How hoteliers use data science to manage rates
 
How hoteliers use data science to manage rates
03 JUNE 2019 7:16 AM

Experts explain how data science helps hoteliers manage their property rates and how the discipline has evolved over time.

GLOBAL REPORT—Until recently, revenue managers had to determine hotel rates on their own using as much information as they could handle, balancing historical performance against current market conditions and guest demand.

As technology advanced, hoteliers started turning to data science to take all of these factors under consideration and create forecasts and pricing recommendations.

In terms of rate management, hoteliers are using data science to create algorithms in forecasting to determine future business and set rates accordingly, said Tom Blomquist, director of revenue management at Scandic Hotels. The information includes historical data, such as what people are paying when they’re buying different products. Data science sees how that can be used to determine new pricing products or improving pricing in the future.

”In short, that’s how we look at the data, both forward and backward,” he said.

Scandic implemented its first rate-optimization program in approximately 40 hotels back in 2006, he said. In 2010, the company decided to add optimization systems to all of its properties. It also has a data warehouse that began in 2005 and was revamped in 2009 and 2010. These systems and the warehouse give the company a more holistic view of guest behaviors, he said.

The revenue-management team works with the optimization system, Blomquist said. They use the data from the warehouse to do a deep dive into the analyses and use that for setting long-term strategic goals. The overall company has structures and systems in place, but each country that is home to Scandic properties has its own organization and cluster of revenue managers to set their own price points, he said.

Often when people hear hoteliers talk about data science, they think companies are using it to increase hotel rates, but Blomquist said that isn’t the full truth. The goal is to try to fill the hotels, which means understanding the customer segments to make sure the hotels have the right pricing at each point in time.

Accor launched its approach to data science about six to seven years ago to better understand and improve its rate management, said Agnès Roquefort, SVP of transformation strategy and data at Accor. Her team has two main competencies: one focused on business intelligence, KPI monitoring and performance improvement; and the other with data scientists, engineers and analysts using AI and advanced analytics for guest knowledge and acquisition.

The company enriches its data to become more accurate, using more data in its algorithms to get the right rate for the right date at the right hotel for the right type of guest, she said. It uses its algorithms to look forward and recommend the best maximum price for the hotel. Sometimes a hotel will have several prices per day.

The forecasts take guests’ willingness to pay into consideration when determining prices, she said.

As the forecasts and recommendations become more accurate, revenue managers setting the prices become more confident in and more willing to accept the recommendations, Roquefort said. This will then free up team members to focus their time and energy on more complex business decisions.

The practical use of data science involves machine learning and forecasting, said Dak Liyanearachchi, chief data and analytics officer at Hilton. It combines pricing, products and services, he said.

“Price is one element, but what does the customer get in terms of the value and make it a value for them?” he asked.

The forecasts take into account a number of elements, such as location, reputation, local events, weather and economic conditions, which enable better occupancy predictions.

“Based on that, you can start to understand pricing algorithms and how do they affect the price, both in terms of not just the core price but what are the products to give the best value,” he said.

The revenue-management team at Scandic Hotels uses data from its data warehouses to determine long-term rate strategies and creates standardized, simplified reports for hotel GMs to help them determine the best rates at their properties. (Photo: Scandic Hotels)

Human influence
As Hilton evolves its use of data science, it will move faster with greater accuracy and allow a greater focus on the guest experience, Liyanearachchi said. As the company builds out that production system, executives will see how that feeds into the revenue-management teams and those at the property level will push that out on a daily basis.

The data and analysis can get the company to a certain point, but it should still be easy for a person looking it over to make a correction if necessary, he said.

“We’re early days,” he said. “We are still somewhat a ways away from the capabilities to get a thing like a NeuroNet to think like a human being. … We still need a mixture of man and machine.”

Accor’s position is to consider data science as a tool for rate management, not to replace human decision-making, Roquefort said. Even though the data science team enriches its algorithms with many factors to produce a more accurate recommendation, the best pricing decision is made by the one who knows the hotel, she said.

The revenue manager who receives the recommendation can either accept it and set the price or make a change, she said.

“We could have automated everything, but this is not our position,” she said. “We consider AI and algorithms as a way to make humans more intelligent and not replace it.”

Scandic’s data scientists create standardized, simplified reports for the company’s GMs to show a quick overview of what business is looking like to help them make decisions about rate, Blomquist said. The system allows for them to override the recommendation. The company tries to avoid “output overrides” in which someone changes the rate, he said. Instead, if the rate doesn’t seem correct, someone can access the system and add information he or she feels is missing to produce a more accurate recommendation.

The system does much of the pre-work and forecasts on a granular level, Blomquist said. The revenue managers evaluate what these say instead of just setting the outputs on their own.

“This was one of the things that maybe took the longest when we took that shift in 2010 and 2011,” he said. “Revenue managers had been setting rates themselves and had a view of this is how it should be.”

Evolution of data science
Since Scandic started using data science for determining rates, the system has evolved quite a bit, Blomquist said. The pricing structures are more dynamic and involve more parameters. The fencing has become clearer and the structures are more complex, but they are not more complex for the customer if it is fenced correctly, he said.

As the tools improve, the system can take more data to improve its capabilities to become more granular than in the beginning when it was just aggregated models, he said. The company is looking at the conjunction of the price and products through different distribution channels and segmentation.

“We can see when or what type of pricing and see different room-type behaviors, who buys what and where,” he said. “This is where it has become more and more important, especially with the way business has developed. As we’re more transparent with distribution, this has become a much bigger part than it was previously.”

To reach this level in optimization, everyone goes through a long training process to understand what the system is doing and how it works, Blomquist said. Users have to understand the pricing structure and why it looks like it does to understand the whole picture. When it comes to the data warehouse in particular, the company has some data scientists who build different models. They simplify it to make it easier for the revenue-management team to easily access the data and create tools to visualize what they’re trying to look for or build, he said.

“They don’t have to write long SQLs,” he said. “They can drag and drop what they need to have. All the relationships have to be built into the data warehouse to make it an easily used and accessible tool.”

The future of data science
Blomquist said that up until now, data science has been focused on hotels’ main product: the guestrooms. But it’s starting to move into other areas, such as F&B and meetings.

“Overall, we will have that holistic view of customers,” he said. “Not just customer types booking rooms, but total purchase behavior.”

As the technology improves, the recommendations will become more and more accurate, Roquefort said. This will allow hotel companies to move closer to dynamic pricing, allowing hoteliers to change pricing several times a day.

As the AI learns more about guest behavior, that will allow better and more customized recommendations for different guest offers, packages and discounts depending on the guest segment, she said. These will also depend on whether the guest is traveling on business or leisure and whether they are alone or traveling with family.

“These are things we are thinking about beyond perfect price prediction,” she said.

Currently these models are complicated and were once reserved for the big players in the market, Roquefort said. As the technology continues to improve and the costs for it decrease, this type of price optimization will become much more accessible to smaller, independent hotels.

No Comments

Comments that include blatant advertisements or links to products or company websites will be removed to avoid instances of spam. Also, comments that include profanity, lewdness, personal attacks, solicitations or advertising, or other similarly inappropriate or offensive comments or material will be removed from the site. You are fully responsible for the content you post. The opinions expressed in comments do not necessarily reflect the opinions of Hotel News Now or its parent company, STR and its affiliated companies. Please report any violations to our editorial staff.