Saturday, September 10, 2011

In Memoriam: 10 years after O.R's Apollo-13 mission

Two years ago, this tab recognized the amazing but well-hidden contribution of OR practitioners to the post 9-11 recovery plans at US airlines. Does your company have the right people to handle such an emergency? It's almost ten years to the day since OR's finest hour, and after journeying through a couple of different industries since then, my personal conclusion is that it was not a coincidence that those US airlines that best overcame this nightmare also had on their payroll seasoned OR experts.
So it's worth revisiting this topic in a Q&A form to examine the impact that OR can have during such situations. It is not intended to be an 'OR hagiography' - the stock of 'traditional' OR has in fact shrunk drastically since 2001, while some different new avenues have opened up.

Q. Was OR really critical to the post 9-11 schedule recovery effort?
Absolutely. The key ingredient to finding a solution to this 'mother of all one-time decision problems' was the presence of a significant number of experienced personnel in the airline with strong OR skills. why? Large airlines that operate thousands of flights on a daily basis use expensive aircraft that burn precious jet fuel, and are crewed by tens of thousands of trained professionals both on and off the ground. All this requires a well-synchronized and manageable 'program of activities', i.e., schedule of tasks that airplanes and crews follow. Every tiny portion of schedule involves complex combinatorial and constrained decision making to make the best use of these scarce and expensive resources. Nobody in an airline understands and applies the fundamental science behind this (yes, there is one) better than OR folks.

Q. Other than the tragic occasion itself, what was so unique and challenging about this problem in particular?
The scope: mammoth - hundreds of planes and thousands of crews stranded across North America in all kinds of airports.

The stakes: huge - the quicker an airline could get back on its feet to resume normal operations, the less money it bled (the slide into eventual bankruptcy began around the same time period).

The technical challenge: very high, but this was just a part of the problem. Research to design a brand new decision feasibility and optimization model that would be actually operational, and manage a gazillion possibilities that satisfy several new constraints, work well within a brand new implementation that gets around tons of IT and engineering issues within a couple of days - then to be used for a few days before being discarded for ever. This was the perfect R&D storm.

The task: a two-stage problem that would first bring our planes and crews back to their domiciles and then get them back to a regular operational mode through the end of the month.

Q. I'm intrigued. What the heck is "OR" and what kind of OR skills were needed to weather this perfect storm?
I forgot. O.R stands for Operations Research. OR folks tend to be reclusive but do make some noise about duality once in a while. As a rule, we generally believe in hiding our best achievements as well as the real name for our field. Dissident, mediocre OR types believe that we ever so often give away our flagship award ('Edelman') to bungling senior non-OR execs (who then hire OR grads to fix the problems they created), and we all know how well they have run this planet the past decade. Sour grapes aside, OR folks deal with the science and engineering of better. For example, we try our best not to tell you how to run your business, but we can and will tell you how you could run it more efficiently and effectively. Visit informs.org for details.

To the second part of your question: Airline-OR research experience combined with solid implementation skills was the need of the hour. During the high-flying days, airline R&D intellectual property in certain business areas was 2-3 years ahead of the best output from any university. Consequently, the OR crisis team was able to quickly take apart existing business-critical decision systems (that OR guys built and managed in those days) and extract big chunks off the million line monolith of mess-C code for reuse within this one-time model.

Q. I'm now an OR expert, so i'm jumping gears. Computationally, what optimization technique was crucial?
Column generation (and emacs) turned out to be the most important weapon in the arsenal 10 years ago. We were already routinely solving mammoth monthly planning instances and steadily pushing the envelope through intense research into improved algorithms as well as million$ cutting edge parallel hardware that airlines could afford then, so multiplying the problem size by another trillion did not really matter in the end. Now there is flexibility for you.

Q. I never realized that OR, in the right hands, and for the right reasons, would be so powerful and practical. Should every company maintain an OR team?
Any medium to large sized company, and not just airlines, would be well advised to maintain and hold on to an experienced and skilled OR team having some implementation experience. OR is ultimately about practical problem solving rather than an consultancy exercise that leaves critical (and insightful) details as an exercise to the reader. Routine planning problems may well be handled using OR-commodity software packages, but the day that even a mild Apollo-13 like decision problem threatens to disrupt your company's fundamental operations and plans, nobody can help put Humpty-Dumpty together again (and in an optimal and efficient manner) like an OR team can; seen this happen just ten years ago.

Thanks, OR guy. I heard the analytics was the real flavor of this decade and ...
Sorry to cut you short, we are analytics too, for a long time now. We change our names like winamp skins.

Winamp?
yup.

9/13/11: fixed typos

No comments:

Post a Comment