In March 2020, what had been a two-day wait for grocery deliveries turned into a two-week ordeal. Hospitals trying to secure protective equipment found themselves in the same online queues as people ordering garden tools. The digital infrastructure had no way to distinguish between urgent needs and routine purchases, and millions of essential workers and at-risk families suffered delays that spanned from days to weeks.
The problem exposed a fundamental limitation in e-commerce architecture. Most major platforms operated on first-come, first-served logic, a system that worked fine during normal demand but collapsed when necessity and convenience competed for the same delivery slots.
The technical challenge wasn’t simply about capacity. It required rethinking how delivery systems categorize, prioritize, and route products when demand surges beyond available infrastructure. Traditional models couldn’t distinguish between a customer ordering hand sanitizer for an assisted living facility and someone stocking up on non-essentials. Every order entered the same queue.
Urvish Pandya, then a technical program manager working on delivery experience systems at a major e-commerce platform, saw the problem from an unusual angle. His role sat at the intersection of product development, engineering execution, and user experience design. When the pandemic hit, he was working on Business Unknown Promise (BUP) products, items for which delivery commitments couldn’t be calculated in advance due to supply chain variables. These products suddenly became the majority of platform inventory.
Building a priority engine inside existing infrastructure
The solution he developed didn’t require rebuilding the platform. Instead, he created a classification and routing system that could identify essential items and adjust their position within the existing delivery promise logic. The approach involved three technical components: product taxonomy refinement to flag medical supplies, groceries, and household essentials; algorithmic adjustments to delivery promise calculations that weighted essential items higher in the scheduling queue; and a monitoring layer that tracked fulfillment rates across categories to prevent the system from over-promising capacity.
“The constraint was time,” Urvish Pandya said. “We couldn’t rebuild the platform architecture. We had to work within the existing delivery systems and make them smart enough to understand context without adding latency to transaction processing.”
The technical execution required coordination across multiple teams. Engineering is needed to implement the classification logic without degrading system performance. Operations teams had to validate that the routing changes wouldn’t create bottlenecks at fulfillment centers. Product managers tracked whether the prioritization actually shortened delivery windows for essential goods without creating unacceptable delays elsewhere.
Implementation happened in phases over eight weeks. Early iterations focused on medical supplies and groceries, categories where urgency was clearest. The later stages brought in household essentials and fine-tuned the weighting system using actual fulfillment numbers. With millions of orders running through the system each day, even a small error could trigger widespread problems. The numbers told a different story within weeks. Medical supplies and groceries that had been taking 10 to 14 days started arriving in three to five days after the first month. The system adjusted on its own as inventory and warehouse capacity shifted across regions. Regular items still took longer, but now there was an intentional difference between what people needed immediately and what could wait.
Expanding capability through voice integration
Parallel to the delivery prioritization work, he led the technical execution of a voice SDK integration project that extended conversational AI capabilities beyond smart speakers into mobile applications. The work tackled a real constraint: voice assistants performed well on smart speakers but fell apart across mobile apps. Users had to jump between different apps just to play a song, place an order, or check when a package would arrive. Building a software kit that outside developers could actually use meant solving several problems at once. The SDK had to process voice commands, interpret what people meant, and respond quickly, all while running on phones with limited memory and processing power. Research shows most developers spend six to nine months building custom voice features from scratch.
His approach focused on standardization and simplicity. The SDK abstracted the complexity of voice processing into a set of APIs that developers could implement in days rather than months. The team validated the approach by launching integrations with multiple internal mobile applications, music streaming, video, and reading platforms, before opening the SDK to external partners. One early third-party integration involved a food delivery platform, extending voice ordering capabilities to millions of additional users.
“We treated it as infrastructure rather than a feature,” Urvish Pandya said. “The goal was to make voice interaction as straightforward to implement as adding a search bar. Developers shouldn’t need to understand the underlying machine learning models or audio processing pipelines.”
Both projects point to something bigger happening in tech organizations. Instead of building one-off features, engineering teams now focus on creating systems that can change without needing to be rebuilt from scratch. The pandemic pushed companies to speed up plans they had been sitting on for years. McKinsey’s analysis found most businesses compressed three to four years of digital changes into a matter of months, mostly because they had to adapt what they already had instead of waiting to build something new.
This kind of work needs different skills than typical product or engineering jobs. You have to know the technical setup well enough to spot where small changes can make a big difference. You need to get teams that rarely talk to each other working together. And you have to be comfortable with solutions that aren’t perfect but actually work, given real limitations.
His work bridging delivery systems and voice technology shows what this looks like in practice. His experience covers software management, technical design, user experience, and quality testing, areas that usually stay separate in most companies. That breadth helped him see connections others missed and pull together solutions that needed multiple teams working in sync.
Over three years, he earned the top performance rating, “Exceeds High Bar” for both 2020 and 2021. He also received stock awards for three straight years, recognition given to people who deliver results well beyond their job description. He shipped more than ten major customer features and programs, all on tight deadlines during a time when most companies were just trying to keep things running.
The need to work across traditional job boundaries keeps growing as platforms get more complicated and users expect more. The pandemic created problems that couldn’t wait for ideal solutions. Companies that moved fast had a few things in common: they let people make decisions without endless approvals, they accepted that first versions would need refinement, and they judged success by what actually helped users rather than technical perfection.
Future problems will probably need the same approach. Supply chains still break easily, people use more devices and apps than ever, and technology changes faster each year. Companies that can find engineers who understand whole systems, work across different areas, and deliver real improvements despite constraints will handle whatever comes next better than those who can’t.
Be First to Comment