IAAA Finalists

Track: INFORMS Prizes – IAAA Finalists

The IAAA Finalists will present at the conference on Tuesday, April 16 in the INFORMS Prizes Track from 9:10am – 12:20pm EST. Judges will then review the rank and announce the winner at the INFORMS Analytics Society luncheon.

Amplifying Marketing Impact: Optimizing Network TV Advertising Inventory with Advanced Analytics

Sebastian Souyris, Lally School of Management, Rensselaer Polytechnic Institute
Sridhar Seshadri, Gies College of Business, University of Illinois Urbana-Champaign
Dana Popescu, Darden Business School, University of Virginia
Vinod Kumawat, RSG Media
Anand Punjabi, RSG Media
Shiv Sehgal, RSG Media
Divyani Sharma, RSG Media
Varun Verma, RSG Media

Promotional advertisements are a fundamental marketing instrument for television networks, enabling them to foster audience engagement for upcoming television programs. The efficient planning of such promotional material is crucial for a show’s success, generating revenue opportunities for networks via commercial advertising. Traditionally, networks have employed manual processes for devising multiple promotional campaigns; however, these methods are labor-intensive, time-consuming, and inefficient.
RSG Audience Marketing Prophet application optimizes and automates the promotional planning process for networks, ensuring each campaign attains its desired outcomes while adhering to the constraints imposed by a limited marketing inventory and network conditions. To address this challenge, we have developed an innovative approach that synergizes mathematical programming and machine learning techniques to produce comprehensive long-term promotional campaign strategies and daily promotional schedules. These plans and schedules are of high quality, as evidenced by conventional business metrics and a minimal integer programming gap. Using our methodology, 24 TV cable and one broadcasting network (Paramount Global and A&E Networks) have observed a 5% to 17% increment in marketing inventory for commercial advertising.

Battling Nursing Shortage Crisis: Building Innovative Staff Deployment Solutions with Human Expertise and Machine Learning

Pengyi Shi, Associate Professor of Operations Management, Mitchell E. Daniels, Jr. School of Business, Purdue University
Jonathan E. Helm, Professor, Operations & Decision Technologies, Kelley School of Business, Indiana University; Industry Partner and Contact: Indiana University Health
Mary Drewes, Vice President, Associate Chief Nurse Executive, Operations
Jacob Cecil, Data Analyst, Workforce Central
William Tindall, Director System Resource Center and Staffing Optimization
Troy Tinsley, Regional Director of Supply Chain at Riley Children’s Health. (Former) Director of Workforce Strategy and Operational Excellence

We partnered with the largest health system in Indiana, IU Health (IUH), to co-develop and implement an advanced analytics-driven decision support solution (DSS) to support their novel practice of mobilizing resource nurses across its 16 hospitals. The aim is to reduce staffing shortage, improve patient care, reduce harm events, and increase retention for nurses through improved working conditions. Our DSS combines state-of-the-art generative AI models with multi-stage reinforcement learning, creating a framework that handles both forecasting and decision optimization in environments with complex spatial-temporal correlations. The DSS provides multi-tiered recommendations, including resource nurse on-call lists, real-time redeployment between hospitals, and future hiring decisions. We conducted a pilot from May to June 2023, which produced a remarkable 17% reduction in understaffing, with estimated annual savings of $2.5-3.5 million in IUH alone, with over $1.5 billion projected savings if adopted on a national scale.

Interdependencies and Cascading Effects of Disasters on Critical Infrastructures

Eva K Lee, PhD, Director, Whitaker-NSF Center for Operations Research in Medicine and HealthCare;
Chief Scientific Officer, the Data and Analytics Innovation Institute; Senior Data Scientist, AccuHealth
Technologies; Subject Matter Expert, DHS Medical & Public Health Information Sharing Environment.
William Wang, PhD, Georgia Institute of Technology; Research Scientist, Amazon.
Taylor Leonard, PhD, Lieutenant Colonel, United States Air Force; Chief, Strike Branch, Department of
the Air Force, Office of Studies and Analysis.
Jerry Booker, Former Director, Enterprise Risk &. Performance, TSA–Strategy, Policy Coordination, and
Innovation.

Critical Infrastructure (CI) is vital for a modern economy, and ensuring its security is crucial. The potential incapacitation or destruction of CI could have severe consequences for national security, the economy, public health, and safety. This project specifically examines the primary effects of cyber-attacks on critical infrastructure, investigating hidden vulnerabilities, cyber-physical disruptions, and their cascading impacts. Our goal is to develop analytical methods that provide valuable insights for policymakers, aiding in informed decision-making. Our accomplishments include (a) establishing the first influence system network model to reveal vulnerabilities with maximum cascading effects in communication critical infrastructure, utilizing scalable computational advances to model the entire U.S. communication network; (b) creating a multi-layer influence network incorporating interdependencies across various CI sectors; and (c) proving theoretical results about the monotonicity and submodularity of the influence function, with implications for computational complexity. The analytic tools and computational framework enhance the capabilities of cybersecurity and infrastructure security leaders, providing improved assessment for decision-making. This contributes to better system analysis, content relevancy, and long-term objectives of assisting decision-makers in assessing disruptions, establishing policies, and designing effective mitigation strategies. Identifying critical nodes enhances both defense and offense strategies, optimizing resource allocation for maximum protection.

