Revenue Management in the Age of Big Data:
Is More Data Equal to Better Data?
By Paul van Meerendonk, Director of Advisory Services, IDeaS Revenue Solutions
For some time, the biggest buzz in business has been around the influx of Big Data and its application to hospitality - and specifically to revenue management systems. Historically, revenue management systems (RMS) were already the biggest data owners within the hospitality enterprise, with two or more years of detailed reservations data consumed by the system, across a variety of room types, customer segments, length of stays and more. With this data, RMS analytics generated billions of forecasts used for further optimization, subsequently producing billions of pricing, availability and overbooking decisions. That is to say, Big Data existed in revenue management systems before it was even known as "Big Data."
But today's Big Data story tends to focus on the increased variety of data sources available for revenue management, including data from social media, reputation management engines, web traffic sources, weather, and information related to your competitors. This data is increasingly easy to access or purchase, and arguments can be persuasive that all of these new data sources should be incorporated into the revenue management process. "More is better," according to some in the industry. And that is true in a universe where every new data element is statistically relevant - where new data improves forecast accuracy and increases property profitability by improving RMS pricing decisions. "More data? It can't hurt!" some loudly proclaim. But, in fact - from a statistical perspective - sometimes it can.
Relevance Is Key in Big Data: Consider the Case of Rate Shopping Data
More data is better only when the RMS analytics improve price-demand estimates, provide controls for your particular business mix and pricing strategy, and enhance the optimization process. A good example of this is the use of rate shopping data for competitive pricing. Revenue managers have long known that incorporating all of their competitors' prices rather than their primary competitors' in their market place is not always the wisest pricing strategy. An analytical approach is necessary to determine which competitive properties are actually relevant to a customer's willingness to pay and to the type of demand, in contrast to using all competitor rate information equally. In the absence of statistical relevance, inclusion of more rate shopping data can significantly impact your competitive and brand positions, and diminish your pricing strategy.
Use of Customer-Centric Data in Hospitality
Recent innovations in RMS technology have also shown that reputation-related Big Data is growing in importance within hospitality. This growth stems from the various research (including studies by Kelly McGuire and Breffni Noone, and by Chris Anderson) which indicated that online reputation and price are two of the most important considerations for guests to make their booking decisions. Access to reputation-related data has become more available to hotels from reputation vendors, and today there are many RMS providers that display a property's reputation and rate in relation to their competitive set for decision support. In the case of online reputation data, the key is incorporating it into demand modeling and optimization processes, rather than merely reporting it or utilizing it as post-decision support. The insurmountable amount of structured and unstructured (such as sentiments) reputation data makes it a very complex and unsustainable process for revenue managers to use it as an ongoing post-decision support mechanism. Thus in the case of incorporating customer-centric data in pricing decisions, revenue managers must consider demand as a function of price, where the demand is also a function of the specific customer-centric data type to be added to the mix.
Incorporating Regrets and Denials Data in the Mix: Simply Unreliable
There has been significant hype about using customer traffic data for unconstraining the demand forecast. Particularly the attempts to use regrets and denials in unconstraining have been largely unsuccessful where the success of unconstraining affects the entire pricing and revenue management process. Regrets and denials data makes a good case for challenging the quality of data when it comes to incorporating more Big Data. First, there is an important distinction between "denials" that are due to unavailability and "regrets" that are due to price or other factors. Second, many reservation or booking systems are unable to automatically capture the difference between regrets and denials, and add to that the fact that denials are captured manually after the fact. Using brand.com data, which TravelClick reports to make up only 27 percent of the reservations for transient nights, also calls the data quality to question because it uses only the unqualified transient demand without sufficient regard for the demand for wholesale, group, corporate negotiated, and unqualified business. In addition, today it is known that guests' look-to-book ratio is extremely high and continues to increase. Studies show that many guests use multiple websites for making booking decisions. Even if one is confident about the methodology for assessing denials on a site, it is not at all reliable to forecast without the cross-usage of additional websites or data from multiple visits per buyer. That is primarily why leading data scientists refer to regrets and denials as "dirty data" and consider regrets and denials data an example of where more data will likely hurt demand forecasting rather than making it more accurate.
