AI scheduling tools take each job's location, likely duration, required skills and the customer's time window, then suggest the engineer and running order that fit everyone's day with the least driving. When a job overruns, an emergency comes in or someone calls in sick, they reshuffle the rest of the day. A dispatcher approves the changes; the software does the arithmetic.
How well it works depends less on the software than on the data you feed it. Most small firms' job durations are guesses ("a boiler service is an hour") and their skills lists live in the owner's head. A scheduler working from those will produce confident timetables that fall apart by 11am. Fixing durations and skills is the unglamorous first step, and it improves the schedule even before any AI is involved.
What "AI dispatch" is doing under the hood
Most dispatch software solves a puzzle with two kinds of rule. Hard constraints can never be broken: an engineer must hold the right qualification for the job, visits must fall inside the customer's window, nobody works beyond their hours. Soft goals are traded off against each other: less driving, more jobs per day, higher-value jobs to the engineers who close them, the same engineer returning to a regular customer.
The software tries thousands of combinations and picks the one that breaks no hard rule and scores best on the soft ones. Much of this is established optimisation maths rather than the kind of AI that writes text. The "AI" label increasingly covers two extra layers: prediction (estimating how long a job will really take, or how much it's likely to be worth, from your history) and language models working around the schedule (turning a customer's email into a job, texting customers, answering the phone).
ServiceTitan's Dispatch Pro, a paid add-on to its platform, is the clearest example of the prediction layer: it recommends assignments using technician skills and performance, location and drive time, and predicted job value, and lets dispatchers choose anything from on-demand suggestions to fully automatic dispatching. Smaller platforms mostly offer the routing layer, which is where most small firms should start anyway.
The data a scheduler needs from you
Before trialling any tool, get these six things written down. Each one, if missing, causes a specific kind of bad schedule:
| Data | What goes wrong without it | Where to get it |
|---|---|---|
| Real job durations by type | Every day overruns; customers wait | Last 3 months of job start and finish times |
| Skills and qualifications per person | The wrong person arrives, or worse, an unqualified one | Certificates, plus the owner's knowledge |
| Start and finish locations | First and last drives ignored | Home addresses or the yard |
| Working hours and breaks | Schedules that need overtime | Contracts and rota |
| Customer time windows | Arrivals outside promised times | How you book: AM/PM, 2-hour slots, anytime |
| Van stock and parts per job | First visit can't finish the job | Stock lists; parts on the job card |
Durations are where the surprises are. An illustrative six-engineer plumbing and heating firm pulled three months of job records and found its booked durations were off in both directions:
| Job type | Booked as | Actual median | Actual 80th percentile |
|---|---|---|---|
| Boiler service | 60 min | 50 min | 70 min |
| Boiler breakdown | 90 min | 105 min | 160 min |
| Tap or toilet repair | 60 min | 40 min | 55 min |
| Radiator replacement | 120 min | 150 min | 190 min |
The 80th percentile column (the time four in five jobs finish within) matters for breakdowns, where the spread is wide. Booking breakdowns at the median means one in two runs late. The firm booked breakdowns at 150 minutes and services at 55, and the day stopped collapsing before lunch. A chat assistant can do this analysis if you paste in an export of job types with start and finish times, but check its medians with a spreadsheet's MEDIAN and PERCENTILE functions; don't trust figures it calculates in its head.
A dispatch day at a six-engineer heating firm
Before the change, the owner built the next day's schedule every evening and redid parts of it at 7am, around 75 minutes a day. A typical Tuesday had 23 jobs: 9 services, 6 breakdowns, 5 small repairs and 3 installations or replacements.
In a two-week trial, the owner kept scheduling as usual while the software produced its own schedule for the same jobs, a setup often called shadow mode. The comparison for one Tuesday (illustrative):
- Owner's schedule: 5 hours 40 minutes of total driving across six vans; two engineers crossing town twice.
- Suggested schedule: 4 hours 15 minutes of driving, mostly by clustering services by area and giving the two engineers who live north the northern jobs first.
