Reverse Logistics Routing

Reverse logistics routing chooses where each returned unit should go next: a regional returns center, a store with sell-through capacity, a refurbishment partner, a liquidation channel, a donation path, or disposal when recovery is not viable. The decision is not the same as disposition classification. Disposition answers what to do with the item. Routing answers which physical destination should execute that path at the lowest total landed cost without destroying recovery value.

For a reverse-logistics planner, the practical question is narrow. Given condition, current location, destination options, and cost drivers, which ship-to node minimizes cost while still meeting service windows, sustainability rules, and operational capacity? The model produces a ranked destination recommendation. Warehouse and carrier teams still book the shipment, apply labels, and move the freight. The planner reviews exceptions, overrides when local knowledge conflicts with the score, and owns the final release.

This page sits next to related return workflows. Disposition logic often runs first; see Automated Return Disposition. Policy and refund eligibility can gate whether a return is even in scope; see Refund Policy Compliance Checker. Fraud signals can change whether high-value recovery routes are allowed; see Return Fraud Risk Scoring.

Routing quality depends on three input families. If any family is missing or unusable, the model should not invent a destination.

Condition and recovery state. Grade, packaging integrity, completeness (accessories, tags, serials), and any quarantine or safety flags determine which destinations are eligible. A unit graded for full-price restock cannot be scored against liquidators as if those options were equivalent. A hazmat or recall flag can remove most recovery nodes entirely.

Location and network state. Origin node (customer pickup point, store, locker, or inbound dock), eligible destinations, transit lanes, cutoffs, and current capacity or backlog at each node. Without a usable origin and at least one eligible destination, there is nothing to optimize.

Cost and constraint parameters. Linehaul and last-mile estimates, handling and processing rates by node, expected recovery value by grade and channel, SLA clocks (customer refund timing, carrier pickup windows), sustainability rules (prefer local restock over long-haul when policy requires it), and hard operational limits (node closed, SKU blocked, partner not contracted).

Planners should treat incomplete manifests the same way they treat incomplete ASN data on inbound. Do not force a route. Request the missing field, hold the unit in a pending queue, or escalate to a manual desk. Empty output is the correct system behavior when condition, location, or cost inputs are absent, stale, or contradictory.

The optimization objective is total expected net cost for that unit (or consolidated carton when multi-line returns share a lane): transportation plus handling plus expected value loss from delay or wrong channel, minus expected recovery value at the destination. In practice, most retail implementations score a short candidate list rather than searching the entire network.

A typical flow:

1. Eligibility filter. Drop destinations that violate grade rules, contracts, capacity, geography, or compliance holds.
2. Cost and value estimate. Compute landed cost and expected recovery for each remaining node using current rate cards, zone tables, and recovery curves by category and grade.
3. Constraint check. Enforce service windows, sustainability preferences encoded as soft or hard constraints, and operational locks.
4. Rank and explain. Return a primary destination plus one or two alternates with the cost delta and the binding constraint that eliminated cheaper options.

Explanations matter as much as the rank. A planner who sees “nearest DC is $4 cheaper but exceeds restock SLA by 36 hours” can accept the recommendation or override with a documented reason. Opaque scores erode trust and push teams back to spreadsheet heuristics.

Consolidation rules should be explicit. Shipping each unit alone to its individually optimal node can raise carton count and per-package fees. The model may prefer a slightly suboptimal single destination for a multi-item return when split shipments violate carton or carrier policies. Those rules belong in configuration, not in ad hoc planner judgment after the fact.

Lowest freight is not always the right answer. Several constraint classes routinely dominate the cost term:

Resale-value preservation. Fashion, electronics, and seasonal goods lose recovery value with time and with the wrong channel. Routing a near-new unit to liquidation because that lane is cheap can wipe out more margin than the freight savings. Grade-to-channel maps should be hard gates, not soft preferences.

Service and customer clocks. Refund or exchange SLAs, carrier pickup appointments, and store capacity for customer drop-off all create time boxes. A cheaper destination that misses the clock is ineligible.

Sustainability and network policy. Some retailers require local restock when grade and demand allow it, or limit long-haul moves for low-value returns. Encode these as constraints or penalty terms so the recommendation stays audit-ready.

Operational reality. Node holidays, labor shortages, quarantine rooms at capacity, and partner turnaround SLAs change eligibility daily. Stale capacity feeds produce confident but wrong routes. Refresh cadence should match how fast backlog moves in your network.

Risk and compliance overlays. Fraud holds, serial mismatches, and regulated categories can force inspection nodes even when a direct-to-store restock looks cheaper. Those overlays should short-circuit optimization rather than compete as another cost line.

When a constraint removes every candidate, the model should return empty output with a reason code (no eligible destination, missing cost card, grade unknown) instead of falling back to a default DC. Defaults hide data gaps and teach the network the wrong behavior.

Empty output is a first-class result. Trigger it when condition grade is missing or unmapped, origin or destination geocodes cannot be resolved, rate or processing costs are unavailable for all candidates, or eligibility filters leave zero nodes. Downstream systems should park the return in a work queue, not auto-label to a house DC.

Human-in-the-loop remains mandatory for execution. The model recommends a destination (and optionally a lane and service level). Operations still creates the shipment, chooses carrier pickup vs drop-off when both are allowed, and confirms the physical handoff. Planners override when store managers report sell-through the model has not seen yet, when a partner outage is known offline, or when a VIP or regulated case needs a supervised path.

Measure the workflow on planner-relevant metrics: share of returns with a scored route on first pass, override rate by reason, cost per recovered unit versus a fixed-node baseline, recovery value retained by grade, and time from receipt to outbound label. Track empty-output volume by missing-field type so data owners can close the gaps that keep units stuck.

Start with a bounded pilot: one category, one origin region, and a small destination set with clean rate cards. Align disposition grades with routing eligibility before you turn on optimization, or the model will optimize the wrong candidate set. Keep refund and fraud systems upstream so ineligible returns never enter the routing queue.

Publish the recommendation contract to warehouse systems: primary destination ID, alternates, cost delta, constraint notes, and empty-reason codes. Train planners to treat overrides as labeled events, not silent label edits. Review weekly which constraints bind most often; chronic soft-constraint violations usually mean the policy and the cost model disagree, and one of them needs to change.

Done well, reverse logistics routing turns return moves from a fixed “send everything to the central RC” habit into a unit-level cost and recovery decision that still leaves shipping authority with operations. The model proposes the lowest-cost feasible destination. People ship only after the recommendation, or the empty hold, is resolved.