Same-day dispatch for HVAC and plumbing: how AI scheduling actually works

Same-day dispatch is the highest-impact promise a home-services business can make. "We'll be there today" closes more service calls than any marketing copy ever has.

It's also the easiest promise to mess up. Promise it, miss it, and the customer leaves a 1-star review that costs you the next 40 customers. Most of what looks like "scheduling software" is really just a calendar with text alerts; the actual logic that makes same-day dispatch work is below the UI.

The four moving parts

Same-day dispatch is one optimization problem made of four sub-problems. Get any one wrong and the whole thing breaks.

1. Geo-clustering

You don't dispatch the closest tech to each job in isolation. You cluster jobs by neighborhood and route them in sequence. A truck that does five jobs in a 6-mile radius is worth two trucks doing five jobs across a 30-mile radius. The dispatch software should be solving a small traveling-salesman problem every time a new job lands, not just sorting by "next available slot."

2. Skill matching

Not every tech can do every job. A junior tech can handle a routine service call but not a panel rebuild; a plumbing tech can't take an HVAC charge call. The dispatch logic has to filter by skill before it filters by proximity.

3. Truck stock

The single biggest cause of "we couldn't finish the job today" is the right tech showing up with the wrong parts. The scheduling system should know what's on each truck — a kitchen faucet replacement isn't actually same-day if the tech doesn't have the cartridge in stock. Real same-day dispatch software either tracks truck stock or warns you when it can't.

4. Promise math

The hardest part. Every booked job has a promised arrival window ("between 10am and 2pm today"). New emergency calls have to fit into that schedule without breaking the existing promises. The dispatch engine needs to know:

• How long does the new job take, with a confidence band?
• What's the buffer between currently-scheduled jobs?
• Which existing job has the most slack in its window if reshuffling is needed?
• Which customers explicitly opted into "we may call to reschedule" vs. which are locked?

Where AI helps

The AI win in scheduling isn't "the AI picks the best truck." Linear programming has solved that since the 1960s. The AI win is everything around the dispatch call itself:

Customer intake. The AI receptionist asks the right triage questions in the right order, then captures the address, problem, urgency, and access notes — all the inputs the dispatcher would normally pull out of the customer one piece at a time.

Triage classification. The AI listens to the customer's problem description and classifies it: emergency vs. routine, single-fixture vs. main-line, parts-on-truck vs. parts-order. This drops on the dispatcher's screen pre-categorized so they're not re-asking the customer.

Conflict notification. When a new emergency call comes in and there's a scheduling conflict with an existing job, the AI handles the outbound call to the existing customer: "We're running about 45 minutes behind, would 3:30 instead of 2:45 work?" This is the highest-ROI use of AI in the dispatch flow because it's the most painful task for human dispatchers — apologizing for delays one customer at a time.

ETA texts. "Your tech (Mike, blue F-150) is 18 minutes away." These cut no-shows on the customer side (the homeowner who left for the grocery store and didn't get back in time) by ~40% in our data.

Where AI doesn't help (yet)

The actual dispatch decision. The optimization math is well-understood. AI-vs.-linear-programming benchmarks on small dispatch sets (under 50 jobs/day) show no meaningful improvement from AI; the right answer is usually the obvious one.

Stock estimation. The AI can flag "this job likely needs a cartridge" but it can't tell you what's actually on the truck. That's a truck-stock-tracking problem, not an AI problem.

Same-day promise calibration. The AI doesn't know whether your dispatcher has been over-promising for three weeks running. A scheduling tool that lets you set explicit "no more than 8 promises/day" guardrails matters more than any AI feature.

What good looks like

A scheduling system that:

• Pulls intake from the AI receptionist with full triage and address pre-filled
• Suggests the dispatch slot but lets the dispatcher override
• Auto-fires the ETA text 20 minutes before arrival
• Handles the conflict-reschedule outbound call (with dispatcher approval before sending)
• Tracks truck stock against the parts catalog
• Logs the actual on-site duration so future promise windows calibrate to real data

FAQ

Does AI scheduling replace a dispatcher? No. It removes the busywork (intake transcription, ETA texts, conflict-reschedule calls) so the dispatcher can focus on the judgment calls. For most operations the right ratio is one dispatcher per 4-6 trucks, AI handles ~60% of dispatch-adjacent work.

Can AI scheduling handle emergency calls? Yes — the AI receptionist triages and routes to the on-call tech. The actual dispatch decision still flows through the dispatcher (or after-hours, the on-call lead).

What about route optimization? Most small operations (under 8 trucks) don't need full route optimization; geo-clustering by zone is enough. Larger fleets (15+ trucks) benefit from dedicated routing tools.

Does Pylor handle dispatch? Pylor's receptionist (Paige) does the intake, triage, and customer-facing reschedule calls. The actual dispatch UI lives in /scheduling. Both ship in the Workforce tier.