- AT&T reported saving $10M a year in route and fuel optimization with models that aligned local technician skills with local jobs while minimizing mileage.
- AT&T described a fleet of 7,000 trucks, where the mix of service requests changed every day.
- Models predict job duration and travel time and rank technicians on skill fit and distance, so fewer trucks cross town for a job.
- The platform runs in four deployment options: on-premises, air-gapped, private cloud and cloud.
Every extra mile is fuel, time and a later appointment
Your fuel bill is set one assignment at a time. When the nearest qualified technician is not the one sent, the truck drives farther, the customer waits longer, and the next job slips.
The best match is hard to see by hand. Technicians have different skills, start from different places and finish jobs at different speeds, and the day's requests change after the plan is made. Fixed rules help with the simple cases, but they break on the daily mix of installs, repairs and outages that a large territory produces.
How route and fuel optimization works
Your operations team trains models on past jobs and routes and feeds the predictions to the dispatch and routing system you already run.
Step 1: join the history
Combine completed jobs, technician skills and home locations, GPS or telematics trips, job locations, fuel records and weather.
Step 2: predict time on site
Regression models estimate how long each job type takes for each skill level, so plans stop assuming every job is the same length.
Step 3: predict travel
Time series data from past trips trains models of drive time by area, day and hour.
Step 4: rank the match
For each job, the models rank qualified technicians on skill fit, predicted duration and distance.
Step 5: write back and decide
Scores go to your routing system, which builds the routes. Your dispatchers review the plan and approve it.
Step 6: watch for drift
New services, new territories and seasons change the pattern. H2O MLOps tracks model drift and flags a model for retraining.
Working with the routing system you already run
H2O models send predictions and rankings to the routing and dispatch system you already run. That system builds the routes, and your dispatchers review and approve the plan. Location and trip data stay in the environment you choose.
Results
AT&T, an H2O customer since 2015, described a fleet of 7,000 trucks with technicians of different skills working at varying distances from the day's jobs, where the mix of requests changed daily. AT&T used H2O to build a set of models that aligned local skills with local jobs while minimizing mileage, and reported saving $10M a year in route and fuel optimization.