February 2016, Atlanta
When revenue management leads to reasonable doubt test, test, test
As 2015 draws to a close, Tom Bacon finds that history lessons can also be applied to revenue management
Reasonable doubt is a legal concept, not a business concept. US juries are instructed to review the evidence and find a defendant ‘not guilty’ if there is reasonable doubt that he committed the crime. Thus, the jury can conclude that he ‘probably’ or ‘likely’ committed the crime but, if there is reasonable doubt, they still must find the defendant ‘not guilty’.
As the saying goes: “It is better for ten guilty defendants to go free than to sentence one innocent citizen”. In the legal system, we assess upside (convict all criminals) versus downside (jail the wrong guy) and opt to be conservative.
This concept applies to risk taking in business. In general, we are taught to make decisions based on ‘most likely’. However, in many cases, when the upside of an initiative is much different from the downside, we need to exercise ‘reasonable doubt’ with regard to the ‘most likely’ scenario. Comparing upside versus downside in a business situation often justifies pursuing a strategy even if in the ‘most likely’ or ‘probable’ scenarios, it won’t work!
RM systems already measure upside versus downside, and estimate an entire probability distribution so in theory they would already quantitatively and scientifically assess ‘reasonable doubt’, and not act solely based on ‘most likely’.
· ‘Most likely’ demand for a $1,000 passenger may be less than one. However, RM systems may still save a seat for such a high fare passenger based on a less-than-50% chance that he will show up. The upside of ensuring a seat is available may be too compelling.
Nevertheless, RM systems are often slow to respond to market changes and are not designed to capture major turning points.
· In many cases, ‘most likely’ is based on conventional views or linear extrapolations that may not hold given the complex, frequently non-linear, dynamics of the market.
So, ‘reasonable doubt’ should form the basis for constant testing. Tests, by definition, limit the downside and if they identify an otherwise overlooked demand opportunity can lead to broader application, and broader success.
Take these two examples:
What if…there is a larger mix of high fare passengers than is represented in the historic database? Testing for more high fare passengers is actually a feature of some RM systems. Rather than relying on historical observed high fare demand, these systems ‘test’ additional higher fare demand by restricting low fare inventory and forcing ‘sell-up’. These systems specifically override ‘most likely’ based on history, to capitalise on more high fare demand if it exists. Of course, as ‘tests’ these typically require significant manual intervention, both with helping to validate the reasonableness of the ‘tests’ and turning them off quickly if they do not work.
What if …there is a dramatic turning point in demand, for example a new higher level of demand across fares that is not captured in history? Airlines typically test such turning points by adjusting the model – for a time period or for a subset of markets – by a fixed percentage. Again, this requires extreme care – in monitoring the results and removing the assumed increase if it doesn’t drive improved revenue results.
In essence then, airlines need to apply ‘reasonable doubt’ to demand forecasts based on history. A practical way to do this is with continual testing, tests designed to determine if actual demand follows a path somewhat different from ‘most likely’, based on history. So, tests that override RM model recommendations are a necessary part of optimal revenue management. ‘Reasonable doubt’ has a role in airline RM, not just the courtroom.
Tom Bacon is a 25-year airline veteran and industry consultant in revenue optimisation.