A small workflow change that reduced dispatch overhead, survived real-world messiness, and earned enough trust to expand with the business.
Before
~2.5 hrs/day
Manual reading and reassignment of job notes.
After
~40 min/day
A lighter triage loop that still felt like the dispatcher’s own tool.
Result
5 months running
The owner later added two vans and kept the workflow live.
The dispatcher spent roughly 2.5 hours each day reading job notes and manually reassigning vans. Over the course of a year, two people left because of the work. The problem was not just time pressure; it was the constant cognitive load of doing the same triage task by hand.
Scheduling data lived in a hosted tool with a read-only API. The dispatcher would not use a new application, so the workflow had to live inside SMS. The owner also would not approve any approach that touched customer payment data.
We did not build auto-reassignment. That would have been the most “intelligent” solution on paper, but it would have been the wrong one in practice. The dispatcher did not trust automation, and the business would have turned it off in week one. We chose suggestion plus one-tap approval instead.
The first real failure mode was a data quality issue: job notes used inconsistent van nicknames such as “big blue” and “V-3.” That broke the matching logic in the exact place where trust mattered most. We fixed it with an alias table the dispatcher can edit herself, which is the kind of operational detail that separates something that ships from something that only demos.
This was not a model problem. It was a workflow problem. The win came from designing around human trust, operational reality, and the messy data that exists in the real world.