clear path: start with the operational decision, build a data recipe that connects historical service volume to an external driver like weather, translate forecasts into dispatch board changes and staffing recommendations, then measure success with KPIs that matter to the business.

Mark also touched on expanding this approach to other trades (plumbing, electrical) and other weather-driven signals like rainfall. The same logic applies wherever external conditions drive demand.

If you're running field operations and tired of reactive scrambling, AI forecasting for field operations offers a way out. Watch the full session to see Mark walk through the app and hear the details firsthand.