Introduction: 2020 Franz Edelman Award for Achievement in Advanced Analytics, Operations Research, and Management Science

It is an honor for us to serve as chair and special issue editor, respectively, of the 49th annual international competition for the Franz Edelman Award for Achievement in Advanced Analytics, Operations Research, and Management Science. The 2020 competition had been scheduled to be held on April 27 at the INFORMS Analytics Conference in Denver, Colorado. The COVID-19 pandemic disrupted this plan and caused turmoil for organizations across the world. Businesses that had leveraged analytics for great success earlier leaned heavily on analytics teams to respond to the pandemic. This led to a delay in the Edelman Award competition, which ultimately took place virtually on September 29. (One finalist team had to delay its participation until the 2021 competition.) To avoid further pressure on the finalists, the INFORMS Board decided to let finalists present the financial impact of their work through 2019 (rather than updating it to reflect the effect of the pandemic).

Five finalist teams described how they applied operations research (OR) and advanced analytics to solve difficult decision problems. The problem domains include designing products, pricing cruise line packages, reducing computer server downtime, scheduling trains, and setting prices to clear retail shelf inventory. These entries highlight not only the high-impact models the proponents have created but also the remarkable and diverse ways these proponents have made advanced analytics modeling and analysis a part of strategic thinking and operational practice at their organizations.

About the Edelman Award Competition

The Franz Edelman Award competition is jointly sponsored by INFORMS and the INFORMS Section on Practice. The purpose of the competition is to bring forward and recognize outstanding examples of advanced analytics, OR, and management science (MS) practice. The award is named in honor of Franz Edelman, who established one of the earliest industrial analytics and OR/MS groups in North America at RCA. He worked at RCA for more than 30 years and is counted among the fathers of innovation in analytics and OR/MS.

The first-place and finalist awards are for implemented work that has had significant, verifiable, and measurable impact. The impact may be beneficial to the organization winning the award (e.g., by increasing its revenues) or to others (e.g., by reducing pollution). INFORMS presents trophies commemorating the finalist awards to the client organizations that used the finalists’ work and presents medals to the finalist authors. This year, the prize money totaled $14,000, with $10,000 going to the first-place winner. More important, the finalists have the honor and satisfaction of knowing their work has been recognized by their peers as the best in the profession. In addition to having their efforts described in this special issue of the INFORMS Journal on Applied Analytics, all finalists have their presentations available at https://www.informs.org/Resource-Center/Video-Library/Edelman-Competition-Videos.

The Process

The Edelman Award process begins with a call for entries in the summer prior to the scheduled competition date (the summer of 2019 for this competition). The selection committee reviews all entrants and selects a set of semifinalists. Verifiers then work behind the scenes to validate the claims made by each semifinalist and to convey this information to the rest of the selection committee. Verification is performed by a mix of practice-oriented academics and full-time practitioners. The verifiers communicate directly with the entrant’s team, the users of the work, and client management. Support from client executives is important. Verification is a crucial element of the competition because it ensures that only the highest-achieving OR/MS and advanced analytics work makes it to the Edelman Award finals. All verifiers are provided with written guidelines and sample verification reports, and novice verifiers are paired with experienced verifiers.

The selection committee studies and discusses the verification reports to select six finalists. Coaches are assigned to each finalist team; these coaches help the finalists improve their papers and presentations for the competition. Typically, multiple iterations of paper and presentation drafts are required to clearly convey the work to a general INFORMS readership and audience within a limited number of pages and presentation lengths.

Judges study the papers, listen to the presentations, and then discuss the finalists’ accomplishments until they reach a decision on which finalist is most deserving of the Franz Edelman Award for Achievement in Advanced Analytics, Operations Research, and Management Science. Relevant factors include the overall impact and value of the application, the level of technical innovation, the difficulty of the obstacles surmounted, and the work's portability to other application contexts.

The Virtual Competition and Awards Ceremony

This year, the Edelman finalists were honored at a virtual awards ceremony held on September 30, the day following the competition. Authors of the Edelman finalist papers are designated as Franz Edelman Laureates and were mailed medals in advance of the ceremony so that they could wear them during the awards ceremony. The culmination of the ceremony was the announcement of the 2020 first-place team from Intel.

