Article Event Banner

Tom Bacon provides some lessons from the fashion industry and delves deep into pricing practices to make some helpful recommendations

Revenue management systems forecast demand by O&D (origin-destination) pairing by flight, by day and by fare level 300 days out. There is, quite literally, a stunning amount of big data and complex analytics behind all this. What is more, given the challenge of any algorithm to accurately predict the future from historic performance, the system calculates uncertainty (forecast variance) around each forecast and optimises pricing based on both the forecast and the uncertainty of the forecast. Greater certainty (less uncertainty) justifies more inventory associated with that fare level; less certainty justifies less inventory for that demand level.

Lessons from the fashion industry

A professor at Harvard, Dr Kris Ferreira, in studying the fashion industry, another industry famous for oft-times unpredictable behavior by its customers, recommends another approach to dealing with inherent uncertainty. In addition to measuring forecast uncertainty and incorporating it in planning, an explicit process needs to be implemented for ‘learning’. Rather than wait for actual performance to come in and the model adjusted accordingly, she suggests fashion companies should implement numerous price changes early in the cycle, explicitly to learn and thus to better assess demand. Of course, often with fashion, price comes down over time – early in the cycle, prices are relatively high and prices come down only as inventory is depleted or the selling season comes to an end. Nevertheless, she observes that forecast misses often drive extremely heavy markdowns late in the process. She recommends retailers test market responses to price changes early in the cycle – and continuously thereafter. She argues that a series of small tests across the buying cycle can provide better insight into demand than a baseline forecast that doesn’t adjust until it’s too late, and where the only option is heavy discounting.

Perhaps, like in fashion, airlines need to implement a system of more frequent price changes

How does this apply to airline pricing? Perhaps, like in fashion, airlines need to implement a system of more frequent price changes – both fare levels and inventory allocations – before most of the demand has already occurred. We too forecast demand for the end of the cycle and price accordingly throughout the cycle; we limit low fare inventory if the total demand is projected to be strong or sell more seats at extremely low fares if we project low load factors. We adjust demand forecasts by fare level nightly but any such adjustment is specifically designed to ignore ‘noise’ and adjust slowly. And, of course, if the model recommends selling half the plane at the lowest fare, no change is likely until most of that inventory is sold out. As such, for many flights there is no real change in the demand forecast until 45 days before the flight – too late for a fundamental adjustment in inventory allocations. Similarly, each sell-up fare or ‘bucket’ needs to be offered and close to sold out before it, too, triggers a projected increase in total demand.

The fashion industry often prices too high early on and then deals with high inventory and desperation pricing to get rid of it. For airlines, the current process is biased toward fares that are too low early in the cycle. It allocates a certain number of seats to the lowest fare and is slow to allocate less. It will, however, allocate more if it takes more time to fill the seats – similar to the ‘desperation pricing’ inherent in fashion.

‘Learning’ opportunities for airlines

A series of tests at various times could provide additional insight into actual demand. More than 45 days before departure – from say, 120 to 45 days before the flight date - is a period when the lowest fares are the default and inventory analysts are not generally actively involved. This period has tremendous potential for ‘learning’. Here are some tips.

  • Don’t simply offer the lowest fares for this entire period. Instead break these down into multiple sub-periods with different pricing strategies.

  • Know that at various pre-defined test periods before the 45-day mark, for example, prices can be randomly $10 or $20 higher.

  • Understand that if average daily booking volume (velocity) stays the same at higher price points, the airline can implement the higher price point as the default and implement subsequent tests around the new higher fare.

Competitive pricing is, of course, always an issue for airlines. Airlines are reluctant to maintain a price premium versus competitors to risk losing share. However, many airlines today are dominant in certain markets and modest price premiums in these markets may have little effect on share of the market.

‘Learning’ needs to be an explicit objective of the revenue management process.

Closer in to flight departure – from 45 days to 14 days before the flight – is a tougher period for such testing. During this period, fares continually increase; each fare could only be available for less than a week at a time. Still, velocity can still be tested by closing off the inventory randomly for a day, or a few hours, at a time. Again, if velocity remains at target, the airline can more confidently move immediately to the higher fare.

Such tests – applied across hundreds of flights for short periods of time – can offer considerable insight into demand before it’s too late – and drive higher fares earlier in the cycle.

As with fashion, such tests operate outside of a sophisticated, history-based model. Such tests are specifically designed to determine whether a model is optimal – before all the demand comes in. Such testing acknowledges the uncertainty of demand and the need to build ‘learning’ into the process. ‘Learning’ needs to be an explicit objective of the revenue management process.

Tom Bacon has been in the airline business for 25 years and will be. He is a regular columnist for EyeforTravel and an industry consultant in revenue optimisation.  Questions?  Email Tom at or visit his website

Related Reads

comments powered by Disqus