Revenue management has always been a “big data” problem. The level of detail needed to make pricing recommendations means even a small chain’s revenue management system runs hundreds of thousands of forecasts every night. These forecasts then are used as inputs for a massive optimization problem. Originally the problem was too hard to solve in time to update prices, so traditional revenue management systems had to make simplifying assumptions. These simplifications ranged from data aggregation for forecasting, to breaking up optimization problems, to calculating key metrics like overbooking levels or price sensitivity after the optimization is complete, or to restricting the ability to optimize “on the fly”. Revenue management systems no longer need to make these sacrifices.
Data storage has become relatively cheap and processing power has increased exponentially. SAS has taken advantage of these technology trends to help companies capture, store and use all of their “big data” without sacrificing speed or accuracy. The performance gains are tremendous. Problems that used to take days to solve now take minutes. We’ve incorporated these advancements in SAS Revenue Management and Price Optimization Analytics, so that we can provide more detailed analysis, faster – at the speed of today’s dynamic market.
Speeding up Forecasting
The most obvious use of this enhanced technology is a major overhaul of revenue management forecasting. Every revenue manager knows that forecasting is the essential foundation of profitable pricing recommendations – yet forecasting is often a major weakness in traditional revenue management approaches. Dealing with all of the market changes we’ve been discussing requires more detailed forecasting, incorporating all relevant factors such as room type, market segment, length of stay, price sensitivity, special events and so on. However, simply incorporating more, and more detailed, data is not enough. In fact, detailed data can add a new set of challenges. Specifically it introduces the issue of sparse data - not enough observations to provide a meaningful or significant forecast. Current revenue management systems have avoided issues like data sparcity by aggregating, but this came at the cost of accuracy. For the SAS approach, we use pooling and automated pattern recognition in innovative new ways, so that detailed data is handled properly, providing accurate and statistically sound forecasts at the right level of detail for truly optimal pricing recommendations.
As more granular data is included, today’s revenue management systems’ approach of “one-size-fits-all” forecasting becomes even more problematic. The most accurate forecasting method for each forecast differs with factors like time before arrival, the presence of seasonality, amount of available data, and the volatility of the data. Restricting forecasting to only one method sacrifices forecast accuracy, especially with granular data. SAS’s approach is to incorporate dozens of best of breed forecasting methods so that the system can automatically select the best forecasting method for each individual forecast. Clearly this results in tremendous improvement in forecast accuracy and forecast performance. With hundreds of thousands of forecasts to run, the system must be able select the right model, optimize the associated parameters efficiently and deliver results quickly. SAS Revenue Management and Price Optimization Analytics is applying leading edge “big data” technology and high-performance analytics to accomplish this.
Ultimately, accurate forecasts at a more granular level of detail enable more accurate pricing recommendations. However, this level of granularity also puts more pressure on optimization algorithms. We’ve also taken advantage of increases in processing power and high performance enabled technology to expand the scope and size of revenue management optimization. This allowed us to overcome two of the biggest weaknesses areas of today’s revenue management systems: price sensitivity of demand and competitive pricing. Over the next couple of weeks Alex will talk in more detail about the analytic advances that we made to incorporate these new factors into the RM decision.