Eleven weeks from signed contract to first live vendor. Here is the honest account of how we did it — and what we would do differently.
The Brief
MrGamArts needed a multi-vendor marketplace where independent retailers could list products, manage their own inventory, process orders, and receive payouts — all under a unified storefront. The client had a hard deadline: a trade event where the platform would be demonstrated to potential vendors and investors.
Week 1–2: Architecture Sprint
We spent the first two weeks entirely on architecture. No application code. We designed the vendor isolation model (each vendor's data is logically separated at the database level with row-level tenant IDs), the commission and payout engine, the product catalogue structure, and the storefront routing. We also identified the three highest-risk components: real-time order routing, split-payment settlement, and vendor onboarding verification.
This upfront investment paid for itself many times over. By the time we wrote the first line of application code, every developer knew exactly what they were building and how it connected to everything else.
Weeks 3–8: Core Build
We ran two parallel tracks: backend API and storefront. The API team built vendor management, product catalogue, order processing, and the payout engine. The storefront team built the customer-facing shopping experience, vendor storefronts, and the admin dashboard. Both teams worked against the same OpenAPI specification we had written in week two.
We hit one serious problem in week six: the split-payment settlement logic had a flaw in how it handled partial refunds on multi-vendor orders. A return of one item from a three-vendor basket had to correctly reverse commissions, restock inventory for the right vendor, and trigger a partial refund to the customer — all atomically. We spent four days on this. The fix required restructuring the order line item model. Painful, but caught before production.
Weeks 9–11: Integration, QA, Hardening
The final three weeks were integration testing, performance testing under simulated load, vendor onboarding flow QA, and production infrastructure setup. We ran 48-hour load tests simulating peak traffic before the deadline.
What We Would Do Differently
We underestimated the complexity of the vendor onboarding verification workflow. We had planned three days; it took eight. Identity verification, document upload, bank account validation, and admin approval each had edge cases we had not fully specified. Next time, vendor onboarding gets its own sprint in week one.
The Outcome
The platform launched on time. Forty-three vendors onboarded in the first week. The trade event demonstration ran without incident. The client reported that three investors requested follow-up meetings after seeing the platform live.