Ask anyone outside logistics what route optimization is and you will hear some version of the same answer: find the shortest path. It is an intuitive picture, and it is almost entirely wrong for the way consumer goods actually move. The shortest-path framing treats a delivery day as a clean mathematical object — a set of points on a map, a single cost to minimize, one correct answer waiting to be computed. The real day is nothing like that. It is a negotiation between dozens of constraints that contradict each other, where the best plan is rarely the shortest one and almost never the one a pure distance solver would produce.
The reason is that distance is only one cost among many, and usually not the one that hurts most when you get it wrong. A van that takes the geometrically shortest loop but arrives at a grocery account during its receiving blackout has not saved anything; it has burned a slot, irritated a buyer, and possibly triggered a redelivery that costs more than the kilometers it saved. The interesting problems in route and delivery are not about geometry. They are about the trade-offs between commitments you have made to customers, limits imposed by your fleet, and the realities of the people doing the driving.
This is why practitioners draw a hard line between hard constraints and soft constraints. A hard constraint is non-negotiable: a refrigerated load cannot ride in a non-refrigerated van, a vehicle cannot exceed its legal weight, a driver cannot work past a regulated shift limit. Violate one and the plan is simply invalid. Soft constraints are different. They are preferences and costs that you would like to honor but can trade against each other when honoring all of them at once is impossible. Almost everything that makes routing hard lives in this second category, and the craft is in deciding how much each one is worth.
Time windows are the canonical example. Many accounts will only receive between, say, nine and eleven in the morning, and a serious distributor has hundreds of these constraints fanning out across a city, each pulling the schedule in a different direction. Treat every window as inviolable and you may find no feasible plan exists at all. Treat them as soft, with a penalty that grows the later you arrive, and the solver can make a defensible call: hold the high-value account whose buyer is strict, and absorb a small, agreed slip at the corner store that genuinely does not mind. The output stops being a yes-or-no feasibility check and becomes a ranked set of compromises.
Driver familiarity is the soft constraint that planners feel most and model least. A driver who has run the same territory for two years knows which loading dock is around the back, which receiver signs quickly and which stalls, where you can park a van without a ticket, and which account will quietly take an extra case if you ask. None of that lives in a distance matrix, yet it is real time and real service quality. A plan that reshuffles every route each morning to shave theoretical minutes can be slower and worse in practice because it throws away this accumulated local knowledge. Good optimization treats route continuity as a value to preserve and charges a cost for churn, rather than optimizing each day from a blank map.
Vehicle limits add a second dimension that distance alone never captures. Capacity is not a single number; it is volume and weight and sometimes compartment layout all at once, and a load can be light but bulky or heavy but compact. Cold-chain compartments, tail-lift requirements, and access restrictions for larger vehicles in dense neighborhoods all reshape what any given van can actually do. The honest version of the problem is multi-dimensional packing happening at the same time as sequencing, and the two interact: a smarter packing decision can unlock a shorter, kinder sequence.
Because these objectives genuinely conflict, the right way to think about a route plan is as a point on a trade-off frontier rather than a single optimum. You can usually buy a little more on-time performance by spending a little more distance, or protect driver routes at a small efficiency cost. There is no universal answer to how those exchange rates should be set; they encode a commercial strategy. A plan tuned for a premium service promise weights time-window adherence heavily. A plan defending margin in a low-cost channel may let windows flex and squeeze the fleet harder. The optimizer's job is not to hide this choice but to make it explicit and adjustable.
This is the philosophy behind Route and Delivery in our platform, and it is why it sits on the same shared data model — the ConnectX data layer — as field sales, ordering, and shelf execution rather than running as an isolated engine. The signals that should bend a route live in those other systems: the order a rep just placed, the account's real receiving behavior, the priority of a promotion landing this week. FMCG Cloud Intelligence reads that shared context so the route reflects the actual commercial picture instead of a stale, standalone map.
The takeaway for anyone designing or buying this capability is to be suspicious of any tool that promises one optimal route. Real route optimization does not eliminate trade-offs; it makes them visible, prices them in your terms, and lets you choose. Shortest path is a starting heuristic. Everything that matters is in the soft constraints.