AI Dispatching for HVAC: Routing By Skill, Stock, and Zone

Why dispatch is the highest-leverage operations lever in HVAC

Every HVAC owner has watched the same scene play out. It is 9:47am, the board is already two stops behind, the no-cool call that came in at 7:15 is sitting in zone 4 with a tech who specializes in commercial RTUs heading to it, and the heat-pump-certified tech twelve minutes from the customer is forty miles away running a tune-up that could have been any tech on the truck. By the time the rebalance happens it is 11:30 and the day is gone.

That morning is the business case for AI dispatching. Dispatch is the workflow where capacity is created or destroyed, and on most mid-market HVAC boards a measurable share of capacity is destroyed every day by routing decisions a smart engine would make differently.

This guide walks the mechanics, the tools, and the implementation pattern.

What a smart dispatch engine actually does

A naive dispatcher orders by earliest-available technician. A smart AI dispatcher scores every candidate stop against every available tech on four dimensions, re-scores when conditions change, and surfaces the top assignment to a human dispatcher who keeps final say.

Dimension 1 — tech specialty match

HVAC techs are not interchangeable. A 12-year heat-pump installer is a different asset than a strong residential service tech who has never touched a mini-split. A commercial RTU tech and a boiler tech overlap less than the schedule pretends. The engine reads the inbound job's system type, the customer's equipment history from the FSM, and the symptom keywords from the call notes, then scores tech-specialty match on a 0–100 scale per candidate.

Dimension 2 — parts-on-truck probability

Each truck carries a different parts mix. The engine reads the symptom and the historical first-time-fix rate for that symptom against each truck's typical inventory. If the tech with the strongest specialty match is also 78% likely to have the part, they are the right assignment. If they are only 35% likely, the engine may surface a slightly weaker specialty match with 90% parts likelihood because the visit will close on the first call rather than rolling a return trip.

Dimension 3 — zone density and drive time

Route optimization is the table-stakes layer. The engine clusters stops geographically, accounts for live traffic, and proposes the sequence that minimizes drive time across the day rather than per-call. The difference between locally-optimal (per call) and globally-optimal (per day) routing is typically 22–28 minutes a tech across an 8-hour shift.

Dimension 4 — SLA pressure and customer priority

Maintenance-agreement members get priority. No-cool customers without conditioned air get priority. Commercial accounts with SLA terms get priority. The engine encodes these rules and refuses to send a member to the back of the queue because a transactional caller dialed in earlier.

The metrics that move

Three numbers tell you whether the dispatch engine is working.

Revenue per truck per day. The cleanest indicator. Pre-AI baseline for a mid-market residential HVAC shop is $1,650–$2,100 per truck per day. A well-tuned dispatch engine adds 11–17% inside the first 60 days, from one additional stop per tech per day and higher average tickets on stops where specialty matching enables an upsell the wrong tech would have missed.

First-time fix rate. Industry baseline sits at 71–78% for residential service calls. Parts-on-truck-aware dispatching lifts this to 84–89%, which is the single largest CSAT driver in the business and the input to review velocity that nothing else moves.

Drive time as a share of paid hours. Pre-AI: 28–34%. Post-AI with a tuned engine: 19–22%. That is recovered productive labor at fully-loaded tech cost — a real cost-out line in addition to the revenue lift.

Tools an HVAC operator should evaluate in 2026

ServiceTitan Dispatch Pro

The deepest FSM-native dispatch AI. Strong on specialty matching once tech-skill profiles are populated, weaker out of the box on parts-on-truck because most shops do not maintain truck inventory tightly enough in ServiceTitan. Best-in-class for shops already on ServiceTitan.

FieldEdge intelligent scheduling

Strong for 2–8 truck shops. Less granular than ServiceTitan on specialty scoring but easier to configure.

Sera

Dispatch and revenue-per-stop specialist used by larger residential shops. Best fit over 12 trucks.

Routific or OptimoRoute on the FSM API

For shops that want best-in-class route optimization without replacing the FSM dispatcher. Requires more engineering setup.

Successware + middleware

Successware shops need a middleware layer for modern dispatch UX. Budget an extra 30 days.

The implementation pattern that works

A staged rollout. Three weeks, three milestones.

