There are common pitfalls in how big data analytics can be applied to travel.
In many industries often the use of big data analytics can be misapplied and the same applies to travel. Why? Because as big data expands and models become more and more sophisticated, there is a danger that companies oversimplify – or, alternatively, over-complicate – customer behavior.
One example of over-simplification is ‘last click attribution’. Attributing sales success to ‘last click’ ignores the fact that travel booking is a series of steps and that the driving factor may occur before the ‘last click’. Many analytics teams are analysing bookings based on a more complex decision-making process than just ‘last click’. Teams may be able to statistically identify the key clicks (sites that predict ultimate booking) or key patterns (combinations of sites) are often better predictors than ‘last click’.
Many relationships in the real world are non-linear so a linear model, that fits the historic period, can dramatically miss the forecast
Similarly, simple regressions often don’t fit the real world. Many relationships in the real world are non-linear so a linear model, that fits the historic period, can dramatically miss the forecast when the explanatory variable moves out of a narrow range. Also, certain variables may appear correlated but are not causal — in many cases, factors are found to be correlated over a historical period but are not actually predictive. Finally, as in ‘last click attribution’ there may be a series of steps, not one or two major ones, that are determinative of customer behavior.
‘Tortured’ data is often an outcome of such over-simplification. Tortured data is data that is forced to align with analyst expectations. ‘Numbers can say anything you want’. Basic regressions lack validity when too many possible explanatory variables are investigated simultaneously and when the analyst, only through dozens of iterations, ‘finds’ the right model. One airline analyst confided to me that his department had worked on a model for months striving to prove a simple hypothesis that, in the end, wasn’t there.