AI for dispatch in plumbing: what to automate and what stays with the dispatcher

The 10:40 board looks fine until it doesn't. One tech is still on a water heater swap that was supposed to end at 10:30. A drain call is waiting for a 11:00 arrival. Another tech has a gap but the wrong skill set for the job, and the customer on the 1:00 slot has no idea anyone is running behind.

AI for dispatch in plumbing means software watches the board, the job files and each tech's day, and does the legwork around it: suggesting which tech fits a call, flagging jobs that are running late, drafting the heads-up text, and confirming tomorrow's schedule tonight. The dispatcher keeps the final say on every move. This page covers what that looks like, what your dispatch software may already do, and where an agent adds something.

What a plumbing dispatcher is juggling

Dispatch is a set of small matching problems that change every hour. Which tech has the skill for this job, in this zone, with this much time left in the day. How urgent it is. How far away the truck is. Whether the job in progress will finish on time. Whether the customer who was told "between 12 and 2" still knows that.

Most of this happens in the dispatcher's head and in a group chat. That works until the day gets busy, which is exactly when it matters.

What your software already automates

Check this before you pay for anything. ServiceTitan's Dispatch Pro, as one example, runs every 10 minutes during business hours and hourly outside them, and it raises alerts when it can't place a job well. The list includes Arriving Too Late (it can only schedule the job late), Max Drive Time (the job is too far from the current or previous job), No Availability (all skilled technicians in the right zone are busy), Urgent Job Conflict, No Matching Technicians (no one has the required skill set) and No Shifts. Each alert comes with a suggested dispatcher action, such as assigning manually, extending the arrival window or fixing the job's required skills.

ServiceTitan's documentation also says that recurring alerts of the same type point to a settings issue, not a one-off job problem, and that an administrator should review them. That is a useful rule for any dispatch tool: an alert that keeps firing is telling you something about how the system is set up.

The Daily Dispatch Board adds a visual layer. Appointments where the technician is late are outlined in red once board alerts are configured, hovering over a technician shows memos and skills, and appointments are colored by status from scheduled through confirmed to done. Dispatchers drag appointments between technicians to reassign them.

Other field service tools have their own versions of this. If yours already flags a late job and suggests a tech, an agent should not duplicate it.

Four jobs an AI agent can take on

These are the gaps between what the board shows and what a dispatcher has time to chase.

1. Suggest the right tech for a new call. The agent reads the job type, the zone, each tech's skills and remaining time, and proposes one or two options with the reason. The dispatcher picks.
2. Flag a late job and draft the heads-up. When a job is running past its slot, the agent notices before the next customer does, and drafts a text for the dispatcher to send.
3. Confirm tomorrow's board tonight. It lists unconfirmed stops, unassigned visits, overbooked days and thin days, so the morning does not start with a surprise. Night-before confirmations are nobody's job in many offices, which is why they don't happen.
4. Keep the job file true. If the tech sat in traffic and the file still says on time, tomorrow's plan is built on the wrong numbers. The agent compares scheduled and actual times and notes the difference.

Cortex describes this pattern on its dispatch and schedule page: the agent sees the risk and drafts the update, and a dispatcher still runs the day.

A worked example (illustration)

This is a made-up board, not a client.

Tuesday, 10:40. Marcus is on a water heater swap booked until 10:30 and now expected to run until noon. The 11:00 drain call (Alvarez, 22 Birch) won't make it. Dana has a 45-minute gap and the drain skill, 12 minutes away. The agent suggests moving Alvarez to Dana, and drafts a text to the 1:00 customer (Kim) saying Marcus may arrive 20 minutes late. Both wait for the dispatcher.

The agent never moved anything. It saw the problem before a customer had to call and ask where the tech was, and put a decision in front of a person with the work already done.

What stays with the dispatcher

• Which emergency wins. A burst pipe and a slow drain are not the same call, and the rule for ranking them is yours, not the model's.
• Bumping a customer. Moving someone's appointment is a promise being broken. A person makes that call.
• Arrival promises. The agent can suggest a window. The dispatcher commits to it.
• Anything unsafe. If a tech reports a gas smell or flooding, the dispatcher takes over.
• Overriding the suggestion. If Dana is the right call for a reason the data does not show, her dispatcher knows it.

Why speed matters at the front of dispatch

Dispatch starts when the phone rings. Invoca's 2026 home services benchmark found a person answered 52% of inbound home services calls, so a lot of dispatch problems start before there is a job to dispatch. See our guide on missed calls for that half of the problem.

On the homeowner's side, Housecall Pro's 2026 survey of more than 1,100 US households, reported by ACHR News, found that 72% would pay more if their emergency were resolved within 24 hours. Customers with a leak are not shopping on price first. They want to know someone is coming.

What it takes to set up

• Read access to the schedule, job types, technician skills and zones. For ServiceTitan that means a registered app, an application key and client credentials generated by an admin.
• Write access limited to drafts. The agent can propose a move or write a text. It cannot send or reassign on its own.
• The dispatcher's rules, written down. What counts as an emergency, how far a tech can drive, which jobs need two people.
• Two weeks of suggest-only. Compare its suggestions with what your dispatcher actually did. Where they differ, find out why.

The Federal Reserve's 2026 small business survey found that time to implement and train employees was named by 37% of firms planning to start AI. Starting with one board and one dispatcher keeps that time short.

Mistakes to avoid

• Automating the assignment before you trust the suggestion.
• Texting customers about delays with wording that promises a new time you cannot keep.
• Skipping the dispatcher when writing the rules. They know which exceptions are real.
• Duplicating alerts your software already raises.

Next step

Look at the last two weeks of late jobs and ask what warning would have helped. If most of the answers are already in your dispatch software's alerts, tune those first. If the answers are "someone should have told the next customer" or "nobody confirmed tomorrow," that is where an agent helps. The plumbing industry page lists the other jobs that slip, the free AI Operations Audit for plumbing ranks them, and the HVAC version of this topic is in AI for dispatch in HVAC.