Optimizing Milk Pools at the Rogers Hixon Ontario Human Milk Bank

Timothy C. Y. Chan, Department of Mechanical and Industrial Engineering, University of Toronto, Toronto, ON, Canada
Rafid Mahmood, Telfer School of Management, University of Ottawa, Ottawa, ON, Canada
Deborah L O’Connor, Department of Nutritional Sciences, University of Toronto, Toronto, ON, Canada
Debbie Stone, Rogers Hixon Ontario Human Milk Bank, Toronto, ON, Canada
Sharon Unger, Department of Nutritional Sciences, University of Toronto, Toronto, ON, Canada
Rachel K. Wong, Department of Mechanical and Industrial Engineering, University of Toronto, Toronto, ON, Canada
Ian Yihang Zhu NUS Business School, National University of Singapore, Singapore

Human donor milk provides critical nutrition for millions of infants that are born preterm each year. Donor milk is collected, processed, and distributed by milk banks that pool donations together to improve the overall macronutrient content. Approximately half of all milk banks in North America do not have the resources to measure the macronutrient content in donations, which means pooling is done heuristically. We collaborated with the Rogers Hixon Ontario Human Milk Bank (RHOHMB) to create a data-driven milk pooling optimization framework via predictive and prescriptive modeling. We implemented this framework over a year-long trial and observed that pools created by our approach met clinical macronutrient targets approximately 31% more often than the previous nurse-led approach, while taking 60% less recipe creation time. This is the first work in the broader blending literature that combines machine learning and optimization. Our results demonstrate that such pipelines are feasible to implement in a healthcare setting and can yield significant improvements over current practices.

Policy Learning Utilizing Counterfactuals with an Application to the Dynamic Pricing of Airline Ancillaries

Shivaram Subramanian, IBM Thomas J. Watson Research Ctr.
Wei Sun, IBM Thomas J. Watson Research Ctr.
Markus Ettl, IBM Thomas J. Watson Research Ctr.
Youssef Drissi, IBM Thomas J. Watson Research Ctr.
Zhengliang Xue, IBM Thomas J. Watson Research Ctr.

Recent advances in AI have encouraged organizations to implement data-driven prescriptive analytics to automate and improve their decision-making capabilities. Such prescriptive policies must satisfy a variety of operational and fairness constraints that are ubiquitous in practice. A simple and interpretable policy is preferable as they can be easily verified and integrated into existing systems. Prior literature focused on constructing variants of prescriptive decision trees that are unable to satisfy such constraints. Our application employs a novel causal-teacher prescriptive-student framework to optimize a constrained prescriptive policy generation problem. We employ a counterfactual estimator that allows the application to be agnostic to the choice of the prediction methods, and deployable across diverse domains. We solve a novel path-based mixed-integer program (MIP) using column generation to derive an optimal policy that represents a multiway-split decision tree. We demonstrate the efficacy of our application by partnering with a global airline carrier to successfully live-test pricing policies for ancillary products.

Using Innovative Analytics on Two Levels to Increase Vaccination Rates

Theodore T. Allen, Ph.D. FactSpread Columbus, Ohio; The Ohio State University, Integrated Systems Engineering, Columbus, Ohio.
Antor Rashid, formerly FactSpread Columbus, Ohio; The Ohio State University, Integrated Systems Engineering, Columbus, Ohio.
Long Wang, Ph.D. FactSpread Columbus, Ohio.

Our vaccination-related advertisements reached approximately 1 million people (3.9 million impressions) across over eight states. Our analytics-based approach was innovative partly because we prepared analytical information and presented it to millions of people. We also used analytical approaches to design our messages, target the campaigns, and measure the results. Optimization was used to cluster similar counties. Then, optimal experimental design and regression analyses were applied to tune the advertisement selections and policy parameters. A second, more major campaign likely benefited from the improved parameters. Estimates of people caused to become vaccinated include 0.7M over one year (p-value 0.001). Another estimate including information contagion effects is 2.2M (p-value 0.002). Therefore, estimates of the lives saved include 0.7M ÷ 124 = 6,000 and 2.2M ÷ 124 = 18,000. These outcomes occurred with a total direct cost across Meta, Google, Broadcast2World, AYTM, and Pathfinder Consulting of approximately $58,000. Therefore, the cost per United States person saved from death is likely under $10.

Track: INFORMS Prizes – IAAA Finalists

Get the Latest Updates