Every fleet runs miles that earn nothing: the empty leg home, the cross-town sequence that visits the same estate twice, the 8am arrival at the gate that opens at 9. Route optimisation is the unglamorous art of finding those miles and deleting them - and it is where the Future of Freight plan’s data-and-technology ambition meets the Tuesday morning manifest.
Wasted miles come in three forms, and each has its own fix. Empty running: the vehicle travelling unladen between the last drop and the next collection - attacked by backload discipline (the constant hunt for return loads, your own or the freight exchanges’) and by network thinking: where the vehicles are versus where tomorrow’s loads are. Congestion and timing: the route that threads the rush hour, the delivery booked for the wrong side of a site’s peak - attacked by time-window intelligence: knowing each customer’s receiving rhythm and each road’s bad hours, and sequencing around both. And sequence waste: the drops ordered by habit or geography-as-the-crow-flies rather than by actual roads - attacked by proper route planning, which on anything beyond a handful of drops means software, because the number of possible sequences defeats intuition quickly.
The measurement that reveals all three: miles per drop and fuel per drop, by route and by planner, watched over time. The numbers name the leaky routes without blame.
The optimisation toolkit scales with the fleet. The basics cost nothing: customer time-windows documented, a live traffic habit in the planning office, the empty-leg review weekly. The software tier earns its licence quickly beyond a few vehicles: route optimisation tools that sequence drops against time windows, vehicle capacities, driver hours and real road speeds - and the hours constraint is the one planners most often underweight, because a route that breaks the driver’s day is not optimised, it is illegal. The data tier completes the loop: the telematics’ actual journey times feeding back into the plan, so next month’s routes are built on measured roads, not assumed ones.
The human layer decides whether any of it works: planners who know the customers’ sites - the gate that is really at the back, the yard that cannot take an artic at 4pm - and drivers whose local knowledge is solicited rather than overridden, because the optimised route that ignores the driver who runs it weekly is optimised only on screen.
Routing waste is fuel, time and carbon bought for nothing: empty legs, congestion, bad sequence. Measure miles and fuel per drop, fix time-windows and backloads, let software sequence beyond a handful of drops inside the hours rules - and keep planners and drivers smarter than the algorithm about the sites it has never visited.
The optimisation output is only as good as its inputs, and three inputs repay attention before any software decision. The service times: how long a drop actually takes at each customer - the single biggest source of fantasy routing is the assumption of five-minute stops on accounts that take forty. The travel times: learned from the fleet’s own telematics by time of day rather than a generic road-speed table, because the M25 at 7am and at 11am are different roads. And the constraints written down: the time windows, the vehicle restrictions, the driver breaks, the customer quirks (the yard that cannot take artics, the site that closes for lunch) - the planner carries these in his head, and software cannot improve on knowledge it was never given. The fleets that spend a month cleaning these three inputs get more from cheap software than fleets with expensive software and dirty data ever do.
Routing projects fail on people more often than on mathematics. The planner whose twenty years of route knowledge is being ‘replaced by software’ will find the exceptions the algorithm missed - and there will be exceptions, because the model’s constraints are never complete on day one. The fleets that land optimisation well frame it as the planner’s tool rather than the planner’s replacement: the software drafts, the planner edits, and the edits flow back into the constraint set so the model learns the yard that closes for lunch and the customer who must be first drop. The drivers need the same handling: the optimised route that ignores how the experienced driver sequences the awkward estate will be quietly re-sequenced from the cab, and the mileage saving evaporates. Involving the drivers’ knowledge in the model - which drops clash, which access is really artic-legal - is not courtesy; it is the data the software is missing.
Honesty about the other direction matters too: the small, stable operation - a handful of vehicles on repeat work with customers known for years - may gain little from software, because the planner’s head already holds a near-optimal solution and the data-cleaning cost would exceed the saving. The signal to automate is not fleet size but variability: when the day’s work changes enough that the planner is rebuilding routes from scratch each morning, the algorithm starts paying. The fleets that buy against that signal avoid the industry’s quietest waste - optimisation software faithfully producing the route the planner already knew.
Routing gains compound in a way few fleet projects do: fewer miles is less fuel, but also less tyre and brake wear, fewer driver hours per drop, more drops per vehicle-day, and a smaller fleet for the same work - or the same fleet with capacity for the next contract. That is why the routing conversation belongs at the top of the efficiency list rather than inside the IT budget: it is the one project that touches cost, service and capacity simultaneously, and the fleets that measure all three find the software line is the smallest number in the case.