Data Science Center - Purdue Business

HIGHLIGHT Sales Forecast Calibration Amid the COVID-19 Pandemic for AB Inbev

In 2020, Simchi-Levi and researchers collaborated with AB Inbev to develop a new online non-parametric regression method to calibrate sales forecasts during the COVID-19 pandemic period.

The new method brings together two different approaches, online learning and pandemic modeling, to dramatically improve forecast accuracy. Both the team and AB Inbev independently tested the new method in various business scenarios. When compared with other methods (including the previous approach used by AB Inbev and another approach based on online linear regression) the online non-parametric regression method reduced the forecast error by over 30% in forecasting sales volumes.

For more details, see D. Simchi-Levi, R. Sun, M. X. Wu, and R. Zhu (2020), "Calibrating Sales Forecast in a Pandemic Using Online Non-Parametric Regression Model."

Supply Chain Automation Powered by LLM

Supply chain management requires addressing a variety of complex decision-making challenges, from sourcing strategies to planning and execution. Over the last few decades, advances in computation and information technologies have enabled the transition from manual, intuition-and-experience-based decision-making to more automated and data-driven decisions using a variety of tools that apply optimization techniques.

These techniques use mathematical methods to improve decision-making.

Unfortunately, business planners and executives still need to spend considerable time and effort to:

understand and explain the recommendations coming out of these technologies
analyze various scenarios and answer what-if questions
update the mathematical models used in these tools to reflect current business environments.

Addressing these challenges requires involving data science teams and/or the technology providers to explain results or make the necessary changes in the technology, and hence, significantly slows down decision-making.

Motivated by the recent advances in large language models (LLMs), we report how this disruptive technology can democratize supply chain technology — namely, facilitate the understanding of tools’ outcomes, as well as the interaction with supply chain tools without human-in-the-loop. Specifically, we report how we apply LLMs to address the three challenges described above, thereby substantially reducing the time to decision from days and weeks to minutes and hours, and dramatically increasing planners’ and executives’ productivity and impact.

Read our HBR article, How Generative AI Improves Supply Chain Management, where we report implementation at Microsoft Cloud Supply Chain.

You can also review our HBR article, When Supply Chains Become Autonomous, on the application of LLM and generative AI for autonomous supply chain management.

The Resiliency of the Global Semiconductor Supply Chain

In 2021-2022, Professor Simchi-Levi and his PhD student, Feng Zhu, collaborated with Denso Auto Parts, the world’s second-largest automotive parts manufacturing company, to understand risk exposure in semiconductor supply chains.

In this research, we analyzed rich data sets from the semiconductor industry and applied the supply chain resilience and stress-test technology developed by the MIT Data Science Lab. Because of the unique challenges of this industry, we introduced a new concept, referred to as Time-to-Recover Inventory (or TTRInv), that measures how long it takes the supply chain to return to normal (target) inventory levels after a disruption. Our analysis shows, for example, that a short disruption of a semiconductor fabrication facility, or “fab,” in Taiwan for 10 days, could cause a flurry of additional disruptions and shortages across the entire supply chain that would last almost a year. Our research also reveals that expanding the number of semiconductor fabrication facilities in the United States alone will not suffice to prevent such shortages from occurring again. Equally important, our analysis, insights and recommendations could also be applied to supply chains for other products, such as the one for batteries and magnets used in electric vehicles (EVs).

For more detail, see David Simchi-Levi, with Feng Zhu and Matthew Loy, Fixing the U.S. Semiconductor Supply Chain (hbr.org).

Trading Inventory Cost for Online Sales Revenue

In 2021, Professor Simchi-Levi and researchers collaborated with Accenture and Home Depot to study an inventory placement optimization problem under the online retail setting, where demand is sensitive to service response time.

