Reflecting on the 2025 Franz Edelman Competition with Flipkart

Ashley K:

Welcome to the final episode in our special series of resoundingly human podcasts, highlighting the finalist teams for the 2025 Franz Aleman Award, the Nobel Prize of Analytics. These finalist projects represent teams from around the world who have leveraged advanced analytics to transform their organizations, address their most significant challenges and better serve their customers and communities. Prior to the 2025 Analytics Plus conference, I had the opportunity to sit down with five of the six teams, and today I’m joined by members of the team representing Flipkart to share some insight on their incredible work. Thank you all so much for joining me. We’ve already had the opportunity to meet in person at this year’s Analytics Plus conference, but I’d love to kick things off by having you briefly introduce yourselves and the Franz Edelman Award finalist project you were recognized for.

Vikas Goel:

Hi, I’ll go first. My name is Vikas Goel. I am senior director for data science at Flipkart.

Gowtham Bellala:

Hi, I am Gowtham Bellala. I’m a principal data scientist at Flipkart.

Prateek Agrawal:

I am PR Kal. I’m director of product for supply chain and inventory planning here at Flipkart. We quickly talked to you about introduce our project to you. So our project focuses on how we revolutionize the supply chain operations at Flipkart, which is India’s largest e-commerce company by building a central planning platform, being an e-commerce company, our bread and butter, of course is procuring and delivering products to our customers. And supply chain planning plays a key role in enabling us to deliver these shipments at a fast speed, at low cost, and at a high reliability to our customers. We started this initiative back in 2020 at a time when our planning process was quite manual and all the different pieces were very loosely coupled. There are a lot of inefficiencies which limited our ability to respond effectively to the changing market dynamics and as the diversity and size of the flip card business grew, this approach was not really able to keep up with the increasing complexity.

Different teams were often shooting in different directions, and every time the size would increase, we would need to supplement and add more people to the team. All of this came to a head in 2020 with the pandemic where a massive jump in e-commerce penetration happened in India. This coupled with regular and unpredictable supply chain disruptions and the challenges of coordinating with people working from home tested the entire manual planning process to the limit. We realized that we needed a centralized planning platform to overcome these challenges, and we spent the last three to four years building this out. We’ve leveraged machine learning and operations research to transform our supply chain and delivered significant cost savings and incremental net sales for the company. All of this during a time of rapid growth and evolution for the business due to the pandemic along the way, we’ve also enabled around a 50% increase in our one day deliveries and enabled Flipkart to compete in the rapidly growing 10 minute delivery market.

Ashley K:

Did you use analytics in any unique or unexpected ways to approach this challenge?

Vikas Goel:

So before I get into that, maybe I’ll talk a little bit about the platform that we built. Like Pratik said, we built a central planning platform and it has helped transform supply chain planning at Flipkart. So this platform actually has two layers. There’s a forecasting layer, which helps us to predict the demand of our products accurately at different granularities using deep learning and machine learning methods. The second layer is what we call an optimization layer. This includes a set of optimization models that optimize the three key sets of decisions in supply chain planning, which include planning capacity at different nodes in our network planning how much inventory we want to have at each node in our network and network flow planning, which is about planning how shipments move through our network. So together these two layers and they are our overall platform ensures that our overall supply chain is fully optimized in terms of speed, cost and reliability to our customer.

So now coming back to the question that you asked, which was what is unique about how we used analytics in this platform, and I’ll say a couple of things. So first, our platform combines the predictive power of machine learning with the optimization prowess of operations research, and together this combination helps us make optimized decisions at every stage in our supply chain. That’s one. Secondly, we didn’t just build one model, we built an entire platform that integrates a collection of models that optimize decisions at every stage in the supply chain. And these decisions aren’t just point in time decisions. These are decisions, these are models that optimize decisions at a monthly scale all the way to real-time decisions. So it enables optimization all across our platform. The third thing I would say is a lot of the work that we’ve done is highly innovative. We’ve used state-of-the-art deep learning models for forecasting, and I’ll give also an example from the optimization site at the conference.

I think there was a conversation between a few senior leaders that I overheard about how stochastic programming is a great technology, but it’s very, very difficult to actually land it into something useful. One of the models that we’ve built and we presented at Deman competition is based on stochastic programming. It essentially runs every 15 minutes at every one of our warehouses, and it’s used to allocate capacity so that we maximize delivery speed for every shipment, right? So it’s actually influencing every shipment in our entire supply chain. That’s kind of the highest possible scale that we could have thrown at stochastic programming, but we were able to make it work and it works pretty smoothly for us and it’s driven a lot of value for our company.

Ashley K:

So how did your team balance technical complexity with ensuring the solution was practical and implementable on the ground?

