Should Marriott International acquire Starwood Hotels & Resorts Worldwide as is widely expected, it will gain a company whose smartphone check-in, key-less entry and robot bellhops are progressive in the increasingly technology-focused hospitality sector. But Starwood’s digital crown jewel is invisible to guests: Its analytics software platform enables the hotelier to automatically recalibrate pricing of its properties using hundreds of variables that influence supply and demand.
Starwood spent more than $50 million over three years on this revenue optimization system, or ROS as the company calls it, which has helped the chain improve demand forecasting by 20 percent since 2015, says David Flueck, Starwood’s vice president of global revenue. As ROS “learns” it will price rooms more efficiently, ideally boosting revenues and profitability. “What we’re trying to do within our industry space is to say: How do we exploit access to data to create a more robust capability?” says Starwood CIO Martha Poulter.
Dynamic pricing has long served as the chief function of revenue management systems operated by airline and rental car industries, as well as the hospitality industry. For example, Marriott, Hilton and Intercontinental Hotels Group adjusted rates one or two times a day as far back as 2003. Thanks to advances in high-performance computing and analytics software, pricing now changes from minute to minute. If you’ve shopped online and noticed changes such as higher or lower rates (or suddenly no availability) after you refreshed your browser, you’ve seen dynamic pricing at work. Marriott’s system, called Marsha, displays rates based on how long someone will be staying, in which market they’ll be staying and when they’re staying.
Dynamic pricing includes ‘man and machine’ approach
ROS works similarly. As you click through your options on an online reservation system, ROS is performing a dizzying number of calculations behind the scenes in Starwood’s data center. It is cross-checking past and present reservation data, booking pattern data, cancellation data, occupancy data, room type, daily rate data, as well as transient or group status (whether you’re a solo traveler blowing into Manhattan for a couple of days or a gang of 100 coming in for a convention). ROS also analyzes external data, such as competitive pricing data, weather and climate data, and booking patterns on other sites.
And in what Flueck calls a “man and machine” approach to revenue management, ROS is also programmable by each Starwood property’s director of revenue management, or DORM, via a Web dashboard. The display shows bar graphs representing total rooms available, transient vs. group rates, projected net occupancy and more granular variables such as the “hurdle rate” or the lowest rate for a given date based on demand. If the DORM booking the St. Regis in Manhattan knows that Justin Bieber is performing at Madison Square Garden, he or she can raise the rate, anticipating that demand will exceed supply, or adjust rates in the event they expect more cancellations. A DORM may also program ROS to alert her if the rival across the street raises price by, say, more than 20 percent. “The dashboard allows DORMs to have a snapshot of the business,” Flueck says. ROS is operational in 1,000-plus Starwood properties, 550 days into the future.
The all-you-can-eat-right-now approach to analyzing data is a departure from how Starwood conducted pricing prior to 2014. Like most hospitality chains, Starwood relied on its revenue managers to manually compile spreadsheets and recommend prices to set for each room, and turn the data over to corporate to run sophisticated analytics. Today, ROS captures all of that data in one central system and analyzes it in real time. “All of those technological advances created the right recipe for us to build the system,” Poulter says. “It is a constant analytic engine refinement.”
ROS makes Starwood more attractive to Marriott
ROS will serve Starwood well as it prepares to merge with Marriott in a $13.3 billion deal shareholders approved last month. The combined entity will boast 1.1 million rooms across 30 brands, all of which will be under pressure to maximize sales. “[ROS] was certainly an ingredient in the mix that made Starwood attractive,” says Atmosphere Research Group analyst Henry Harteveldt. “All of the larger brands are exploring ways to improve their revenue optimization.”
Systems like ROS are also necessary to keep pace with data-driven companies that disintermediate the travel and hospitality sectors. Online travel agents such as Priceline and Expedia, and even Google, have ruffled many feathers for its frequently evolving and various iterations of both flight and hotel booking services. These companies help guests find rooms at Starwood, as well as other well-heeled chains, albeit with little loyalty. Starwood see ROS as a way to keep pace with these disruptive companies, which “sit between us and our guests,” Flueck says.
Starwood has a distinct advantage. Flueck says that while those competitors see the broad market, they don’t see such data as groups, corporate contracts or wholesale bookings Starwood has acquired over the years. “Google, Priceline and Expedia all have analytics at the core of their strategy,” Flueck adds. “We need to make sure that we are staying at the leading edge of that as well.”
ROS will also help Starwood counter a challenging trend. Hotel booking windows are shrinking, perhaps as a result of the get-it-quicker expectations cultivated by Amazon.com, Uber and other on-demand companies that are accelerating the pace of fulfilling orders, say Poulter and Flueck. A consumer who previously booked a vacation several months in advance might be booking their June vacation in April. Events managers booking room blocks for corporate conventions are also waiting much longer.
What does ROS’ architecture look like? Neither Poulter nor Flueck will say much beyond that it is comprised of a multi-tier architecture that uses R, the statistical modeling language, IBM’s C-Plex optimization software, as well as proprietary machine learning algorithms. Flueck says that software components for variables such as forecasts and booking windows are self-contained within modules. When Starwood wants to refine an algorithm it can pull out the module and plug in a new one. “This will become more accurate as we refine the system,” Flueck says.