Revenue management is a human activity and as such opens itself up to preconceived notions, emotion and natural biases. On this subject, guest columnist Tom Bacon has a lot of questions

It should go without saying that you shouldn’t let your statistical model run without human intervention. After all, sophisticated forecast and optimisation models available today are fundamentally based on statistical analysis of history – and the travel market is constantly changing in ways that a model cannot anticipate. But even so, there are questions that need answering.

·         Does your operation regularly suffer from intervention that hurts your overall revenue performance?   

·         If so, what is the proper level of intervention and how do you maintain the model without incorporating human bias in the process? 

·         When should you override the demand forecast? 

·         Should you intervene when it automatically closes off the lowest fares and stops booking velocity - doesn’t that just reflect a ‘too optimistic’ demand forecast? 

·         Or does intervention, in this, case reflect a bias that the model is able to avoid?

There have been many books written about human decision-making and bias – and bias appears in virtually every human context, including RM. So, it is important to have procedures that serve to reduce or eliminate bias that does not improve revenue results.

In fact, some studies have concluded that analyst intervention actually worsens performance rather than improving forecast accuracy. So, unsurprisingly, many RM system vendors caution against excessive analyst intervention.

There are a number of common flaws in our thinking that make us less accurate than the computer. Here are four examples:

1. Human emotion: Emotion is part of every human process. For example, RM managers typically fear their flights will go out with too many empty seats. It’s far more obvious that something went wrong when there are empty seats than when the plane is full (even the plane is full of too many low fare passengers). 

2. Confirmation bias: We often give too much credence to information that confirms what we hypothesized and we tend to ignore contrary information. If we ‘think’ the booked load factor is too low, we find lots of data to support our intuition without trying to find reasons why the booked load factor might actually be too high.

3. Fighting the last battle: Similarly, we naturally respond to ‘the last battle’. In this case, we are biased to action! If the average fare turned out much lower than expectations last month, we tend to want to act to increase average fares this month in an effort to ‘hit the target’. Unfortunately, we’re dealing with dynamic markets with complex causality so all too often fighting the last battle is counter-productive. Overcompensating fits into this category too as we often overcorrect for a phenomenon, so as not to suffer from the same over-or under-forecast two periods in a row.

4. Too narrow framing of problems: Open or closed? Raise fares or not? Strive for more upsell or not? Often we simplify decisions to binary options when there could be multiple causes and, potentially, a continuum of solutions. 

Top tips for improved Intervention and how to combat RM ‘bias’

1) Insist on data based intervention: Given that emotion plays a role in all human activity, any intervention must be based on solid facts. So ask the questions: exactly how much has the load factor underperformed and for how long? How pervasive is the shortfall? Why wouldn’t the RM model ‘catch up’ with the actual? Any intervention should match – in direction and magnitude – the new information.

2) Widen our options: If bookings are coming in strong, it could be that demand is higher than expected (a logical conclusion) …or that the booking curve has moved outward …or that more bookings are speculative or …many other possibilities. Since we may not know which of many options are most valid, our response should tend to be more muted – until we have more information.  Small tests designed around multiple alternatives are often a better approach than too quickly pursuing one option.

3) Look for the opposite: Similarly, to avoid confirmation bias, we need to constantly look for opposing data. We should simultaneously seek data that, for example, supports both that demand is stronger than expected and that demand may not be.  ‘Devil’s advocacy’ should be a fundamental part of all analysis. 

 4) Seek alternative perspectives: Rather than quickly reaching a conclusion on a new trend, actively seek others’ interpretations. Use your RM team and the broader experience across markets that your colleagues represent to reduce individual analyst bias.  Have daily team meetings to discuss actions and results; this provides a forum for more inclusive thinking – as well as accountability. Regular meetings with other functional areas (marketing, sales, finance) are important too as they can also be a forum for gaining new perspectives.

5) Set ‘stop loss’ limits: In financial markets, traders who cannot predict a result utilise ‘stop loss’ limits to ensure they don’t lose too much when their system isn’t working optimally. RM managers, too, should establish alerts so that a wrong assumption doesn’t become too costly. Don’t let your ‘convictions’ lead to a closed mind! Go ahead and intervene based on the best data you have at that point in time but, once you’ve done so, be sure to maintain an openness to counterveiling information.

6) At the most basic level, measure ‘bias’: In addition to measuring overall forecast accuracy, monitor the impact of analyst intervention – and whether there is consistent over or under-forecasting. Often, analysts over adjust forecasts based on information they receive outside the RM system; by measuring bias, analysts can learn how to more appropriately process adjustments to the automated results.

Human intervention is a critical part of RM. After all, our RM models cannot respond quickly enough to the fast-changing marketplace. However, we need to manage that intervention and recognise that too could, unfortunately, be counter-productive.  Implementing processes to reduce bias in RM is a way to balance the need for intervention, with the danger of human bias.

This article was produced by Tom Bacon, 25-year airline veteran and industry consultant in revenue optimisation. Questions?  Contact Tom at tom.bacon@yahoo.com or visit his website http://makeairlineprofitssoar.wordpress.com/