The owner overrode three of the 23 suggestions, and the reasons are instructive:
- A regular landlord customer always gets the same engineer. The software didn't know; the fix was a "preferred engineer" field on that customer.
- A radiator job went to an engineer whose van didn't carry that radiator size. The fix was adding the part to the job card so the tool could check van stock.
- The software put two loft installations on the same engineer on the same day. Technically possible; physically brutal. The fix was a rule of one heavy install per engineer per day.
Each override became a rule, which is the point of shadow mode. By the second week the owner was overriding one or two jobs a day and spending about 20 minutes approving the schedule instead of 75 building it. Over a five-day week that's more than four and a half hours back, before counting the drive time saved.
When the day breaks: re-dispatching
A schedule is only a plan until 8:15am. The value of a good tool shows when something goes wrong and it recalculates the rest of the day in seconds. Three situations from the same firm, and what to check before accepting the new plan:
- An emergency at 10:40. A customer with no heating and a newborn baby. The tool suggested the nearest breakdown-qualified engineer and pushed his two afternoon services to a colleague. Check: were the moved customers promised a specific engineer or time? If so, phone them rather than relying on an automatic text.
- A job overrunning by 90 minutes. A breakdown needing a part. The tool moved the engineer's next job to someone nearby and flagged that a 4-6pm slot would now be missed. Check: is the slot really lost, or can the customer take a later visit? A person asks; the software can't.
- A sick call at 7am. Four jobs to reallocate. The tool spread them across three engineers and suggested moving one non-urgent service to Thursday. Check: are the rescheduled customers told today, and does Thursday have capacity?
A realistic danger in re-dispatching is speed. Accepting a reshuffle in one click without reading it is how a job needing a qualification ends up with someone who doesn't hold it. At the heating firm, a gas appliance job was once reassigned to a newly hired plumber who wasn't yet gas-qualified, because his profile had been copied from another engineer's and still listed the qualification. The office manager spotted it on the approval screen. The fix was a monthly check of the skills list against certificates, and a rule that new starters' profiles are built from scratch.
Where language models help around the schedule
The text-writing kind of AI is useful at the edges of scheduling, where jobs come in and customers need updating.
Turning an enquiry into a job card. Paste a customer email or a voice-note transcript into a chat assistant with a fixed format:
Turn this customer request into a job card for a plumbing and heating firm.
Return: job type (from this list: [list]) | likely duration (from our table: [paste]) |
skills needed | parts to check | customer time constraints | questions to ask the
customer before booking.
If anything is unclear, list it under "questions" instead of guessing.
Request: [paste]An illustrative result for "Our boiler keeps losing pressure and the radiators upstairs are cold, we're in all day Thursday but not before 10":
Job type: boiler breakdown | Duration: 150 min (booked) | Skills: gas-qualified |
Parts to check: pressure relief valve, expansion vessel | Time: Thursday, after 10:00 |
Questions: boiler make and model? any visible leaks under the boiler? how old is it?What needed fixing: nothing in the job card, but the office always asks whether the customer is a landlord, because landlords' jobs are invoiced differently. That question was added to the prompt. Converting voice notes from site into job sheets the same way is covered in turning on-site voice notes into job sheets.
Customer messages. "On my way" texts, reschedule notices and day-before reminders are standard features in most job software. AI helps write the templates so they sound like you, and some tools personalise them per job.
Answering "when is my engineer coming?" AI phone agents can answer that from the schedule. Jobber Receptionist, for example, costs $29 a month for 30 conversations on eligible plans, then $0.79 each, and can book visits, capture requests and take messages.
What the main trades platforms offer
Features change often, so treat this as a starting list to check against each vendor's current pages:
- Jobber: route optimisation in its New Schedule on the Connect, Grow and Plus plans. It can optimise one person, a group or the whole team for up to seven days at once, using real driving distances, but only reorders "anytime" visits; visits with set times stay fixed. It runs on the web version, by admin users.
