True Grit: telling our stORy

This was my first year seeing the Edelman award competition and my first time at the Analytics meeting. My last post was about how the Edelman competition finalists tell the stORy of operations research.

Disclaimer: I was one of the judges. Discussions about the finalists are confidential, so I have to leave out some details. I only speak for myself in this post.

All the projects were very impressive and challenging in their own way. Some stories are about using OR to solve an important problem and add value to an organization (360i, Army, Chilean Professional Soccer Association) , and others are all about implementation in a difficult setting. I want to showcase some of the stories that had to do with implementation, because those finalists showed true grit.

UPS for “UPS On Road Integrated Optimization and Navigation (Orion) Project”

This project was all about the implementation. One of the most challenging parts of the project was to create a map that could be used to automatically route the drivers. Their early map had the same problem you have when you use GPS to route you to destinations: the maps often do not find the right driveways and entrances to destinations. This snowballs and becomes a huge issue when there are more than 100 stops on a route. Optimizing with incorrect data is not very helpful. Improving the maps was an enormous effort that required years of collecting and refining data. This daunting task of replacing the expert knowledge possessed by UPS drivers almost led to UPS abandoning the project. Once a map was usable, UPS worked on the other aspects of the project: creating automatic routes for the drivers to increase the number of deliveries they could include in a route without asking drivers to drive longer. Ultimately, they need their drivers (55,000 of them!) to use interface with the analytics on an app to make it all happen. It was interesting to hear about how much UPS invests in training their drivers with the apps. Training is measured in weeks.

Eventually it all came together. The algorithmic contributions of this project seemed modest — UPS used a heuristic to solve a Traveling salesman problem with time windows. Getting everything to work and to work well was the real contribution. True grit.

BNY Mellon for “Transition State and End State Optimization Used in the BNY Mellon U.S. Tri-Party Repo Infrastructure Reform Program”

This project was fascinating. After the 2008 financial crisis, the U.S. Tri-Party Repo Infrastructure Reform Task Force sponsored by the Federal Reserve Bank of New York exposed a weakness in how banks do loans that could lead to huge bank failures, and they enacted government interventions to reduce the risk. As a third party custodial bank, BNY Mellon acts as an agent and intermediary between two parties in a loan, facilitating settlement between dealers and investors. A lot of money (in the trillions of dollars!) is left “out there” during the loan, and this amount was several times the size of the market cap of BNY Mellon. Another market crash could lead to BNY Mellon going out of business, which would have ripple effects to other banks. To stay in business, BNY Mellon had to almost entirely eliminate risk in one aspect of their operation (intraday credit risk) and fundamentally change business as they knew it. They made the change using optimization models that basically solving a big matching problem and then sequenced the trades to allow the matching to happen in time. The optimization model was solved using column generation algorithms. A second optimization sequenced the loan pairings to ensure that time-feasible matchings. At the end of the day, BNY Mellon had to get buy in with the people they do business with to implement the solutions. Some of the clients even ran the optimization(!) Column generation, you’ve come a long way, baby.

The New York City Police Department (NYPD) for “Domain Awareness System (DAS)”

I love public sector OR and have even been on a ridealong with police, so I naturally loved the idea of using OR to create safer communities. The NYPD developed an app (the Domain Awareness System (DAS)) that was a huge IT project with analytics lying on top of it. The DAS draws information from two main sources: (1) historical call records and surveillance cameras and (2) sensor data.

A big part of the project was creating an app that could quickly draw data from many databases and sources. That was an IT contribution. Prior to the NYPD’s adoption of the DAS, much of the Department’s information was only available to officers in the precinct house with permission to access standalone siloed software applications. Getting that information on to the app was an enormous effort that demonstrated true grit. It had a huge impact (officer safety and productivity), but the other aspects of the app were much more interesting to me.

The sensor data was fascinating. The app passively collects and processes sensor data to generate new calls that do not enter the system through 911. The sensors on the apps would “hear” and triangulate gunshots as well as test for radiation and other contaminants. Only 25-30% of gunshots are called into 911, so the app itself could help to discover important events for police to respond to. Officers are not dealing with a continuous stream of false alarms and true alarms when using the app. Officers can customize the sensor alerts and other types of alerts to do their job efficiently.

Finally, the app tracks data to be used for prescriptive policing. The data collected about what officers were doing and the types of calls received. They used historical data to forecast future demand at a fine level of granularity that could be used to better schedule the staff offline.

The “true grit” part of the app is the enormous level of buy-in from the officers in a very short amount of time. The app is deployed across every police precinct in the five boroughs, and will shortly be on all 36,000 officers’ smartphones and all 2,000 police vehicle tablets. I credit this to the design team who seemed to have the idea of optimizing police officer usage in mind from the beginning.