Maximizing Profits vs. Maximizing Revenue Using Big Data
Another evolution-in-the-making in revenue management technology is the optimization of profitability rather than revenue. Profitability optimization can be tackled by obtaining ancillary revenue and cost data to generate profit contributions by various customer segments. Ancillary revenues range the spectrum of food and beverage revenue streams as well as golf, spa, events and more. In the casino business, player theoretical loss - i.e. the amount of money a player is expected to lose based on a casino's statistical advantage - can also be incorporated into pricing decisions.
Cost or margin data is required across each customer segment when the RMS maximizes total profitability, since certain customer segments, while contributing extra revenue, can also incur additional variable costs. This is especially true when considering group and function space requests, as different revenue sources associated with these requests often have significantly different margins. This data can be enlightening throughout the hospitality organization - marketing and operations departments come to mind - but depending on the contribution of ancillary revenues with respect to rooms, integrating ancillary revenue and cost data can cause significant changes in RMS optimization.
Incorporating Weather Data in Hotel Demand Forecasting: Which Hotel?
As an input to hospitality demand models, weather data may improve the short term demand fit if, and only if, its immediate impact can be assigned to a particular market or property. Imagine a snow storm hitting Boston Logan Airport and closing down all highways and airports. Is it good or bad for the hotels in the area? The answer: it depends. If all departing guests have nowhere to go and extend their stays while the number of expected arrivals are low - it might be good for some of the hotels as the expected occupancy has increased. However, if all departing guests have left and arrivals have stalled it would be really bad for most area hotels. Not only the impact of such weather events depend on the specific circumstances but also it will vary greatly for an airport property versus a property far away from the airport. All of this is to say weather data trends may be impactful to travel patterns at large but their relationship to business or leisure bookings at a particular location is loosely coupled.
Statistical Relevance Is Key to Big Data Inclusion in RMS
In many cases, much of the "Big Data" begging to be incorporated in RMS is demand-related data; that is, data that is assumed to improve forecast accuracy. Some examples of Big Data having an impact on forecast accuracy are by improving price-elasticity estimations, recommending better competitive pricing decisions, changing the objective (profitability vs. revenue) used by optimization algorithms, and adding the user-centric information that guests actually use in selecting hotels.
As it can be concluded from the examples cited above, RMS technology must incorporate Big Data into analytics not just because the data is available, but because the addition of more data is statistically significant in the RMS process. RMS providers have to be extremely careful when continuing to add more and more data into the RMS forecasting algorithms. In many cases, using more data may improve the fit of the initial training data, but will not generalize well to future dates or scenarios. While one may be reducing the variance in the historical data very well, the inclusion of the additional data ends up "over-fitting" to historical data thus resulting in bias and unreliable forecasts.
To summarize, it is very important to treat each new data source carefully:
- Does it contribute to new information that has not been provided in the data currently used?
- Does it change the nature of the decisions that you are making by offering a new way of thinking about the problem?
- Does it meet any performance standards you have set such as reducing forecast variance or having a reactive pricing strategy?
- Is the new data set being used as input to the demand forecasting or as post-decision support?
In this day of easy access to data and progressively lower costs to store and process this data, it is easy to believe and act upon the idea that RMS' should incorporate the newest available data immediately. But, like any investment decision, it is a decision that should be considered carefully, ensuring that the hotels' pricing strategy and decisions are improved because of these additions. In conclusion, some RMS providers have yet to prove how new data types drive better revenue.
As Director of Advisory Services for IDeaS Revenue Solutions, Paul van Meerendonk leads a global team of revenue management advisors focused on hotel revenue optimization projects. Mr. van Meerendonk is responsible for global development, management and operations of the Advisory Services team. He oversees the hiring, training and management of industry-leading consultants located in London, Beijing, Singapore and Atlanta. Mr. van Meerendonk also represents IDeaS on industry thought-leadership initiatives related to trends and best practices within revenue management, including authoring a number of white papers, conducting public speaking engagements, as well as leading key client webinars with an average audience of over 200 global representatives. Mr. van Meerendonk can be contacted at +44 (0) 118-82-8100 or Extended Bio...
HotelExecutive.com retains the copyright to the articles published in the Hotel Business Review. Articles cannot be republished without prior written consent by HotelExecutive.com.