| October 20, 2006 - HotelExecutive.com Hotel Business Review
By Timothy C. D'Auria, Director of Data Analytics, Carnus Systems & Vicky Lee Bradshaw, Founder, Carnus Systems
F&B forecasting is well known throughout the industry to be notoriously difficult, often leading to inaccurate staffing. This challenge is the result of the large variety of variables that effect F&B operations, the complex interactions between these variables, and the frequently large fluctuations in daily cover volumes. Banquet information is often in flux due to the requests by guests, and restaurant walk-ins are a common occurrence. Together, these factors make the forecasting complexity overwhelming and nearly impossible for the human mind to grasp, even with the added assistance of a capture ratio or spreadsheet program. Indeed, calculating the contributions made by each of the scores of variables is a daunting task best left to advanced techniques with greater precision. Almost as long as the U.S. hotel industry has been in existence, owners and operators have relied on managerial intuition and experience to forecast F&B demand and labor. Occasionally, managers may use capture ratio calculations or other methods available in a spreadsheet program such as Microsoft Excel in their planning. However, the complexity of variables and seemingly unpredictable fluctuation in demand makes the application of many mathematical forecasting techniques inappropriate. Until now, forecasting F&B outlets have been accepted as a weekly "crap-shoot." With the advent of novel artificial intelligence solutions in other industries that yield strong forecasting accuracy in these highly variable environments, such technologies will likely disseminate to the hotel industry and eventually become the standard within F&B departments. Similar to revenue managers� reliance on management systems and complex analytical tools to forecast and schedule, F&B directors and hospitality ownership will eventually demand the same level of technical analysis in an effort to provide improved service quality in a competitive environment. This contention is supported by studies showing that both intuition and capture ratios are inaccurate methodologies to forecasting demand and are each outperformed by artificial intelligence methods. Artificial intelligence (AI) is the process of simulating human thought processes using computers. Since the early 1950?s, AI has been a topic of great interest in the field of computer science. However, only recently has a transition been made from theory to practical application involving everyday problems, a delay likely caused by initial limits in computational power and our own understanding of the human brain. Today, the integration of AI into everyday business remains in its infancy despite demonstrated promise and success. New elevator systems in high-rise buildings, including the one found at the New York Times corporate headquarters, use artificial intelligence to learn patterns in traffic flow and become increasingly efficient in transporting people to appropriate floors in the least amount of time, leading to a happier work force. In medicine, AI has been successfully implemented to detect the presence of prostate cancer, with results that exceed those yielded by current techniques. In many instances, AI has been shown to exceed the reasoning ability of humans. In 1997, IBM-designed AI computer "Deep Blue," defeated world champion chess player Garry Kasparov. Additionally, a NASA-designed AI computer classified distant space objects as stars or galaxies with accuracy that exceeds that of humans. In the hospitality sector, the Hyatt Regency Riverfront in Jacksonville, FL, one of the largest hotels on the U.S. East coast with 966 rooms, is an early adopter of AI technology to forecast and schedule staff. An AI forecasting tool has allowed the hotel to both forecast patronage and generate on-the-fly schedules with superhuman accuracy while reducing weekly scheduling time from 3-5 hours to mere seconds. Similar to the Hyatt Regency, the Pan Pacific in San Francisco was on track to save an estimated 4% of labor cost during the first year of AI implementation with a daily overall average accuracy of 90%. Similar accuracy was observed with room service and banquet forecasting, allowing the hotel to appropriately staff and reduce costs. Artificial Intelligence has been shown to yield the greatest hospitality forecasting accuracy compared to alternative techniques in current industry-wide use, including capture ratios, human guesswork, and multiple linear regression, each of which fail to detect the complex non-linear patterns within hotels, thus leading to poor accuracy. In the case of capture ratios, its underlying assumption is that F&B demand is based exclusively on in-house occupancy, and that the relationship between in-house occupancy and F&B demand is fixed; However, if in-house occupancy was the exclusive factor that drove F&B demand, this would imply that in-house events, surrounding attractions, weather, competition from nearby restaurants, and other variables play no role in determining F&B demand. Similarly, the relationship between occupancy and demand will generally not be fixed since other variables interact with occupancy to generate demand. The need for better tools has led some F&B departments to implement software that uses linear regression to make forecasts. However, the weakness of using this method in an F&B setting is that it is a simplified linear technique being used in a complex non-linear setting. Furthermore, most regression models assume that the relationship between F&B demand and other variables do not change. For example, weather accounts for 4% of demand, arrivals accounts for 11%, etc. However, this static relationship is an inaccurate oversimplification of a much more complex problem. As an example, rainy Wednesdays in the month of September affect patronage differently than sunny Wednesdays in September, which is also different from both rainy and sunny Wednesdays in March. The effect a particular variable has on demand changes depending on the value of another variable. Time series methods, which some hotels have expressed interest in, demonstrate the poorest accuracy due to their univariate nature that uses only previous values of the variable of interest to forecast, thereby ignoring all other information like convention information, weather, room availability, etc. While guessing games and crap-shoots may have long been the status quo,
this standard is soon about to change. The resulting inaccuracy caused
by manual forecasting, capture ratios, and other poor techniques should
no longer be considered "part of doing business." The use of these unsophisticated
and inaccurate forecasting methods is troubling since most full-service
hotels rely heavily on the service quality and profitability of their multi-million
dollar F&B operation. AI�s ability to learn and determine the complex
interactions between a large numbers of variables make it extremely versatile
and effective in complex environments such as hotels, allowing it to reduce
costs and improve service quality.
About the authors Timothy D�Auria is Director of Data Analytics for Carnus Systems. Mr. D�Auria is responsible for all aspects of quantitative analysis and data solution development at Carnus Systems. As a business entrepreneur and published author in the area of statistical application, Mr. D�Auria�s work formed the cornerstone of the analytical methodology used at Carnus Systems. Originally a business owner in New York City, Mr. D�Auria moved on to assist with the development and teaching of the statistics course in the Cornell School of Hotel Administration. Mr. D�Auria is a graduate from Cornell University with a double-major in Statistics and Biology and a Bachelors of Science degree with Research Distinction. Carnus Systems was founded by Ms. Vicky Lee Bradshaw. Ms. Bradshaw has extensive research and practical experience in hotel operations, and has led forecasting and labor management projects with major U.S. hotel chains, including the Ritz Carlton Hotel Company. Ms. Bradshaw�s domain expertise in hospitality operations and labor management led to her vision of designing automated artificial intelligence forecasting and scheduling software to minimize labor costs and to enhance service quality for the hotel industry. Ms. Bradshaw has a Masters degree from Cornell University s School of Hotel Administration. <http://www.hotelexecutive.com/bus_rev/pub/002/624.asp#top> Copyright © 1997 - 2005 HotelExecutive.com |
