This Partner Powerup Series session features Anuj Matal, Chief Architect at Flipkart Commerce Cloud, discussing why traditional single-tenant order management systems create unsustainable costs and complexity. Learn how multi-tenant architecture enables retailers to support B2B, B2C, marketplace, and quick commerce on a unified platform handling millions of daily orders.
Major themes: multi-tenant architecture, single-tenant hidden costs, B2B vs B2C unified platform, inventory fragmentation, order orchestration vs optimization, real-time inventory aggregation, AI-driven order routing, omnichannel architecture, event-driven systems, microservices challenges, eventual consistency.
The discussion reveals that running separate OMS instances for different business models creates 3x training costs, project feature implementations, compliance overhead, and slower innovation cycles. Flipkart's multi-tenant approach supports dramatically different use cases (tracking individual iPhone IMEI numbers vs bulk brick orders) on the same core primitives through configurable policies.
Q: What are the hidden costs of running separate single-tenant OMS instances?
A: The real costs aren't license fees—they're training operations teams twice, maintaining feature parity across N instances (every feature becomes N projects), compliance and QA duplication, and slower release cycles. Like Southwest Airlines' Boeing 737 strategy, running three different systems for the same purpose costs 3x without 3x benefit. Innovation slows to a crawl.
Q: Why are companies replacing their order management systems?
A: The $4 billion OMS market is growing 7% yearly with 30% seeking replacements. Reasons: traditional monolithic stacks not web-scale, inability to handle diverse business models (B2B, B2C, marketplace, quick commerce), band-aid solutions over fundamental limitations, and need for optimization (not just orchestration) at each inflection point.
Q: What is multi-tenant OMS architecture?
A: Multi-tenant architecture runs dramatically different business use cases on shared core primitives with configurable policies. Flipkart supports selling books, 10-minute iPhone delivery, fresh vegetables, air tickets, and influencer commerce on one platform. Same order management "brain" handles tracking individual iPhone IMEI numbers (B2C compliance) and bulk brick orders (B2B efficiency).
Q: What is the Shopkeeper Test for evaluating OMS platforms?
A: The Shopkeeper Test asks: "What input do I need to sell one pen to one customer instantly?" If the answer is complex, the system has inherent bias that will create bespoke customizations as business grows. This "hello world" test case should be trivial—if it's not, the platform wasn't designed for flexibility.
Q: What is the Monsoon Mumbai Test for OMS capabilities?
A: Real-world scenario: 30-minute delivery promise, one hub has pen (no notebook), another has soft-cover substitution, third has both but rain delays delivery. Can your OMS make optimal decisions based on different policies in the same scenario? Can it leverage AI with past data? Testing if platform provides different outcomes for different policies validates flexibility.
Q: How does Flipkart handle inventory fragmentation across business models?
A: Traditional OMS creates fragmentation like OS memory fragmentation—inventory unused in one context while needed in another. Flipkart's "single brain" approach aggregates real-time inventory across all sources, channels (owned/marketplace, B2B/B2C), labor capacity, and fulfillment rates. Event streams feed one place for global optimization, not local silos.
Q: What OMS limitations appear in the pre-order journey?
A: Cracks show before order placement: available-to-promise vs available-to-ship inventory gaps, real-time aggregation failures across fragmented channels, and huge gaps between order promises and fulfillment ground reality. Band-aid fixes multiply exception scenarios instead of addressing root cause: lack of unified inventory view.
Q: How should OMS systems be bidirectional, not unidirectional?
A: Traditional OMS dictates ground operations ("software says you can only do this"), passing system constraints to workforce. Modern OMS must surface ground realities upward—business policies and rules decide workflows, system is just enabler. Field data, workforce capacity, real-time conditions should inform the system, not vice versa.
Q: What scale does Flipkart Commerce Cloud handle?
A: Infrastructure: 1.5 million CPU cores, 50-60 petabytes RAM, 1000+ petabytes disk, 40Gbps network traffic. Performance: millions daily orders, 3,000-4,000 orders per second at peak, million+ concurrent browsers. Use cases: books to 10-minute delivery (groceries, iPhones), marketplace, fashion, electronics, refurbished goods, tickets, influencer commerce.
Q: How does AI change order management from optimization to autonomous decisions?
A: Traditional optimization used configuration rules and DSLs. AI enables optimization without human-in-loop: empty store automatically receiving delivery orders from overloaded nearby store during sales, maintaining promise dates within cost guardrails. Hybrid approach: logic acts as guardrails, AI provides outcome options, system chooses optimal path.
Q: Why can't traditional OMS handle B2B and B2C tracking requirements simultaneously?
A: B2C compliance requires tracking three iPhones individually (unique IMEI numbers for returns), but B2B 1,000 iPhones makes IMEI tracking overkill. Same with 1,000 bricks vs 10 diamond sets (track individually) and 100 tomatoes (no individual tracking). Traditional systems can't dynamically adjust granularity—AI-driven policy engines can switch tracking levels based on product value and use case.
Q: What is the noisy neighbor problem in multi-tenant OMS?
A: When iPhones sell as hot cakes on same stack as fashion/grocery, how do you guard other verticals' experience? Flipkart solved this through containerization revolution, horizontal scaling without hesitation, orchestration keeping API contracts same while isolating workloads. Functional extensions over common core layer as use cases mature.
Q: How does Flipkart convince leadership to allow market disruption?
A: Use "reference frame" concept from physics—explain in business language they understand. Show 0-to-1 journey possible but 1-to-10 requires different approach. Communicate tech debt curve where business becomes unsustainable. Reasonable people want success—tech leadership must explain business growth vs loss clearly. Ownership lies with tech leadership, not business.
