Vehicle routing problems have been studied extensively. There are well-understood algorithms for minimising distance, minimising time, and balancing loads across a fleet. What these algorithms assume — uniform vehicles, consistent speeds, rational objective functions — rarely matches what a real logistics operation looks like.
The constraints that matter in practice
Real fleet operations carry constraints that most routing models don't capture. Driver hours regulations create hard cutoffs that vary by jurisdiction and accumulate across the week, not just the day. Vehicle capability varies within nominally identical equipment — a vehicle with a faulty tailgate lift changes which deliveries it can take. Customer time windows are often softer than stated but variable in ways that experienced drivers know and systems don't.
Planning for HGV operations specifically adds bridge restrictions, weight limits by road type, and turning radius constraints in urban delivery environments. An optimal route in terms of distance may be operationally impossible for a specific vehicle. The routing model has to incorporate the vehicle characteristics, not just the depot-to-delivery graph.
“Real fleet operations carry constraints that most routing models don't capture.”
Historical data vs real-time conditions
Most route optimisation implementations use historical travel time estimates. This works well for stable conditions and provides a usable planning baseline. It breaks down during events — sporting fixtures, weather, infrastructure works — that aren't captured in historical averages. The question is how much real-time correction is operationally useful given driver communication overheads.
In our experience, overnight planning with morning confirmation works better than continuous real-time re-optimisation for most UK logistics operations. Drivers can work with a stable route. Re-optimising mid-route has limited value once the sequence is underway, creates communication overhead, and can generate routes that are mathematically better but operationally disruptive.
What good looks like
A well-implemented route optimisation system doesn't replace dispatcher judgment — it gives dispatchers better starting points and handles the combinatorial complexity they can't handle manually. The system generates initial routes; experienced dispatchers review and adjust based on knowledge the model doesn't have. Over time, that adjustment data feeds back to improve the model.