- Housecall Pro: added route-based scheduling in its May 2026 update, organising each person's day by route, with recurring routes and quicker handling of call-outs. Its CSR AI answers calls and books jobs.
- ServiceTitan: Dispatch Pro, described above, aimed at larger operations with dedicated dispatchers.
A detailed side-by-side is in Jobber, Housecall Pro and ServiceTitan AI features compared. For a firm with one to three vans, a route-optimise button plus good durations often captures most of the benefit. The daily route itself, once jobs are assigned, is covered in AI route planning for small fleets and mobile teams.
The two-person version: an electrician and an apprentice
At the smallest scale, "AI dispatch" is mostly a question of whether the apprentice goes with the electrician or works alone on simpler jobs. Take a two-person electrical business, invented for this example, with 11 jobs for a Wednesday: two consumer unit replacements (which need both of them), five small fixes the apprentice can do supervised by phone, three inspections and a quote visit.
With no job software, the owner pasted the list into a chat assistant with the addresses, the rule "consumer units need both of us, the apprentice can't do inspections", each job's duration and the customer windows, and asked for a plan for two people. The answer was a sensible split: both together for the two morning replacements, then the apprentice on the small fixes in one part of town while the owner did the inspections and the quote in another.
Two things needed checking. The assistant estimated drive times between addresses from general knowledge, and two of them were plainly wrong for a road the owner knew was slow at school pick-up time. And it had scheduled the quote visit at 4:30pm, when the customer had asked for mornings. Once corrected, the plan saved an estimated 35 minutes of driving compared with the owner's first idea, which was to do everything together. For a business this size, that's the realistic ceiling: a helpful second opinion on the day, not a system. Once a third person joins, or the owner is spending more than half an hour a day on it, job software with proper routing starts to earn its subscription.
A holiday-let maintenance team is a field service team too
Scheduling problems look similar well outside the trades. Consider a holiday-let manager (a made-up but typical case) with 30 cottages, running four cleaners and two maintenance staff on changeover days. Its hard constraints are unusually strict: every clean must happen between 10am checkout and 4pm check-in, and a property can't be cleaned while maintenance is working inside.
The manager's rules for the scheduler were: cleans are fixed to the 10am-4pm window; maintenance jobs at occupied cottages only between 10am and 6pm with notice to guests; urgent faults (no hot water, a lockout) jump the queue; and hot tub checks always go to the one person trained on the chemicals. Once the rules were written down, the routing tool in its property-management software handled the order, and the manager's Saturday morning scheduling dropped from about an hour to 20 minutes. The same approach suits a boutique hotel's maintenance team or a campsite's grounds staff, where the "customer window" is a guest's check-in rather than an appointment.
Where AI scheduling goes wrong
- Optimising drive time at the expense of relationships. Regular customers who always see the same engineer may care more about that than about a 10-minute earlier arrival. Put customer preferences in as rules.
- Ignoring what engineers want. A schedule that's efficient on paper but has someone finishing 40 minutes from home every Friday will cost you staff. Ask the team for their three biggest scheduling complaints and turn them into rules.
- Bad durations cascading. One under-booked job pushes every later job back. Review durations quarterly against actual job times.
- Out-of-date skills lists, as in the gas example above, where the consequence can be a safety issue, not just a delay.
- Customers not told about changes. The software moves the job; somebody has to make sure the customer knows.
Shadow mode first, then four numbers to watch
Run two weeks in shadow mode: the tool suggests, you schedule as usual, and you compare. Then two weeks of the tool suggesting and you approving. Only then consider letting it assign jobs by itself, and only for job types with a clean record.
