2-hour delivery platform
across 5 cities

Amazon's same-day promise was held together with duct tape

When I stepped into this role, the same-day platform was fundamentally fragmented. Grocery lived on a separate surface — customers had to navigate into a completely different store, creating a jarring experience. There was no time-slot flexibility, and warehouse throughput was the daily chokepoint strangling order volume.

The quick commerce opportunity was massive — customers were going to Zepto, Blinkit, and Swiggy Instamart for the speed promise. Amazon had the inventory depth and logistics muscle, but the infrastructure wasn't built to hold the promise. A 2-hour delivery window starts at the warehouse door, not the customer's click.

The problem wasn't one thing to fix. It was a complete rebuild — topology, warehouse operations, last-mile routing, and demand intelligence — all simultaneously.

A full-stack quick commerce engine — six interlocking products

This wasn't a feature sprint. I treated it as a platform initiative across six distinct product workstreams, each feeding into the others:

• Topology Product: Used demand heatmaps at geocode level to identify optimal warehouse placement for the SSD network. Right warehouse in the right catchment area is the unlock — no routing algorithm fixes a bad topology.
• Pick-by-Order (WMS): Replaced batch-pick warehouse operations with order-by-order pick flows. Eliminated the staging bottleneck that was adding 45+ minutes to every order's warehouse journey.
• Segregated Stowing: Redesigned how inventory was placed in the FC — quick-commerce eligible SKUs stowed in dedicated zones for instant access by pick-by-order flows.
• Continuous Routing: Replaced periodic dispatch batching with real-time order-to-driver assignment, cutting the routing window from 40 minutes to near-instant. Combined with Pick-by-Order, this took warehouse processing from 2 hours to 30 minutes.
• MMOT (Multi-Modal Transport): AI-powered routing engine dynamically matching orders to bike, van, or truck based on geocode, product size, and real-time fleet utilisation. Eliminated over-dispatching large vehicles for small orders.
• Driver App (Android Migration): Migrated the driver-side app from a fragile iOS-only build to Android, unlocking 25,000+ organic signups from drivers already on Android. More drivers meant better route density and faster delivery SLAs.

Grocery integration vs. separate surface — the decision that moved CVR

The biggest strategic call in this entire initiative: integrate grocery directly into the SSD app, or build a separate quick-commerce-specific surface?

The case for separation was compelling: faster to ship, lower technical risk, no cross-team dependency on the main app team, cleaner UX story for the grocery persona.

I argued for integration — and had to fight for it.** The reason: separate surfaces create customer context-switching. A customer browsing Amazon for a phone charger shouldn't have to "go to the grocery app" for milk. That friction is the silent CVR killer. Every app-switch is a lost basket.

We integrated. The result: +18% conversion rate and a double-digit reduction in total shipments per order — because a single basket now ships as a single delivery. Lower cost, lower carbon, better customer experience.

The numbers that came out

• 57% faster delivery — warehouse processing cut from 2 hours to 30 minutes through Pick-by-Order + Continuous Routing
• 346M+ orders/year fulfilled through the platform across India
• 25,000+ SKUs integrated into the quick commerce network including grocery, electronics, and essentials
• 25,000+ driver signups through Android app migration, improving route density and last-mile SLA
• 5 markets live — India, UAE, KSA, Qatar, Japan — on a single platform architecture
• India's first 2-hour delivery promise at this scale — a competitive moat that Zepto, Blinkit, and competitors have since tried to replicate

Other projects from this era