Design a Reorder & Subscription Service

Problem statement

Design a reorder and subscription service for an e-commerce retailer selling consumables such as groceries, pet food, diapers, and household supplies. Customers subscribe to items on a cadence they choose (weekly, biweekly, monthly, every 6 weeks), receive a discount for committing, and the system automatically generates the next cart, charges the stored payment method, checks inventory, creates a shipment, and notifies the customer. The customer can pause, skip the next delivery, change frequency, swap items, or cancel at any time before a published cutoff.

This is not SaaS billing with a database attached. In a SaaS system, a failed renewal degrades an account. In physical subscription commerce, a failed renewal means the customer's dog food runs out, the meal-kit box never arrives on the chosen day, or the warehouse picks an order for which the stock was already sold. The design therefore couples three hard domains that most billing tutorials ignore: a calendar engine that must fire millions of renewals in tight windows without double-charging anyone, a money engine that must survive gateway ambiguity and card declines, and an inventory engine that must promise goods for a future ship date.

Why the problem is distinctive

Three properties make autoship harder than one-time checkout. First, the write path is initiated by the merchant, not the customer, so every charge is a merchant-initiated transaction under card-network stored-credential rules, with network tokens, account updaters, and regional consent laws in play. Second, the customer retains a standing right to mutate the future: pause, skip, reschedule, and cancel must race safely against a billing job that may already be executing. Third, the promise is physical: by the time money moves, the warehouse must be able to ship on the promised day, which means inventory must be checked and reserved days before the charge.

The four architectural planes

1. Lifecycle plane: subscription state machine, cadence computation, cutoff enforcement, pause/skip/cancel semantics.
2. Billing plane: due-date scheduling, idempotent charge initiation, gateway adapters, retry and dunning, card lifecycle.
3. Fulfillment plane: cart generation, inventory availability-to-promise, reservation, order creation, shipping and tracking.
4. Intelligence plane: churn prediction, save offers, frequency recommendations, replenishment forecasting, unit economics.

A strong answer keeps these planes decoupled. The billing plane must never silently cancel a subscription because a notification failed; the fulfillment plane must never ship against an uncharged order; the intelligence plane may suggest, but only the lifecycle plane mutates subscription truth.

Public operating baseline versus design assumptions

Public figures establish the category. Chewy reported fiscal 2023 net sales of roughly $11.4B with Autoship accounting for about 73% of net sales, and around 20.5 million active customers; these are company-reported earnings figures. HelloFresh reported approximately 7.9 million active customers and around 300 million meals delivered in 2023 under a weekly box with a modification cutoff. Amazon's Subscribe & Save, launched in 2007, discounts recurring consumable deliveries by 5% to 15% with customer-selected frequencies of one to six months. Dollar Shave Club reached roughly 3 million members before its $1B acquisition in 2016.

For capacity planning, this answer explicitly assumes a mature multi-category retailer with 6 million active subscriptions, 150 million subscription shipments per year (about 410K per day average), and a peak day of 1.2 million renewals caused by month-start and week-start anchoring. Unless tied to a citation, every number below is a stated design assumption.