A D2C apparel brand was running online sales, retail, shipping and two returns systems as five separate sources of truth. devx labs built an AWS data warehouse that reconciles all five every night — and answers business questions in plain English.
Replyall is a direct-to-consumer apparel brand designing everyday clothing — tees, polos, joggers, overshirts — engineered around a proprietary "Fabric Library" of performance textiles for temperature regulation, stretch and weather resistance. The brand sells through its Shopify storefront and a retail store in Mumbai.
As volumes grew, the operating picture behind the storefront didn't keep pace. Online sales sat in Shopify. Shipping and delivery status sat with courier partner Proship. Returns were split across two systems — the live portal, Pragma, and its predecessor, Returns Prime, holding everything before it. And Broadway, the retail store's point-of-sale system, ran entirely apart from the online numbers with its own sales, returns and stock.
Answering a single question about the business — what did we actually keep, net of returns, across both channels? — meant assembling five exports by hand, every time.
Every architecture decision traced back to one of three roles at Replyall — and what each of them had to stop doing manually.
Running growth and inventory decisions off GMV, AOV and channel mix — but every figure required an analyst, an export, and a wait. Questions that occurred mid-meeting went unanswered.
Return and exchange cases lived in one system, their history in another, and delivery failures in a third. Reason codes never lined up, so "why are returns up?" had no single answer.
Needed sell-through by SKU, colour and size to catch stock-outs and overstock early — but online and in-store stock were counted separately, and neither refreshed on its own.
The problem was not a missing dashboard. It was that no two systems agreed on what a sale, a return or a customer was — and three of the five sources offered no automatic connection at all, so the data could not simply be pulled on a schedule.
Returns made the gap concrete. A return raised in the current portal, a return that predated it, and a return handed back over the retail counter were three different records in three different shapes. Measured against online sales alone, every return rate on every report was wrong.
devx labs designed, built and deployed a fully managed AWS pipeline — ingestion through to conversational analysis — with the business rules behind every number documented in plain language, not just in code.
A scheduled Lambda calling the Shopify GraphQL API, nightly bots for the courier portal, returns portal and retail POS, and a one-time historical load folding legacy returns into the same schema — five sources, one staggered nightly sequence.
Raw files land dated and untouched in S3, then move Bronze → Silver → Gold inside Redshift, so every published figure can be traced back to exactly what the source system handed over.
Net revenue, return rate, fulfilment TAT and what counts as a genuine customer are calculated once in the warehouse — so every dashboard, and every team, tells the same story.
Sales & Revenue, Returns & Exchanges, Logistics, Customers and Inventory — 19 QuickSight tabs covering revenue-bridge waterfalls, LTV cohorts, RTO analysis and variant-level stock health.
A Claude MCP connection into Redshift lets the team ask questions in plain English and get answers straight from the warehouse — no SQL, no analyst in the loop, no export.
EventBridge schedules run the nightly sequence unattended; any failure alerts with the exact job named. Because each job pulls a fixed window, a re-run never creates gaps or double-counts.
Inventory health bands — Critical, Watch, Healthy, Overstock — and the Returns & Exchanges SLA days for approval, pickup and processing are exposed as adjustable controls, so the team retunes them as the business changes without raising a request.
Proship cost joined to Shopify revenue produces Shipping Cost as % of Revenue. Pragma returns matched to Shopify orders produce a true return rate — neither exists in a single source.
It surfaced 74% of returns as Size & Fit, sizing running large (L → M, XL → L), and WELCOME10 as the highest-returning cohort.
Dashboards read from QuickSight's in-memory SPICE layer rather than querying Redshift on every open, so ten people can work in the reports at once at no additional warehouse cost.
The platform is live in production, refreshing daily and scaling as Replyall adds sources and channels. Founders ask questions in plain English and get answers from the warehouse itself; operations sees returns as one picture instead of three; merchandising reorders against sell-through that was accurate this morning.