By Q2 of 2023, Kumasi Logistics' on-time delivery rate had slipped from 91% to 76%. Operations Director Akosua Boateng suspected the cause lay not in the fleet, but in the routing logic itself — a spreadsheet stitched together over years of incremental fixes.
Akosua framed the question carefully: was the problem the volume of orders, the geography of Kumasi's informal road network, or the way the team was sequencing each driver's stops? The distinction mattered. Each diagnosis pointed to a different intervention — and a different cost.
Over the following weeks, she assembled a small task force to test three alternative routing approaches against a single week of historical delivery data, looking for the configuration that would restore margin without expanding the fleet.
The first approach was the simplest: re-cluster delivery zones around postal-code centroids and let drivers self-sequence. It was cheap to deploy, but it required every driver to make local decisions that compounded across the day.
The second approach replaced the spreadsheet with a constraint-based solver that minimised total drive time across the cohort. Early simulations were promising — a 12% reduction in idle minutes — but the solver assumed road segments that didn't always exist on the ground.
The third approach was a hybrid: a daily plan generated by the solver, then handed to a dispatcher who could override individual legs based on real-time conditions. It cost more in labour, but preserved the institutional knowledge embedded in the team.
By the end of the trial, Akosua had data on cost-per-stop, on-time rate, and driver overtime across all three scenarios. The choice, she realised, was not really about routing. It was about how much discretion the company was willing to delegate to software — and what it was willing to pay to keep humans in the loop.
