The situation
The business runs a regional parcel and pallet operation out of two depots in the Midlands, with around 95 staff and a mixed fleet of about 40 vans and rigids. The ops lead is the kind of operator who knows every postcode in the patch and which driver hates which drop. That depth of knowledge had become the bottleneck.
Every morning started at 05:30 in front of a magnetic whiteboard. Manifests printed off the WMS were sorted by hand into rough geographic clusters, then balanced against driver hours, vehicle capacity and a mental map of which customers had booked time slots. On a normal day this took three hours. On a Monday, or any day after a bank holiday, it crept closer to four and a half. By the time the last van rolled out, the ops lead had already lost half a shift to firefighting.
When volumes spiked, the cracks widened. Two drivers would end up in the same village an hour apart. A 30-drop round would be handed to someone with seven hours left on their card. Time-critical pallets would slip because nobody had spotted the booking note buried in the order. OTIF was sitting at around 91 percent, which the customer base had stopped finding charming. Agency drivers were being booked as a buffer almost every week, not because the work needed them but because the plan could not be trusted.
The ops lead was clear about what they wanted. Not a new TMS. Not a rip and replace. Just to stop spending the best hours of the day moving magnets around a board.
What we did
We spent two days in the depot watching the morning routine and pulling six months of historical manifest data, driver timesheets, vehicle telematics and customer SLA records. The goal was to understand the actual decision logic the ops lead was applying, not the version written down in the SOP.
From there we built a dispatch assistant that sits alongside the existing WMS rather than replacing it. Each evening it pulls the next day's confirmed orders, geocodes the drop points, and produces a recommended allocation across available drivers and vehicles. It optimises for delivery density, driver hours, vehicle capacity and booked time windows, and it flags any drop it is not confident about so a human still makes the call. Crucially, it knows which customers expect which driver and respects those preferences unless the maths gets silly.
We deliberately left several things alone. The WMS stays as the system of record. Proof of delivery, invoicing and customer comms run exactly as before. Drivers still get a paper run sheet alongside their handheld, because that is what they wanted and arguing about it would have wasted everyone's time. The ops lead keeps full override on every plan, and roughly one in ten allocations gets nudged before the vans leave. That is the point. The tool does the grunt work and the human keeps the judgement calls.
Rollout took six weeks from kickoff to live running, including a fortnight of shadow mode where the recommended plan was generated each night but the whiteboard still ran the day. Once the numbers lined up, the board came down.
The result
Daily dispatch planning dropped from three to four hours to under 30 minutes. The ops lead now spends the recovered time on customer issues, driver development and the kind of commercial conversations that used to get pushed to next week and never happen.
Delivery density improved from an average of 38 to 46 drops per van per day on the parcel side, which has absorbed a 14 percent volume increase without adding a single permanent driver. Agency cover has fallen by roughly two shifts a week. OTIF has moved from 91 to 96.4 percent over the first quarter of running, and the late-delivery complaint volume is down by about two thirds. Combined savings on overtime, fuel and agency spend are tracking at around £148k a year against a build cost recovered inside four months.
The honest line on team sentiment is this. The drivers were sceptical for the first fortnight, mostly because they assumed any system that touched their rounds would make their day worse. Once it became clear the plan was usually sensible and they could still get changes through the ops lead in 30 seconds, the grumbling stopped. The ops lead's verdict was shorter. They said it was the first time in years they had eaten lunch sitting down.