The Finalists and the Papers in This Issue

Here is a summary of the finalists listed in the sequence that their papers appear in this special issue.

Intel for Intel Realizes $25 Billion by Applying Advanced Analytics from Product Architecture Design Through Supply Chain Planning

Intel decomposes its product architecture design and supply chain planning into five integrated analytics modules. Product composition is determined using device physics simulations and mixed-integer programming (MIP) to minimize cost while meeting forecasted demand. Genetic algorithms and linear programs generate an efficient frontier of product design options that trade off product performance for manufacturing cost. The three supply chain modules help determine wafer start volumes (the most important supply chain decision), allocation of capacity across the network, and, finally, optimized routing plans for assembly and test operations. Intel uses several techniques to solve the supply chain optimization subproblems efficiently: MIP, cuts, Dantzig–Wolfe decomposition, parallel processing, constraint caching, and path-based constraint formulation. The resulting analytics have increased Intel’s revenues and reduced its costs for a total benefit of over $25 billion since 2009.

Carnival Corporation & plc for Carnival Optimizes Revenue and Inventory Across Heterogeneous Cruise Line Brands

One of the challenges of applying revenue management principles to cruise lines stems from the possibility of using cabin room inventory for multiple cruises (e.g., 14-day and 7-day lengths) simultaneously on a given ship. Carnival uses quadratic programming optimization to jointly determine cruise prices and allocate cabin inventory to cruises. Machine learning algorithms predict the demand that drives the optimization. This led to an incremental uplift in net ticket revenue that varied between 1.5% and 2.5% depending on the cruise brand.

Deutsche Bahn for Deutsche Bahn Schedules Train Rotations Using Hypergraph Optimization

Deutsche Bahn (DB) uses hypergraph optimization to create rotational (cyclic) train plans for its rolling stock of locomotives, wagons, and trains for its passenger and freight railway operations. The rolling stock is combined into trains and scheduled employing a three-level hierarchical decomposition using column generation. To ease planner workload and to improve computation time, reoptimization is used when possible. DB attributes this work with providing annual savings of €74 million, improved regularity of operations, and significantly reduced CO₂ emissions.

IBM for IBMPredictive Analytics Reduces Server Downtime

IBM uses machine learning to classify problems encountered on many of the computer servers that IBM and its clients manage. High-impact problems involving server outages are correlated with problematic server configurations. IBM uses multivariate analysis and simulation to recommend hardware and software upgrades. These recommendations enable IBM’s clients to use their upgrade budgets more effectively, thus saving the clients an estimated $7 billion since 2013.

Walmart for A Multiobjective Optimization for Clearance in Walmart Brick-and-Mortar Stores

Walmart uses optimization to determine (discounted) markdown prices to clear its U.S. stores’ excess inventory by a specified date (often, when product types are being replaced by others). It uses gradient boosting regression and partial differential equations to estimate the relationship between price and demand. Multiple price markdowns for a given product are determined using a deep learning algorithm (double deep Q network algorithm). As a result of the work, Walmart increased the amount of products sold during the markdown period by 21% and reduced its markdown costs (such as relabeling) by 7%.

Conclusion

The Edelman finalists’ papers make this issue of the journal special for both practitioners and academics. Practitioners can benefit in at least four ways. First, they will find better ways of accomplishing their work using advanced analytics models in a diverse group of organizations in both the private and public sectors. Second, they will find better ways to advocate their ideas to others in their organizations by pointing out the impact of adopting analytics modeling. Third, they will learn how to bring about change in an organization to make OR-based modeling and analysis an integral part of the culture. Finally, they can be inspired to tackle challenging problems and make the modeling choices necessary for their effective solution and deployment.

Academics will find validation of the advanced methodology they teach and will be able to demonstrate what can be achieved with OR/MS and advanced analytics. Furthermore, these examples illustrate how model customization is often required to suit the problem context.