There’s a common point of view in divining profit maximization for the travel industry that only perfection of inputs, models and outputs provides the right answer in revenue management. For those of us who are over-achievers, perfect always sounds like the right answer.
The fact remains: Modeling human behavior is an imperfect science and simply isn’t possible. So if 80% feels like failure, it’s time to rethink what that means when we look to optimize profitability. That particular spot at the top of the bell curve may be at the top of the range – and it may be good enough to make smart business decisions. In fact, 80% may be perfect.
By balancing costs and optimality according to the Pareto Principle – known as the 80/20 rule – we’ve identified how to achieve profit maximization. If I can’t convince you, let’s at least start a discussion on this topic.
Remember the scene from “Meet the Parents” where the airline check-in agent taps on her keyboard for what feels like ages, only to come back with a ridiculously high ticket price for Ben Stiller? I love to show this clip when I describe airline revenue management to the uninitiated. However, beyond the intricacy of revenue management, the video hints at the complexity and age of airline systems: Many run on “green screens,” and their answers often depend heavily on the skills of the user, clearly reminiscent of the “Meet the Parents” video.
So how does that affect optimality, you may ask? In a very big way it creates huge impediments for revenue management and similar systems that leverage this information to make recommendations and decisions.
Here’s an example: One of our solutions provides pricing recommendations for airline group sales, which requires fare information as an input to our pricing optimization algorithm. Simple enough, right? Well, not really. The mere fact of requesting fare information – which we all do when we book travel through airline websites, online travel sites or traditional travel agencies – turns out to be immensely complicated because those external pricing systems don’t provide the necessary tools to remotely obtain this information. These systems were designed to be self-sufficient, but not to communicate with other systems, and these restrictions make the availability of information very challenging.
Peter Belobaba, a renowned research scientist at the Massachusetts Institute of Technology (MIT) makes the argument that real life gets in the way of optimality for revenue management solutions that have been developed over the years. According to Belobaba, “provably optimal solutions” are often restrictive and work in labs – but not the real world, where more often than not their assumptions are violated in some form or other. He has an upcoming entry in the Journal of Revenue and Pricing Management – “Optimization models in RM systems: Optimality versus revenue gains” – that will make for great reading, and I encourage you to do so.
The end result: Heuristic solutions – like those we employ at PROS – end up outperforming theoretically optimal solutions. Belobaba summarizes his argument with a review of PODS simulation results, the MIT Passenger Origin Destination Simulator, which shows the performance of a variety of heuristics that make different assumptions and end up with very comparable revenue performance.
On the path to revenue management, airlines are often challenged by the inaccuracy of the data provided by external systems. There’s no means to verify the data, and it’s often stale due to user overrides. One airline we know actually crashed its revenue management system when it attempted to enter more than the allowable 64,000 user influence tables. It simply reached a point where it had to settle for heuristic solutions. Proof again that there’s just no possibility of 100% optimality.
So back to my original example on group pricing. What should airlines do in this case? In my opinion – and this is where I would love to hear your input – here are my recommendations:
1. Get the most accurate data where possible.
2. Evaluate the trade-offs between accurate, optimal inputs and approximations of these inputs.
3. Benchmark heuristic solutions in simulation environments.
This last point is paramount in my opinion. Understanding how heuristic solutions perform in simulations will help us better understand their shortcomings and gradually improve their performance. Furthermore, it will also allow us to guide users to the areas where the heuristics are most likely to need supervision to ensure overall company-wide revenue and profit maximization, if not optimality.
So to my opening point … perfect isn’t always perfect. If you’re optimizing at 80%, you’re still making smart decisions that deliver extraordinary results.