Track four numbers for a month before and after:
| Measure | Heating firm before | After six weeks (illustrative) |
|---|---|---|
| Drive time per job | 15 min | 11 min |
| Jobs per engineer per day | 3.8 | 4.2 |
| Arrivals inside the promised window | 81% | 90% |
| Owner's scheduling time per day | 75 min | 20 min |
Also watch first-time fix rate (jobs finished on the first visit), because a schedule that packs in more jobs but sends engineers without the right parts just moves the work to a second visit. If that number falls while jobs per day rise, the tool is optimising the wrong thing, and it's time to revisit the parts and skills rules rather than celebrate the busier diary.
Scheduling and dispatch questions from trades owners
Is AI dispatch worth it for a one- or two-van business?
Usually not as a separate product. With one or two vans, the owner can see the whole day at a glance, and a route-optimise button in your job software or a multi-stop route planner covers most of the gain. AI dispatch starts paying back once three or more people are out and the owner is spending an hour or more a day on the schedule.
Will engineers accept a schedule built by software?
More readily if they can see why a job was given to them and if their preferences are in the rules: finishing near home on Fridays, not two loft jobs in a day. Run the tool in suggest-only mode first, let the team point out bad assignments, and adjust the rules before letting it assign anything on its own.
Can the scheduler handle emergency call-outs?
Most tools let you add an urgent job and re-optimise the rest of the day, showing which bookings move. The decision about which customer to delay should stay with a person, and customers whose times change need telling straight away, which is where automatic texts help.
Does AI scheduling work without a job management system?
Only in a limited way. A chat assistant can suggest an order for a day's jobs from a pasted list, but it can't see live locations, calendars or job progress, and it may get travel times wrong. For regular use you need scheduling software with the jobs, people and addresses in it.
Further reads
- AI Scheduling for Small Builders: Subcontractors and Deliveries — Scheduling subcontractors and deliveries on building jobs.
- Can AI Triage Emergency Plumbing Calls Out of Hours? — Sorting urgent calls before they reach the schedule.
- How a Plumber Can Save Five Hours a Week on Admin With AI — Where else a plumbing firm can save admin time.
- Fleet Admin for Small Businesses: Automate Services, Checks and Fuel — Keep the vans themselves serviced and checked.
- AI Shift Scheduling vs a Rota Spreadsheet for Small Teams — Rota planning for staff who don't travel between jobs.
- AI Route Planning for Landscaping Crews: Time and Fuel Savings — Measure a baseline week, group customers into area-days, let a routing tool order the stops, and work out what the saved miles and hours are really worth.
- What Tradespeople Actually Use AI For: 12 Real Jobs It Handles — Twelve everyday trade jobs AI handles well, from voice-note job sheets to invoice chasers, each with a real-looking example and the mistake to watch for.
- How Cleaning Companies Use AI to Quote and Book Jobs in Minutes — Instant booking forms, AI triage of messy enquiries and calendar rules: how a cleaning firm gets from enquiry to booked slot in minutes.
- Why Most Landscapers Haven't Adopted AI Yet, and Where to Start — The honest reasons landscaping firms hold off on AI, the three office jobs to start with, and a first-month plan for a four-crew firm.
- Can AI Replace the Home Survey for a Removal Company? — Which moves an AI video survey can quote without a visit, what it misses, and a hybrid survey that keeps most of the time saved.
- Can a Small Courier Firm Use AI to Plan Delivery Routes? — What route-optimisation software really does for a small courier firm, what it costs per stop or per driver, and a four-van worked example.
- AI Appointment Scheduling for Small Businesses: Tools and Setup — Booking engine first, rules second, AI at the front door last. Tools, prices, a six-stage setup and an architect practice's bookings worked through.
- How to Reduce No-Shows With AI Reminders and Automatic Rebooking — Make rearranging easier than not turning up. Reminder timings and wording, AI reply handling, automatic rebooking and a heating firm's month costed.
- Capacity Planning With AI: Know When Your Team Is Full — Find the role, day or machine that limits your next booking, with a practical weekly capacity calculation and checked AI scenarios.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Jobber help article on route optimisation in the New Schedule and Jobber's Receptionist help article; Housecall Pro May 2026 product updates; ServiceTitan Dispatch Pro product page.