Week 1 — data foundation

Audit the tech-skill profiles in the FSM. Most shops discover that profiles are 40–60% accurate — Jim is listed as heat-pump-certified because he took the course in 2018 but has not installed one in two years; Maria is not listed as commercial-capable but has been running commercial calls all summer. Fix the profiles. Audit the truck-inventory records. If they do not exist, build a typical-inventory baseline by truck. Without clean inputs the engine produces clean-looking but useless outputs.

Week 2 — shadow mode

Run the engine in shadow. It scores every stop and proposes an assignment; the human dispatcher continues to assign as they always have. Compare the engine's proposals against the human's assignments at the end of every day. Capture every disagreement and ask why. About 65–75% of disagreements turn out to be engine-correct — the human dispatcher was working from incomplete information. The remaining 25–35% reveal rules the engine needs to encode (this customer always wants Mike; this neighborhood routes badly after 3pm; this commercial account has a no-tech-named-Steve preference).

Week 3 — partial cutover

The engine assigns; the human dispatcher reviews and overrides. Override rate starts at 30–40% and drops to 8–12% by end of week three as the rules get encoded. Once override rate is under 15% consistently, the engine is the primary; the human is the supervisor.

Pitfalls that kill dispatch deployments

Bad tech-skill data

Single largest failure mode. If the profiles are wrong, the engine optimizes for the wrong objective. Fix profiles before turning on the engine.

No parts-on-truck data

Engines that cannot see parts inventory route by specialty alone and miss the first-time-fix lift. If your FSM does not track truck inventory, this is the cleanup task to do before the dispatch project starts.

Ignoring dispatcher institutional knowledge

Senior dispatchers know things that are not in the FSM — which customer fights with which tech, which neighborhood has a parking nightmare, which commercial account requires the safety briefing. Capture these as configuration rules during shadow mode. Otherwise the engine will violate them and erode trust.

Optimizing per-call instead of per-day

Greedy per-call dispatching often beats local optimization but loses globally. Confirm the engine optimizes the full day's route, not the next stop.

Failing to re-run mid-day

Conditions change. Cancellations, no-shows, emergencies, traffic. The engine has to re-run every 30–45 minutes against live state, not just at 7am.

FAQ

How does AI dispatch differ from route optimization?

Route optimization minimizes drive time within an already-assigned set of stops. AI dispatch picks which tech gets which stop in the first place, then layers route optimization on top. You want both; they solve different problems.

Will the dispatcher lose their job?

No. The dispatcher's role shifts from per-stop decision-making to exception handling, customer-facing communication, and rule maintenance. Most shops report the senior dispatcher becoming dramatically more effective once relieved of the per-stop assignment grind.

How much capacity does a mid-market shop actually recover?

0.7–1.2 additional revenue-producing stops per truck per day. For an 8-truck shop, that compounds to ~1,800 additional stops a year. For the dollar math, see the HVAC AI ROI breakdown.

What about commercial HVAC dispatch?

Commercial dispatch is structurally different — lower call volume, higher complexity, account-manager involvement. The engine helps, but the rules layer is heavier. Plan for an extra week of configuration.

Does the dispatch engine integrate with the AI receptionist?

Yes — and it should. The receptionist captures qualification data; the dispatch engine reads it and proposes the right tech. The two together is the workflow you want; for receptionist context see the AI receptionist for HVAC guide.

What is the realistic cost?

ServiceTitan Dispatch Pro is included in the higher ServiceTitan tiers; standalone overlays like Routific run $200–$600 a month for a mid-market shop. Implementation cost — primarily the data-cleanup work — is typically $4k–$12k.

For the full operator playbook including where dispatch sits inside the bigger AI stack, see AI for HVAC contractors: the 2026 operator playbook.

Ready to scope dispatch for your shop? Visit AI for HVAC contractors for the operator context, or book a 30-minute board audit and we will benchmark your tech-skill data, parts-on-truck records, and current revenue per truck per day against the levers in this guide. The bigger frame lives at AI enablement.

SOURCES

Cited and consulted.

1. 01ServiceTitan Dispatch Pro — Product Overviewservicetitan.com · accessed May 8, 2026
2. 02FieldEdge Operations — Scheduling and Dispatchfieldedge.com · accessed May 8, 2026
3. 03ACCA Operating Benchmarks — Dispatch and Productivityacca.org · accessed May 8, 2026
4. 04HVACR Business — Dispatch Operationshvacrbusiness.com · accessed May 8, 2026