Multi-Location Inventory System
ByteDance / Tokopedia · Senior Product Manager / Acting Squad Lead · Jul 2022 – Feb 2024
Problem
Sellers on the platform could only manage inventory from a single location. For sellers operating multiple warehouses or branches, this created stock inaccuracies, fulfillment delays, and missed sales. The platform had no infrastructure for multi-location inventory management.
Approach
I scoped the full system: data model changes, seller onboarding flows, inventory synchronization across locations, and downstream impacts on checkout, fulfillment, and pricing. I chose to build the data layer first and layer the seller-facing tools on top, rather than shipping a UI that masked a fragile backend.
Key decisions: we built location-level stock tracking (not just aggregate), integrated with existing fulfillment routing, and designed a migration path for existing sellers with zero downtime.
The judgment call
The team wanted to ship the seller-facing UI first — it was visible, demo-able, and would look good in the quarterly review. I pushed back. I had learned from endurance training that the foundation you skip in month one costs you in month six. We spent two extra weeks building the data reconciliation layer before touching the UI. Two engineers were frustrated. My manager asked why we were “behind schedule.” I showed them the data model and explained what would break without it.
Outcome
The capability grew into one of the platform’s most significant transaction drivers. Sellers using multi-location saw substantial increases in order volume and item value. It became a core part of the platform’s commerce infrastructure.
What I think about differently now
The hardest part of zero-to-one is resisting pressure to ship the visible thing before the invisible thing is solid. It is the same discipline as base-building in endurance training: the aerobic foundation is invisible and unglamorous, but it determines whether you can sustain pace at kilometer 35 or collapse at kilometer 25. Those two weeks of data layer work saved us months of reconciliation issues. I now build every product the same way: foundation first, features second.