Gowtham Bellala:

So I’ll highlight a few ways we achieved that. Right, so first we designed our solution focusing on adoption as an example, as because mentioned demand forecasting was one of the core components of our platform. We started with simple statistical models such as Arima very early on in our journey and focused heavily on business adoption. As our systems and processes matured and the business adoption of our forecasting platform increased, we then pivoted to using machine learning models and more complex deep learning architectures for enhanced forecasts. Second, we also focused on interpretability as an example, our inventory, inter warehouse inventory rebalancing solution. It is based on a constrained optimization model. We build explicit routines and reports that help explain the model output. Specifically, we generated a report to help the operations team who use the model output to understand what fraction of the inventory movement is being limited by each constraint. This helps the operation team get an intuitive understanding of the output and also provide an actionable view. And third, we set up well-defined governance process. We continuously monitor the platform and model metrics. As an example, in the case of our forecasting model, we compute the error in our forecast after the fact, after the demand, after we get the demand, we complete the forecast errors. This view is generated at different granularities and cuts and it ensures full transparency and accountability, and by generating these model metrics, it also serves as a continuous feedback for our tech teams to identify where the systems are degrading and also to improve on those degradations.

Prateek Agrawal:

I think just adding to what Gotham said, right, a big aspect apart from the approach was also the team structure itself. The product engineering, data science and business teams were all working together, often sitting on the same floor right by each other and working iterating very quickly. So it wasn’t a case where these teams would meet every two weeks or every two months, and where somebody would create a big project plan and come back with doing status subjects. These were a bunch of colleagues that were sitting together solving a problem together. So it made sure that the solutions were grounded. It made sure that the business teams took ownership of the solutions and made sure that they were implemented on it.

Ashley K:

Could you share some of the tangible benefits or improvements Flipkart has experienced since implementing your solution?

Prateek Agrawal:

So the platform has been transformational for Flipkart It has enabled significant cost savings and incremental let’s sales for us. Some of the factors that have enabled it is we’ve enabled faster delivery speeds across our network and protected speeds for our high priority customers, delighting these loyal and high priority customers. We’ve driven lower supply chain costs by improving the manpower utilization at our warehouses. We’ve reduced our inventory costs by a reduction in unhealthy inventory and working capital, and most importantly, we’ve had more agile and efficient planning with a drastic reduction in planning cycle types, not just that it’s essentially enabled our planning teams to scale their impact efficiently despite rapid growth. In the flip card business, as I mentioned earlier, whenever the business grew earlier, we would just go ahead and start expanding the teams to plan for this. But in some of the new business lines, we’ve been able to scale our operations five x in the last six months without having to add a single person. This systemization and automation, it has led to significant reduction in run times of our cycle. Some of our processes that used to take up to 10 days are now taking a day enabling us to run these much more often and overall improving the accuracy of our plan. Overall, these improvements have helped flip car grow in a sustainable manner, which has been critical to the driving the future of flip car.

Ashley K:

Before we wrap up, are there any final thoughts or even a story you’d like to share about the experience of competing in the Franz Edelman award competition?

Vikas Goel:

Yeah, I have a couple of thoughts. So see, I have a background in or so I used to see Edelman papers in interfaces when I was in grad school, so it was great to participate in one of these competitions all the way to the end. It required quite a lot of effort and we got a lot of help from our mentors. I think one of the great outcomes of this for me was that it actually helped me synthesize the amount of work that we had we had put together over the last three, four years, or our teams had done over the last three, four years. So not all of which I think I hadn’t realized the gravity of all of that before we did this, so that made me feel very proud. At the same time, I think going to the conference was both humbling and inspiring. There were some really, really good presentations, both at the Edelman competition and at the other competitions that were going on. There were presentations from different parts of the world, different industries, the USA sports, the USA cycling team. So I mean, looking at the kind of work that people were doing in a very applied setting, dealing with the challenges and still doing very innovative, technically challenging work, that was very, very inspiring and motivating

Prateek Agrawal:

This adding to that, I think so completely opposite to that, because for me it was my first informed conference or any such conference of this magnitude, right? I think for me, there were two experiences that define me right? One was just the entire rollercoaster of the competition where I started off by saying, okay, have we actually done anything which is worth applying to getting everybody together, scrambling to meet the deadline, giving a submission, which makes you pause and think, oh wow, I think we have something here. To then going and finally polishing our presentations, presenting it, where the first thought that came after I stepped down from the stage was, we can really win this. And then following that with five different presentations, which made me realize that how strong everybody’s submission was and also how much there was to learn and explore back at clip card from what I’d seen of those five presentations to finally the gala where while we did not eventually win, all of us got together and celebrated the victory of the USA cycling team, as well as I think the overall went away feeling really well and really happy of having just been a part of this journey, and now we’re back to the grind, taking all the lessons that we learned at informs and figuring out how we can apply that to Flipkart.

So it’s been a whole journey for us and I feel really richer for the experience. So that’s one part of it. The other part, I think just meeting, seeing how different people were applying analytics in such a varied number of applications, and this is when the USA cycling team being a prime example of the same was really inspiring. It’s really made me start thinking a lot more deeply about what are the places that we end up ignoring where we can actually apply analytics and decision sciences.

Ashley K:

I want to thank you again so much for joining me. It was such a pleasure meeting you all in Indianapolis for the competition, and I enjoyed the opportunity to share this special insight on your incredible finalist project. Thank you so much.

Prateek Agrawal:

Thanks a lot, Ashley. It was a pleasure talking to you. Thank you.

Ashley K:

Prateek Agrawal:

Thanks, Ashley.

Gowtham Bellala:

Thanks, Ashley.