An OMS that survives Black Friday
Stood up an order management system that holds at 42× normal load — replaced a vendor product that had failed two peaks in a row.
The problem
Vendor OMS had failed two consecutive holiday peaks, losing the retailer ~₹64Cr in deferred revenue across the two events.
Inventory accuracy was 71% — driving over-promise, under-deliver, and a 14-day cancellation tail.
9-month deadline to the next peak. Replatform was non-negotiable.
Our approach
We took the previous post-mortem and turned every red line into a SLO with an owner.
Built the OMS on Go services, Kafka for state changes, Postgres for the truth, Redis for the speed.
Chaos-tested against synthetic peak traffic 42× baseline. Found and fixed three load-shedding bugs before October.
Shadow-traffic for six weeks, full cutover in the off-peak window, with the old system kept warm for two months.
Results
Held at 42× baseline orders through Black Friday weekend with no degradation.
Zero P1 incidents through three holiday seasons, including the year of the platform migration.
Infra cost down 31% versus the previous OMS at higher throughput.
“They asked harder questions than our previous vendor in the first week than that vendor had asked in three years.”
Have a problem worth solving? Bring it.
Tell us what you're trying to ship. We'll come back inside two working days with people, a plan, or both.