Routing an order to the right warehouse with a scoring function

Once you run more than one warehouse, "which site fulfills this?" becomes an algorithm. A clean start is a scoring function: each candidate gets a score, highest wins.

function pickWarehouse(order, warehouses) {
  let best = null, bestScore = -Infinity;
  for (const w of warehouses) {
    if (!w.canHandle(order)) continue;   // oversized, battery SOP, etc.
    const score =
      0.5 * stockCoverage(w, order) +     // does it have every SKU?
      0.3 * proximityScore(w, order.destination) +
      0.2 * (1 - w.utilization);          // prefer a less loaded site
    if (score > bestScore) { bestScore = score; best = w; }
  }
  return best;
}

The weights are a starting point, not truth. Tune them to what actually hurts: split shipments, slow lanes, or overloaded sites.

Two gotchas:

1. Capability gates before scoring. A battery or oversized item must be filtered out of warehouses that cannot legally or physically handle it before you score. Scoring an ineligible site just gives a wrong answer faster.

2. Partial stock. If no single warehouse covers the whole order, decide explicitly whether to split-ship or hold. Silent splitting doubles freight and confuses customers.

We run three sites in China (Shenzhen for oversized and sea-air, Suzhou for consolidation, Dongguan for ecommerce one-piece), and the routing logic is mostly about matching the order to the site actually built for it.