Design a Returns & Refunds Service

Hard45 min
1 / 30
understanding•10 min read

Problem Statement: Returns Are Reverse Logistics Plus Money Movement

Frames returns as two coupled flows - physical goods moving backwards and money moving backwards - with fraud risk wrapped around both.

Problem statement

Design a Returns & Refunds Service for a large e-commerce platform. Customers initiate returns against delivered orders; the system generates a Return Merchandise Authorization (RMA), issues a shipping label or drop-off credential, tracks the package backwards through carrier and warehouse events, receives and inspects the item, decides the refund amount under policy (taxes, restocking fees, prorated discounts), executes the refund through a payment service provider, updates inventory disposition, notifies the customer in real time, and keeps an audit trail strong enough to survive disputes and chargebacks.

The forward order flow is a pipeline that mostly moves one direction. Returns are harder because both physical goods and money move backwards through that pipeline, and the two flows are decoupled in time, space, and reliability. The package may be lost by the carrier. The warehouse may receive it but fail to scan it. The PSP may accept the refund call and lose the acknowledgement. The customer may return an empty box. A strong design treats returns as a cyber-physical money system: every physical event (label issued, carrier scan, dock receipt, inspection grade) must be correlated with a money decision (partial refund, full refund, denial) under adversarial conditions.

Why the problem is distinctive

An order-management system can retry a failed notification. A refund cannot be quietly retried without risking a double credit, which is pure financial loss. A tracking feed can be stale. A warehouse receipt record cannot be wrong, because it authorizes inventory to be resold and money to leave the company. The design therefore separates return progress (an eventually consistent customer experience) from refund authority (a strongly consistent, idempotent, append-only money decision). Return progress may degrade, go stale, and retry freely. Refund authority must fail closed, reconcile exactly, and produce evidence.

Public data shows the scale is real. The National Retail Federation reported that U.S. retail returns reached approximately $743 billion in 2023, about 14.5% of total retail sales, and e-commerce return rates commonly run 20-30% versus 8-10% for brick-and-mortar. Industry analysts regularly place holiday-season (December-January) return volume at well over $100 billion, creating a January surge that return systems must absorb. These are cited public industry figures, not requirements for our fictional system.

For capacity planning, this answer explicitly assumes a marketplace with 100 million registered customers, 2 million delivered orders per day, a 15% return initiation rate (300,000 returns per day average), a 10x peak factor for the post-holiday surge, and an average refund of $85 (about $25.5 million per day in money movement). Unless a number is tied to a citation, it is a stated design assumption, target, or budget - not a claim about any company's private architecture.

The four architectural planes

  1. Experience plane: eligibility, return reasons, RMA creation, label and drop-off credentials, live status, notifications, support tooling.
  2. Reverse-logistics plane: carrier label purchase, scan ingestion, dock receipt, inspection and grading, disposition back into inventory or liquidation channels.
  3. Money plane: refund calculation, approval thresholds, PSP execution, retries, ledger, settlement reconciliation, chargeback evidence.
  4. Integrity plane: fraud and abuse scoring, policy rules, audit chain, compliance evidence, dispute handling.

A strong interview answer keeps these planes separate. It allows the experience plane to degrade during the January surge without weakening the money plane, and it lets the integrity plane tighten policy without silently changing validated refund arithmetic.

Key Highlights

  • •Returns reverse BOTH physical goods and money; the two flows are decoupled in time and reliability and must be correlated under adversarial conditions.
  • •Model return progress as eventually consistent experience, but refund authority as strongly consistent, idempotent, append-only money decisions.
  • •NRF reported roughly $743B in U.S. retail returns in 2023 (about 14.5% of sales); e-commerce return rates commonly run 20-30%.
  • •Assumed design scale: 300,000 returns/day average, 10x post-holiday peak, $25.5M/day in refund volume.
  • •Four planes: experience, reverse logistics, money, integrity - each with different consistency, latency, and failure rules.
Lead With Money Movement
State in the first two minutes that the refund ledger is the single source of truth and that no carrier scan or warehouse event can credit money directly. This instantly distinguishes a returns architecture from a generic CRUD workflow.
Do Not Draw Returns as CRUD
A design where POST /refund immediately credits the card with no idempotency, ledger, inspection gate, or reconciliation will fail under the first webhook replay or PSP timeout. Returns are a distributed money protocol, not a status update.

Section Rescue Kit

Buzzwords to use:

Reverse LogisticsRefund Authority

Safe statements:

  • "I will separate return progress, which may retry and degrade, from refund authority, which must fail closed and reconcile exactly."
  • "Before drawing services, let me define which physical events are evidence and which component alone can turn evidence into money movement."
Design a Returns & Refunds Service - System Design | WinJob | WinJob