01 the business
Electrical and HVAC service company
45 technicians, two dispatchers
The dispatchers build tomorrow's schedule on a whiteboard, juggling skills, locations and parts. A sick day or an overrunning job means an hour of phone calls. Customers get four-hour windows and ring in to ask where their technician is.
02 before
The problem
The dispatcher builds tomorrow's schedule in their head and on a whiteboard, juggling who has the right skills, who is nearest and which van has the parts. One sick day or overrunning job means an hour of phone calls. Customers get vague time windows and ring in to ask where their technician is.
- Scheduling takes most of each dispatcher's afternoon.
- Travel time is high because jobs are not grouped by area.
- Customers call to chase arrival times.
03 the build
What the AI does
- Step 1: Reads open jobs, crew skills, availability and van stock from ServiceTitan, Jobber or Housecall Pro.
- Step 2: Builds a proposed schedule that groups jobs by area and matches each one to a qualified person, using Google Maps travel times.
- Step 3: Explains each assignment in plain words, such as nearest qualified technician or only crew with the part on board.
- Step 4: Re-plans the rest of the day when a job overruns, a cancellation comes in or someone calls in sick.
- Step 5: Texts customers through Twilio with their arrival window and sends updates when it changes.
paused: waiting for a person to approve
How it was put together
- Read open jobs, technician skills, availability and van stock from the field service system.
- Proposed a schedule grouped by area, using real travel times.
- Explained each assignment in plain words.
- Re-planned the rest of the day on overruns or sick calls, and texted customers updates.
04 human in the loop
Where a person stays in charge
The dispatcher reviews and approves the proposed schedule each day, and confirms any mid-day change before crews or customers are told.
- The dispatcher approves the daily schedule and every mid-day change.
- Customer texts only go out after the dispatcher confirms a change.
- Rules for skills and certifications are hard limits, never traded for travel time.
05 rollout
From kickoff to live in 5–6 weeks
- Weeks 1–2Clean skills and certification data; agree scheduling rules.
- Weeks 3–4Propose schedules alongside the whiteboard and compare.
- Weeks 5–6Switch to approve-the-proposal; add customer arrival texts.
06 after
What changes
The dispatcher starts from a sensible draft instead of a blank board, and changes during the day take minutes rather than a string of calls. Customers know when to expect someone.
Assume 45 technicians and better grouping saving 20 minutes of driving each per day, over 220 working days.
45 × 20 minutes × 220 days = 198,000 minutes, or 3,300 hours a year of driving turned into working time.
What to measure
- Drive time per technician per day.
- Jobs completed per technician per day.
- Dispatcher hours spent scheduling.
- Customer calls asking about arrival time.
07 evidence
What the research says
In a field experiment with 758 BCG consultants, those using GPT-4 on tasks within its abilities finished 12.2% more tasks, 25.1% faster and at over 40% higher quality, but did worse on a task outside its abilities.
About 75% of generative AI's potential value sits in four areas: customer operations, marketing and sales, software engineering, and R&D.
08 lessons
Mistakes to avoid
- Bad skills data. Garbage in, wrong technician out.
- Optimising travel only. Customer priority and skills come first.
- Not explaining choices. Dispatchers trust what they can understand.
09 questions
Common questions
- Does it replace our field service software?
- No. It works with ServiceTitan, Jobber or Housecall Pro and writes the schedule back into them.
- Can it use vehicle tracking?
- Yes, if you use a telematics tool such as Samsara, it can use live locations when re-planning.