To predict the sensitivity of demand to response time, the team developed a novel demand prediction and elasticity model for different product categories. Such a model may suggest products that need to be positioned closer to market demand. Unfortunately, such a strategy will increase inventory cost. To address this challenge, a new method based on data-driven stochastic programming was devised that optimally trades safety stock for service response time. The efficiency of the approach was demonstrated through data provided by one of the largest e-commerce retailers in North America. The new approach led to a more than 10% increase in total profit. Our approach offers supply chain managers a general-purpose decision support tool that optimally positions inventory in the supply chain and generates recommended stocking levels for stores, distribution centers and warehouses on a daily basis.

For more details, see H. Qin, D. Simchi-Levi, R. Ferer, J. Mays, K. Merriam, M. Forrester, A. Hamrick (2022), "Trading Safety Stock for Service Response Time in Inventory Positioning."

Smart Supply Chain Digitization

In 2020, Professor Simchi-Levi, together with Accenture Executives and PhD students from the MIT Data Science Lab, developed a practical supply chain digitization strategy.

The strategy includes three important components - a unified, single, view of demand; supply chain segmentation; and smart planning and execution - all of which are powered by Digitization, Analytics and Automation. This strategy was implemented in a variety of industries including fashion retail, Consumer Packaged Goods manufacturers and high-tech.

For more information, see D. Simchi-Levi and K. Timmermans (Accenture), Deep Transformation with Smart Supply Chain Digitization. See more here.

Model Misspecifications for Oracle Technology

In 2017-2018, Simchi-Levi and his PhD students collaborated with Oracle Retail to develop and test a dynamic pricing algorithm for fashion retail.

The goal was to integrate the pricing algorithm into Oracle Retail’s price optimization products, which mainly optimize markdown pricing for fashion retailers. The proposed algorithm, referred to as the Random Price Shock (RPS) algorithm, includes features such as the ability to correct for a mis-specified model; feature-based pricing; learning and earning on the fly; and a variety of business constraints.

For details, see Nambiar M., D. Simchi-Levi and H. Wang (2019), "Dynamic Learning and Pricing with Model Misspecification." Management Science, Vol. 65, No. 11, November 2019, pp. 4980–5000.

Pricing at Groupon and Rue La La

In 2017, Simchi-Levi and his PhD students collaborated with the two companies and developed new models that integrate machine learning and optimization techniques to optimize pricing decisions, increasing Rue La La’s revenue by about 11% and Groupon’s by about 20%.

More on this in Simchi-Levi, D. (2017), "The New Frontier of Price Optimization." Sloan Management Review, Fall 2017, pp. 22–26.

Customized Offerings in the Airline and Insurance Industries

In 2015, Simchi-Levi and his PhD students developed a new method, the Contextual Treatment Selection (CTS) algorithm, for optimizing customizing offering based on individual characteristics.

Initially, the algorithm, which combines machine learning and optimization techniques, was implemented in the airline industry to offer online customers ancillary products such as priority boarding, seat upgrades or car rentals. Later, the same algorithm has been applied in the insurance industry to offer car, home and life insurance to maximize customer lifetime value.

For details, see Zhao Y., X. Fang and D. Simchi-Levi (2017), "Uplift Modeling with Multiple Treatments and General Response Types." SIAM Data Mining 2017, pp. 588-596.

Supply Chain Risk Management at Ford

In 2014, Simchi-Levi and his Phd students developed a new way for companies to identify hidden risk in their supply chain by employing new concepts, referred to as Time-to-Recover and Time-to-Survive.

The new approach, which was initially implemented at Ford Motor Company, allows management to understand the impact of a disruption originating anywhere in the firm’s supply chain and quantify it using operational and financial performance metrics.

See more in Simchi-Levi, D., W. Schmidt, and Y. Wei (2014), From Superstorms to Factory Fires: Managing Unpredictable Supply Chain Disruptions. Harvard Business Review, January–February 2014, pp. 96–101.

Supply Chain Segmentation at Dell

Simchi-Levi’s 2013 paper with colleagues at Dell developed and implemented the first supply chain segmentation strategy that was able to transform Dell business. The impact, as described in the paper, was dramatic.

See Simchi-Levi, D., A. Clayton and B. Raven (2013), When One Size Does Not Fit All. Sloan Management Review, Volume 54, No. 2, pp. 14–17.

Impacts on Practice: examples