"Simu-casting": Around and Beyond Forecasting
Sergio Mendoza, Ph.D., Co-founder & CEO Airnguru;
Javier Jiménez, CCO & Product Owner Airnguru;
Mauricio Acuña, Sr Research Executive Airnguru
An unprecedented context for the airline industry
During the COVID-19 pandemic, the airline industry has suffered from a dramatically steep fall in passenger demand, but, worst of all, from enormous short and mid-term uncertainties which have crippled its commercial capabilities.
Forecasting, the backbone of key commercial decisions
Arguably one of the most data-driven industries, over the last few decades, airlines have relied on forecasting science, technology, and processes as the backbone of many commercial decisions. From network planning to Yield Management, forecasting has been a critical ingredient in airline decision-making.
In the long and mid-term time window of airline business decisions, fleet allocation decisions and budgeting rely, among other things, on long and mid-term industry demand forecasts.
On the other extreme, that is, the short-term time window, Revenue Management decisions heavily rely on granular, unconstrained airline demand forecasts to determine the optimum number of seats to allocate (inventory) to each origin-destination-fare product in each future flight.
In between these two extremes, itinerary adjustment decisions rely on constrained demand forecasts, that is, the airline demand resulting from inventory optimization under the limited capacity of the itinerary for sale.
All this worked reasonably smoothly and well for decades until the pandemic's onset, with one extraordinary exception: September 11 of 2001. In terms of demand destruction, the COVID-19 pandemic has been the worst nightmare that has hit the commercial aviation industry in its whole history.
However, a few months after the onset of the pandemic, we had learned that the main problem was not the demand destruction itself, but the enormous structural uncertainty introduced to the business: uncertainty of not knowing how the COVID cases would evolve over the next few months, how long would it take for a territory to reach the herd immunity, when would people regain the confidence of traveling again, when and under what conditions would authorities in different countries reopen the borders to international visitors, how will customer preferences and purchase behaviors (basically, what defines the demand segments) change during and after the pandemic.
These and many other unknowns are causal for the travel demand. They define the underlying context of the airline business and the conditions for future air travel demand. And we don’t know how to forecast them because they are highly volatile. Some of them depend on politics, and others depend on demographic, epidemiological, and economic factors, and, above all, this is a first-time-ever event. This highly volatile context produces a highly uncertain demand for future air travel.
The collapse of the airlines' commercial machinery
The most perturbing effect of high contextual volatility is that we have no clue what will happen with the demand for air travel, not even in the short term. Even if we had faith that travel demand would recover (which is a reasonable assumption), we don't know when this may happen and how that demand will behave in terms of anticipation, length of stay, and other relevant customer preferences.
Current airline demand forecasting science, technology, and processes are mainly based on the observed behavior of the demand for past flights and the incoming demand for future flights under stable context conditions. A stable, pre-pandemic context allowed airlines to bet on a single future scenario and made forecasting a valuable tool. The current volatile contextual conditions left demand forecasting models useless, leading to a collapse of the airlines' Revenue Management, fleet allocation, itinerary optimization, and budgeting processes, basically crippling the airlines' commercial planning and decision-making machinery.
We have to be bluntly honest here: given that we have no clue how the context is going to evolve, we have no clue how to forecast demand for air travel. Thus, a new approach to airlines' commercial decisions is needed, and new machinery is required to support such decisions. After all, worse than making bad decisions is not making any decisions at all!
Though we have no clue how the context will evolve, we may have some hints on possible directions such evolution may take. We could identify some key drivers that define possible future scenarios, for example, the date in which the US will reach herd immunity, the pace at which leisure demand will recover in a given OD market, etc. Second, using reasonable ranges for each of these key drivers, we could lay out a comprehensive series of possible future scenarios representing the level of uncertainty we currently have. Third, we may simulate a battery of alternative commercial strategies. Each commercial strategy could include one or more key decision areas like fleet allocation, itinerary, pricing, yield management, frequent flyer program. We would use the simulations to forecast each strategy's impacts in each scenario on expected net revenue and other financial, commercial, and strategic KPIs, and perform analysis of sensitivity to the various parameters of the corresponding strategies. Fourth, we could rank the strategies based on how winners or how resilient they are across the scenarios, or based on their upside versus risk ratios, and, accordingly, decide which strategy to execute. And finally, we could measure the real results of the executed strategy, recalibrate our simulation parameters and get ready to start a new "simu-casting" cycle.
As time evolves and the pandemic approaches an end, we expect contextual volatility to diminish gradually, so airlines will have a better idea of which future scenarios to discard and which ones to ponder with a higher probability. So, at the end, when the markets reach stability and airlines can bet on a single forecasted scenario again, using our proposed approach to select the right strategy would converge to a problem of strategy optimization for the future scenario. If we could simu-cast at the right speed and cost, we could even use this method to optimize tactical decisions. Moreover, a simulator built and calibrated with the right level of granularity could help airlines understand how the underlying demand is evolving in terms of preferences and behaviors, something quite valuable for the purposes of product and price optimization.
All this may have sounded like science fiction a few years ago. With the help of massively parallel computing and Machine Learning on the Cloud, this may not be science fiction anymore.
How to continue
If you are interested in a deeper discussion of these concepts, want to see Airnguru's developments in this front, or have some interesting ideas that you'd like to share, please do not hesitate to contact us at:
Follow us on Linkedin: https://www.linkedin.com/company/airnguru