Swapnil Kambli

• 99.99% — Availability
• 31% — Revenue Growth

As mobile orders increased, total work in progress (WIP) in the system grew without a corresponding increase in throughput, which inevitably led to longer lead times—particularly for walk-in customers. As a result, walk-in orders started ageing in the system, like cars stuck in an ever-growing traffic jam.

I balanced digital efficiency with in-person customer service, as both walk-in and mobile orders contributed to the WIP, while the throughput—number of baristas and their capacity—remained relatively constant.

I implemented the ability for in-store baristas and managers to “pause” or throttle mobile orders on iPads to manage overwhelming demand and control, allowing for better management of in-store throughput. I also enabled cafe order prioritization over digital orders to ensure an elevated in-store customer experience.

The focus was on elevating the in-store coffeehouse customer experience, which translated into prioritizing in-store orders over digital orders. I eliminated digital orders piling, while enabling baristas to focus on in-store orders.

• 1% — Projected Revenue Growth
• 25% — Customer Wait Times Reduction
• 40% — Demand Forecasting Improvement
• 50% — Food Waste Reduction
• 75%a — Store Item Availability
• 99.99% — System Availability

The challenge for this project was that daily and intra-day demand for bakery items is highly variable by SKU, time of day, and store location. There was also a risk of large-scale rollout without validation.

As a solution, I conducted 10-store MVP experiments, iterating model parameters and batch rules based on real-world KPIs before scaling. I also integrated the in-store-to-retail cloud API integration.

I improved baked item SKU availability to 95% and reduced food waste by over 50% through demand planning, boosting revenue.

• 30,000+ — Stores Virtualized in 7 Months
• 60 Million — In Assets Repurposed

The new store launch required expensive Dell PowerEdge DL20 hardware that was backordered, causing delays in store openings. Hardware footprint, while also replacing faulty hardware, was time-consuming and expensive.

I virtualized the store, running applications on a virtual machine, and repurposed the hardware for the new store launch. I also enabled fleet-wide store virtualization across 40,000+ stores and identified target hardware in stores globally.

I led the zero-to-one store virtualization at scale, virtualizing 30,000+ stores in seven months, repurposing around $60 million in assets.

Our order fulfillment platform occasionally experienced intermittent failures—like delayed or retried orders—whose root causes were hard to reproduce and trace in production. These gaps in resilience created risk for high-volume periods and made it difficult to guarantee SLA/SLO commitments. Moreover, chaos engineering was never performed prior to testing the service resiliency.

SOLUTION
• Introduced Azure Chaos Studio in partnership with Microsoft to proactively uncover reliability gaps.
• Defined critical failure scenarios, prioritized high-impact services, and developed step-by-step chaos tests simulating node failures, mocking service unavailability, service crashes, and order-routing retries.
• Partnered with SRE and infrastructure teams to run controlled experiments, validate fixes, and iterate, improving platform resiliency and reducing incident impact on order fulfillment.

I exposed a critical bug in the retry logic. By fixing the bug and improving failover handling, we measured resilience improvements via MTTR (45 minutes to 10), order fulfillment success rate (0.5% failed orders to 0.05%), and fewer Sev2 incidents.