AGIFORS And The Definition Of OR

AGIFORS has kept a public record of its research for more than six decades. At its very foundation, the annual symposium has sought to promote Operations Research (OR) within airlines. In fact, a handful of early papers themselves attempt to find a common definition for OR within the airline industry.

In a 1961 paper entitled The Development And Organization Of OR In Airlines, J. Taylor set about to find this common definition. He began his research with a questionnaire handed out to fourteen airlines. The very first question he posed was this: “Where in an organization should OR have its niche so as to be most effective?” It turned out that it was more common for OR to live somewhere in the Finance & Economics department, however multiple carriers found it most appropriate to place it in Engineering and Maintenance. Other questions such as team size and how different airlines decide which types of projects the OR team will accept all resulted in a mixed bag of answers. After synthesizing the responses, Taylor mused “It used to be fashionable, centuries ago, when life was more leisurely than it is now, to debate the problem of how many angels could dance on the point of a needle. I don’t think an answer has yet been found, even using electronic computers. The modern equivalent to the age-old conundrum would seem to be ‘What is OR?’”

Although difficult to define, OR quickly began to produce insights that were ahead of their time. In 1961, B M Brough of British European Airways opened his paper entitled “Problems in the Design of Automatic Seat Reservation Systems” with the following:

It is now nearly ten years since the first automatic system for the control of airline seat space came into being. This elementary system was based on a simple, though robust, fixed program valve computer with limited magnetic drum backing storage and relied on low-speed telegraph lines for data transmission. The technical success of this venture is well known although it is not so generally appreciated that with this system the airline industry can lay claim to having pioneered the first practical ‘on line’ computer system and one of the first, if not the first, successful commercial data processing application of any type.

Even in 1961, Brough talks about the difficulty of evaluating potential revenue gains to be achieved by automation. He goes on to discuss data collection, record storage requirements, reliability, hardware costs, and evaluating proposals from multiple vendors. You would be forgiven for thinking this was a conversation from 2026. The technology may be different in 2026, but the challenges and opportunities are eerily the same. Some of the early papers from the 1960s almost read like Asimov in their futurism.

Many of the earliest papers focused on crew management, marketing, engineering, staff planning, inventory, and reservations. In 1970, the Beatles were releasing their last album in London, and revenue management papers began to more significantly appear at AGIFORS. In October 1972, Kenneth Littlewood of British Overseas Airways Corporation presented “Forecasting and Control of Passenger Bookings,” the paper that gave revenue management what’s still called Littlewood’s Rule: accept a lower fare only as long as its revenue matches or beats the expected revenue of the higher fare it might displace. Littlewood wrote “Several papers… have been presented at AGIFORS Symposia concerned with the determination of reservations control strategies which maximise the passengers carried by flight… However recent moves in fares policies now make it increasingly important for the airlines objective to be to maximise revenue instead.” Littlewood’s Rule was later expanded upon by Peter Belobaba (Expected Marginal Seat Revenue, EMSR), and upon this were developed the subsequent models and modern ML and AI-driven science that power RM systems. It was put in front of the AGIFORS audience first.

The topic of Revenue Management gained momentum in the 1970s, but in the 1990s it became a separate study group with its own annual meeting. Forecasting was a frequent subject, carried by a tight, recurring cast of characters: Richard Ratliff, Thomas Fiig, Peter Belobaba, Larry Weatherford, Kalyan Talluri, and Ben Vinod, publishing new angles on the same hard problems year after year. Pricing and willingness-to-pay grew from an occasional entry before 2010 into a dominant share of the program by the mid-2010s, tracking the industry’s shift into dynamic and continuous pricing.

Some of the people presenting at the 2026 Revenue Management Study Group meeting were writing the earliest papers on these topics decades ago. Kalyan Talluri is on this year’s program with Muge Tekin, estimating customer behavior from competitor sales data that arrives aggregated, noisy, and missing the no-purchase customer entirely. Talluri’s first AGIFORS Revenue Management Study Group paper on record is from 1999: “Economics of Yield Management.” Talluri has written papers that span twenty-seven years, each paper building on the knowledge gained from prior research.

Thomas Fiig’s AGIFORS record dates back to 2001, authoring or co-authoring thirty papers, including one this year. In 2017, at the Annual Symposium, with Rodrigo Acuna-Agost, Fiig submitted a paper entitled “The end of Airline Revenue Management as we know it? (Deep) Reinforcement Learning for Revenue Management.” This year, Fiig is back with Michael Defoin Platel, asking a question along the same vein: “Artificial Intelligence—What Will Be the Impact on RMS?” The research paper archive is evidence that the anxiety which today is being attributed to AI, predates AI.

Other sessions on this year’s agenda carry a similar theme, without appearing to have coordinated it. Alex Matson’s presentation is titled “The Human in the Loop: Thoughts on the Analyst/RMS System.”, discussing why system overrides are so common despite advanced science in the RMS. John Elder, John Bruer, James Graham, and Randi Griffin ask how you prove after the fact that a human override actually produced a better outcome than doing nothing. Alex Winston appears to accept this challenge, proposing a machine learning method for separating what a revenue management strategy did from what the market would have done anyway.

This all sounds like the latest version of a topic that has never been settled: what does the system get to decide, what does a person get to override, and how does anyone prove afterward which of the two was right?

In a few weeks in Seattle, these ideas will all become part of the same conversation. In subsequent articles, we will see firsthand why these in-person symposiums are so beneficial to the discourse surrounding AGIFORS research. We may not have an industrywide definition of Operations Research, but maybe this is not necessary as long as we continue to see the real life impact.