TORONTO—Keeping things simple might seem like a mantra that is easier said than done when it comes to the world of data science and analytics, but Nancy Pyron, senior director of operations research and revenue management at Marriott International, said that’s exactly what her company tries to do.
Speaking at HSMAI’s recent Revenue Optimization Conference, she said there are a lot of things to consider when trying to turn a sea of raw data into useful information for hoteliers, but it makes sense to try the simple solution before the complex ones.
“Start with the most simple and work forward,” she said.
Pyron noted there are a many incredibly interesting and complex approaches to data science today, including things such as machine learning and neural networks, and they can give businesses sophisticated information, but they require significantly more data and upkeep to operate properly. And the information put out by simpler models can often be easier to interpret and act upon.
She noted there is a time and place for both, but it’s important to not just go to the most complicated options off the bat due to their promise and potential.
She said more complex models like neural networks will spit out answers based on the data provided, but it will be impossible for anyone to track how it came to its conclusions, which can be difficult for the people on the business side to digest.
“If you can’t get (the path to the answer), then there’s a level of acceptance and trust asked of the user base,” Pyron said. “You have to have a huge leap of faith. … There’s this weird thing where users want to understand where these answers came from.”
At the same time, Pyron said the hotel industry needs to embrace a higher degree of sophistication when it comes to data analytics because many simply use the term loosely and inaccurately to make their businesses sound more advanced than they truly are.
“There are a lot (of businesses) who say they’re doing it, but they’re really just doing basic math,” she said.
The types of analytics
Pyron said there are three main buckets for data analytics. The first is reactive, which takes a look at historical performance. The second is predictive, which ventures to predict outcomes and set expectations going forward. Proactive is similarly forward looking, but is more qualitative than quantitative.
Once businesses decide what kinds of goals they have, she said analytical models can do lots of things, from forecasting future room rates to trying to parse out how many of their forward-looking reservations are leisure- versus business-based on limited guest profiles.
She noted there are myriad applications for data science, but people need to know what they’re looking for to be productive.
“The complexity of a model can span from giving you a three-month moving average to building a full neural network,” she said. “But the level of complexity interacts with the accuracy (of the data), the time (it takes to build and operate) and the cost.”
Things to consider when dealing with data
Pyron said there can be many costs when dealing with data, some of which might not be considered right off the bat, but it’s important to weigh those costs versus the financial benefit of the analysis.
She said there are costs associated with sorting through the data, and with the manpower and systems needed to build and operate the models to crunch those numbers. But the piece that is often overlooked, she said, is the costs of acquiring the data itself.
Pyron said the more complex a model, the more data is required to fuel it. That means possibly having to purchase outside data to get the results required.
“If it’s going to cost $100,000 a year just to get that data stream, you have to ask, ‘Is this going to generate $100,000 in incremental life for (the) hotels that need it?’” she said. “It could be a model, but you might not be able to afford the data needed to feed that model.”
Pyron also said communication is key when building models, so teams of IT workers don’t get stuck creating something that in the end nobody wanted.
“You have to make sure teams are working in concert with the business to give the best models and to get the best results,” she said.
She said teams need regular input from a diverse group of executives to operate at the best level. She also suggested creating separate roles for the people who come up with the models and those who go through and program them.
“The person building that brilliant model shouldn’t be the person coding it into production,” Pyron said.
She said one of the hardest but most important things to do when it comes to data analytics is to find the perfect team, which requires highly skilled individuals who are invested in the work.
“Your team needs to be dedicated to building the best model for utilization—the best model to get the best answer,” she said. “They can’t be focused on themselves or how cool they can make it or writing a paper out of it. They need to build a model to meet (the company’s) needs and build it in the most simple way possible.”