2022 Franz Edelman Competition: Alibaba

Ashley Kilgore:

This podcast is part of a special series featuring the 2022 finalist teams for the informs Frans Edelman award for achievement in advanced analytics, operations research, and management science. The most prestigious award for achievement in the practice of OR and advanced analytics. For more than four decades, the Edelman award has recognized contributions that are transforming how we approach some of the world’s most complex problems. Finalists for the Edelman award have contributed to a cumulative impact more than $336 billion since the awards’ inception, as well as countless other non-monetary benefits.

The winner of this year’s award will be announced to the 2022 informs business analytics conference held April 3rd of the fifth in Houston, Texas. Joining me for this episode, are Xinhui Zhang, senior director and head of supply chain management and operations research committee and Yuming Deng, director of digital supply chain to discuss their team Alibaba’s finalist entry. Alibaba, which aims to build the future infrastructure of commerce has designed many multi-platform retail business models. These range from mobile apps to brick and mortars and more. Merchandise on Alibaba’s platforms covers general supplies to fresh produce.

Each of these different channels in their products brings unique features regarding demand forecast and inventory management. These challenges are being solved through a series of algorithms to align supply with demand, which in turn has a significant impact on the business. Xinhui, Yuming, thank you for joining me. I’m looking forward to discussing your team’s work. To start, can you share a little bit more background at Alibaba? What type of services does it provide? And what is the scope of its operations?

Xinhui Zhang:

Alibaba is one of the largest eCommerce companies in the world. And as mentioned, it has designed many multi-platform or omnichannel retail business models. These models range from mobile applications to brick and mortar stores. The purpose is to allow customers to either shop at the store or place online orders and have them delivered to their homes as a designed client. To get a feeling of some of the business models, I would like to give you a few examples.

Fresh people has over 300 brick and motor stores across more than 20 cities. Allows customer to shop for grocery is their offline at one of those stores, all can place online orders for grocery and have their orders deliver to their homes in 30 minutes, if they are within three kilometers of a store. T-Mobile allow customers to shop online for general supplies and have them delivered at their doorsteps by the same day or the next day. At its core, Alibaba aims to transform these traditional retail business digitally provide more options for customers and increase revenues for these businesses.

Ashley Kilgore:

And now what were the challenges facing Alibaba that your work was undertaken to address?

Xinhui Zhang:

Retail is an important part of our economy and has screened many fascinating problems to be solved in the supply chain and logistics discipline. It provides a [inaudible 00:05:45] in for example, forecasting inventory, pricing, equipment planning, et cetera. By the way, we have discussed some of our efforts in warehouse operations and logistics, such as last mile delivery through the solution of variants of vehicle routing problems in the 2021 [inaudible 00:06:10] competition. In this podcast, we are focus on the supply chain site.

The various omnichannel business models bring many opportunities as well as challenges such as weak season or promotion, driven demand patterns, interaction among products, complex inventory operations such as the mixture of inventory retrieval patterns, real time price markdowns with inventory availability, considerations, et cetera. Here we present three challenges. First, retail demand is affected by manufacturers such as weekly patents, seasonal patents, holidays, weather and temperature. Various promotions such as buy one, get one free 20% off, et cetera, as well as complicated interactions such as complimentary and substitution relationships.

For example, changes in pork price affect the sales of B because these products substitute each other. Changes in tortilla chips affect the sales of salsa because these products are often brought together. To capture these factors, to learn and quantify the interaction of amounts, a large set of products have post significant and challenges to the forecasting algorithms. Second, the behavior of products inventory, especially though with short life cycles, such as fresh produced are affected by customer selection patterns.

For example, in store purchases by customers might exhibit a last in, first out pattern yet online orders by pickers could accept the first in, first out pattern. Computation simulations show dramatic difference in inventory settings under these two patterns and the stools overall inventory retrieval pattern could be a mixture of the two. These pictures have a significant impact on inventory management. Define traditional inventory models and are of very importance to the inventory management system.

Third, In practice is impossible to match the buy chain was demand. So, timely markdowns and use of recommendations in channels could be critical. Essentially, we want to change the position of product in recommendation system to direct product with more inventories, more often without recommending products people don’t want.

These considerations are seldom used in traditional recommendation systems, provide fresh levers or opportunities to mitigate these supply and demand mismatches. Yet the modeling and solution of these problems are seldom being addressed. These are online optimization problems and are very hard to solve. These are some of the problems or challenges we face and that these are more problems or challenges to be discovered.

Ashley Kilgore:

Could you share what your approach was to address these challenges? What different factors did you have to take into account?

Xinhui Zhang:

The operations research teams at Alibaba, especially the digital supply chain and fresh EO teams have developed a series of models and algorithm to address these challenges. Especially, we have developed deep learning based forecast, simulation based inventory optimization, online markdowns and product ranking optimization in recommendation system. I would like to introduce my colleague Yuming to start sharing some of these techniques and algorithms.

