In September Irv was interviewed for the “Global Visionaries” podcast by Carlos Ponce. Listen to the full episode here: https://www.advancio.com/advancio-talks-season-2-episode10-with-irv-lustig/. Following are lightly edited excerpts.
Carlos: Tell us about Princeton Consultants. What is your value proposition? What problems do you solve?
Irv: I joined Princeton to help grow the optimization part of the business. We look at problems where companies need to improve and automate decision making. Often, optimization improves decision making about the best allocation of limited resources. Problems we’ve solved include asset allocation in Finance, and patient and workforce scheduling in Healthcare. We helped the United States Census Bureau, which re-engineered the Census in 2020 so that the enumerators would get an algorithm on their phone that would tell them which house to go to and when. That helped take out about $2.5 billion of cost in the 2020 Census by making sure the enumerators were doing their jobs as efficiently as possible. We work on problems like that for our customers to help them transform their businesses so that they are making better decisions, saving money, increasing revenues, and boosting performance.
Carlos: Can you walk us through how optimization works behind the scenes? For example, the airlines, what problems are they solving in their back-office operations?
Irv: For decades, the airlines have applied optimization to the fleet assignment problem. Imagine a whole schedule of flights that you want to execute over the next month—and you need to allocate planes to those flights. What does that mean? If you are an operator like American Airlines or Delta or United, you have planes of different sizes, and they have to go enroute. Let’s say American has a plane going from Miami to Orlando, and then to Dallas, then Los Angeles. The plane has to be scheduled for maintenance. Airlines do not want to assign a big plane with lots of capacity to flights with low demand nor, conversely, a small plane to a flight with high demand. It is a critical optimization problem to efficiently schedule and allocate planes. The optimal solution assigns an actual plane to a flight, while accounting for the sequence of flights that it will go through to make sure that when the plane lands at one airport, there is enough turn-time to get the passengers off, clean the plane, board the next set of passengers, and move on to the next flight. This is a problem the airlines solve every month, maybe even more frequently, to make sure that they can run the airline as efficiently as possible.
A related problem is crew planning. The crew consists of pilots and flight attendants, who are assigned to flights. There are all sorts of rules, such as daily hours of service and qualifications to fly certain types of aircraft. Executives seek the right number of crews in each flight, not too many crews, and to avoid “deadheading,” which is when a pilot or flight attendant is a passenger out of necessity and not working on a particular flight.
In another area, airline customers experience optimization via revenue management, which was introduced by American Airlines back in the 1980’s. To get a basic understanding, consider a flight from Miami to Dallas. An airline team is allocating seats onto that flight according to a desired quantity of tickets to sell at different fare classes—but there are passengers whose final destination is Dallas, and others who are connecting to another flight. Expand this to a huge network. There are other passengers coming from other places into Dallas to make connections. This is a big optimization problem to determine how many seats to allocate at each fare level.
Now, passengers may sometimes observe many vacant seats on a flight and wonder why the price is high. The airline likely estimated that the demand would be higher or that passengers would be willing to pay more. This is all done with analytics behind the scenes.
There is a new revenue management application area to determine what the prices should be on those fare classes using other types of analytics and profiling. For example, the airline may offer you fares at certain classes if you are a frequent flyer.
Because airline revenue management was such a computationally hard problem to solve, applications used to be run overnight or just once a week. Today, these applications, including the allocation of seats to fare classes, are run much more frequently. For popular flights, such as Miami to Dallas on a certain day, an airline might adjust the allocation based on a demand signal. This is the kind of optimization that makes airlines run every day.
Carlos: How do you validate that an optimization solution is actually improving business outcomes versus just producing mathematical optimal results?
Irv: We try to establish a baseline, determining what can be measured. There are usually multiple KPIs that have to be traded off at times. Everyone wants to maximize revenue and minimize costs at the same time, but often a tradeoff is required. In the case of an airline’s disruption due to weather, for example, there are tradeoffs between customer service and operations. When an optimization solution gets implemented and it is making better decisions, we have the opportunity to compare KPIs not just quantitatively, but qualitatively. We are working with a client on a scheduling problem. There are current schedules that need to be modified. There is one metric that measures the quality of the schedule. There is another metric regarding the modifications. If there are too many moves to modify, there is a cost—not a financial cost but a “warm and fuzzy” cost, we could say. The client executives look at answers the solution is recommending and at certain metrics, and they can say “This is the one that feels right.” They do their own qualitative evaluation.
Typically in optimization, we have quantitative measures that allow you to see. Then we also build things that validate. Doing optimization, I mentioned what’s the best way to allocate resources under some constraints? We’re making sure that the decisions we’re making are satisfying a set of constraints. Technology‑wise, we call it a solution validator.
Consider there is a solution that comes out of the optimization engine. We have a non‑optimization team member, a validator, that takes the answer from the optimizer, and says, “Does it follow all the constraints?” We don’t allow the validator to look at the optimization code—the validator looks only at the business requirements that we wrote up from interviews we previously conducted with client executives. In this way, we make sure we are coming up with something that is valid with respect to the business requirements and that it is delivering the value along the appropriate KPIs.
Carlos: I would like to ask you for some final words of wisdom for the audience about optimization in general. Anything you want to say to the audience, that you want to leave them with?
Irv: Today we’re hearing a lot of hype about AI, especially Gen AI. Executives looking for transformational technology should be aware that Optimization has been around for nearly 80 years. Its value is proven every day by many companies. It can be hidden, but it delivers incredible Return on Investment (ROI). Another measure we optimizers use is Time To Payback (TTP). Business leaders considering a project and related investment have to assess if it is worthwhile having a system built that solves a certain problem. Additionally, they should evaluate how long it would take to improve their investment, because better decisions will be made. It’s not just about making better decisions that might improve profits or service—sometimes the benefit is making the decision faster than when it was done previously in a spreadsheet, or on a piece of paper, or with yarn on pins inside of a big gymnasium.
Optimization is all about making better decisions, but it’s often unknown despite its success in delivering value. I hope our listeners today realize they might have an optimization problem because they are figuring out a schedule by themselves on a whiteboard or using something in Excel to make employees do certain work. Companies like ours are going to help you recognize that these are optimization problems that we can help you solve.
