Classical vehicle routing is a beautiful, solved-enough problem: given stops, distances, time windows, and capacities, find the cheapest feasible sequence. The trouble is the word "given." In a live network, travel times are distributions, service times vary, new orders arrive mid-route, and the time window that mattered at 6 a.m. has been renegotiated by 10. The inputs move while the optimizer thinks.
Routing under uncertainty reframes the objective. The question is not "what is the shortest tour through today's stops" but "what sequencing policy performs best across the range of days this network actually experiences." Those produce different answers. The stochastic answer will sometimes accept a slightly longer planned tour because it degrades gracefully: it keeps the tight windows early, holds slack where variance concentrates, and leaves late-day flexibility for the orders that have not arrived yet but statistically will.
The re-optimization trap
A common workaround is aggressive re-optimization: solve deterministically, then re-solve every time reality diverges. This is better than nothing and worse than it looks. Each re-solve is locally optimal and globally amnesiac. It thrashes assignments, breaks promises made an hour earlier, and confuses drivers and customers alike. Stability is a feature of good routing, and stability has to be optimized for, not hoped for. The stochastic formulation does this naturally: a plan chosen for robustness needs fewer revisions in the first place.
The best route is a probability, not a fact.
Resample conditions and watch the winner change.
| A · C · F | 1 | 100% |
One graph, one question, many answers. Frequency is the honest summary.
What operators should measure
The revealing metric is not planned cost. It is realized cost, including the failures: missed windows, overtime, re-deliveries, and the expediting that rescues them. Deterministic plans consistently win on planned cost and lose on realized cost, which is the only one the P&L records. Networks that make this one measurement change almost always discover their true routing cost is 10 to 20 percent above plan, and that the gap is not bad luck. It is the price of optimizing the wrong problem.