Yuming Deng:

Thanks Xinhui. First, in the area of forecasting, we started developing forecasting algorithms as early as 2014. Over the years, we have developed statistical approaches, machine learning approaches, and most recently deep learning forecasting models. The reason is we want to borrow the capability of deep learning structure to model the various nonlinear function, decompose time series into various blocks, representing channed, seasonality price, structure and promotion effect.

We also use a deep learning network to represent the interaction among products. The model represent the breakthrough in forecasting techniques and have achieved superior performance. Our results have shown to be able to provide two to 10% forecasting accuracy improvements relative to traditional statistical and states of art, deep learning forecasting models. Second, in the area of inventory optimization, we have developed simulation based inventory optimization algorithms. Instance, we use simulation to sample various empirical demand, distributions and mimic complex customer consumption patents.

For example, customers picking the freshest products, pickers picking randomly products winging the shelf life ordering calendars, et cetera. Separation allows us to model almost any practical inventory practice and represent a true reflection of reality. We use an effective iterative search technique with novel inventory moves and have dramatically reduced the computational time. In sense, we use simulation optimization to translate an otherwise very mathematical and [inaudible 00:12:55] inventory management problem into an optimization problem.

Simulation optimization results are vital and appealing compared to traditional analytic results, making them accessible to business operators. Third, in the area of recommendation systems, we have developed a series of planning and real time online optimization models and integrated them into personal recommendations.

We developed to use price elasticity across various promotion strategies, such as buy one, get one free, 20% off, 10 of out of 10, $100 and optimize both product ranks and price markdown inventory cautions online item placement ranks based on real time inventory availability. The key here is real time computing and as optimization folks, we were able to do this wrong. This has significantly reduced both shrinkage and auto stocks and at the same time, improve service levels, providing a winging for both customers and retailers.

Ashley Kilgore:

And could you share what was unique about your approach?

Xinhui Zhang:

I think our approach is unique in that it addresses many of the practical problems in retail where demand patterns are not of the typical normal of a song, where productive demand affects each other, where inventory practical retrieval patents could be first in first out last seen first out or randomly of any empirical distributions. These are very important features that have to be addressed in the supply chain of a retail company. Yet we believe the skill are not receiving the full attention from the operation research design.

We believe this is one of the first approaches to solve these problems. We also believe these algorithm are general and are very critical to retail. Grocery products usually have short life cycles, actual inventory at the end of the shelf life, will be disposed of and represent a significant waste. Proper management of the inventory of these products is critical to the overall profit of a retail company. Simulation optimization provide an elegant approach to the solution of these problems that are otherwise and not solvable through any analytical approaches.

We also believe the interaction of product [inaudible 00:16:20] optimization in recommendation system provides some unique opportunities in omnichannel retail, where we could elegantly utilize online search and recommendations to effectively incorporate inventory availability and supply chain capabilities. It is our experience, it offers great opportunity to align supply chain and demand in real time and provides many opportunities for omnichannel retail.

Ashley Kilgore:

And now since implementation, what has been the impact for Alibaba?

Xinhui Zhang:

These algorithms have been deployed in almost all retail business subsidies in Alibaba. At the time of this writing. We became to see more and more business realize the importance of supply chain, and the role of Christian research plays in this process. Since the forecasting inventory and recommendation systems have been implemented, many others used to be placed by inventory personnel have been automated and has result in much higher efficiency and accuracy. Since the implementation of the algorithms in the past two years, both shrinkage and out of stock rates have been steadily and significantly reduced translating into tons of millions of dollars savings in cost annually.

We believe inventory management is very importantly in retail, especially grocery and the fresh produce retail. These algorithms had a direct impact on revenue and profit margins and played a key role to achieve business successes. Overall, these algorithms have achieved significant financial savings with the annual financial benefit of more than hundreds of millions of dollars reduction in shrinkage and inventory costs while maintaining or increasing the service levels in all Alibaba retail business and increases our revenue.

Ashley Kilgore:

Xinhui, Yuming, I want to thank you both again for joining me and wish you the rest of your team, the best of luck in the 2022 Fran Edelman competition. Are there any final thoughts you’d like to share regarding Alibaba’s finalist project?

Xinhui Zhang:

We think retail is a field as fascinating and rewarding to the operation research community. There are still many more problems we’re facing and we haven’t really fully addressed. Retail is also undergoing dramatic changes at the same time when we encourage more operation researchers to delve into this field, to make operation research more applicable and produce more impact to our society.

Ashley Kilgore:

Want to learn more? Visit resoundinghuman.com for additional information on this week’s episode and guest. The podcast is also available for download or streaming from apple podcasts, Google play, Stitcher and Spotify. Wherever you listen, if you enjoy Resounding The Human, please be sure to leave review to help spread the word about the podcast. Until next time, I’m Ashley Kilgore and this is Resoundingly Human.