What’s the Price Now?

What's the Price Now? photo illustration

It is fairly common for travelers to begin their searches for the ideal itinerary online, only to find the price of a plane ticket or hotel room has changed dramatically in just a few hours. Checking back on one of the many travel comparison websites a few days later will only result in more exasperation, as a hotel room price may have gone up, while the cost of the airline ticket has gone down.

These industries rely on pricing models that try to eke the most profit out of a plane ticket or hotel room. The constantly changing selling price is by design: travel providers have decades of experience in using dynamic pricing models based on supply and demand, which help them balance prices with seat and room availability. Yet there is far more to dynamic pricing than basic economics, and a number of industries—including some unexpected ones—are adopting this concept to capitalize on this growing trend of price optimization.

What is really enabling the growth of dynamic pricing is the availability of petabytes of data stemming from the billions of transactions that consumers have conducted with businesses on a daily basis. Coupled with advances in software, inexpensive storage solutions, and computers capable of analyzing thousands of variables, this vast amount of data can be crunched almost instantly to help companies optimize pricing on everything from a pair of shoes to a luxury automobile.

There are dozens of software firms offering intelligent pricing software, many designed with specific vertical markets in mind. At the center of each vendor’s offering is their own proprietary algorithm used to help the deployer maximize profits. The algorithms are an integral part of the software, taking in all of the data and making pricing recommendations in near-real time. The companies deploying the software also can customize the weight of different inputs to help achieve their desired outcome. Online retailers that want to ensure pricing for "hot" items is in line with competitors may choose to adjust prices more frequently than on slower-moving products.

What is enabling the growth of dynamic pricing is the availability of petabytes of data from billions of consumer transactions.

Airlines in particular have become adept at using virtual warehouses of data to help set pricing at the seat level. The result: passengers sitting in the same row on a plane, just inches from one another, may have paid drastically different prices for their seats. Supply and demand are certainly key factors that determine the cost of a ticket, but competitors’ prices, seasonality, the cost of fuel, and a host of other variables go into the computation of prices, all based on complex algorithms.

While airlines were the pioneers in the use of dynamic pricing science, the retail industry has emerged as a leader in innovation and implementation of price optimization systems.

According to a November 2013 Retail Info Systems News survey, 22.6% of online retailers are utilizing pricing intelligence software to assist in price optimization. An additional 35.6% of survey respondents said they expected to implement such systems within the next year. If these retailers followed through with their implementation plans, that means more than half of online retailers are using pricing intelligence systems today.

Arnoud Victor den Boer of the University of Twente’s Stochastic Operations Research group says, "The emergence of the Internet as a sales channel has made it very easy for companies to experiment with selling prices." In the past, companies had to replace price tags on shelves or print new catalogs when prices changed, but online retailers have much more flexibility, and can instantly change prices based on market dynamics. "This flexibility in pricing is one of the main drivers for dynamic pricing," says den Boer.

The ‘Amazon Effect’

What might be called an ‘Amazon effect’ is prompting retailers across many vertical markets to take a closer look at price optimization strategies. Pricing intelligence software can scan competitors’ websites as often as every 10 minutes for their pricing and, using pre-set rules, automatically raise or lower prices on products to stay competitive.

Den Boer says online retailers are leading the pack in the use of dynamic pricing, "but a growing number of companies in ‘offline’ settings are starting to acknowledge the possible advantages of dynamic pricing. The idea that your customer has a ‘personal’ willingness to pay which you would like to learn/exploit is very appealing."

Dynamic pricing is being used in areas that might, at least on the surface, seem to be incompatible with this technology. Such models are being used to determine the cost of the cars we drive, the amount we pay for parking for that car and, perhaps one day, how much we spend on fueling the vehicle.

One of the most complex uses of dynamic pricing may be in the retail automobile market. A car is generally one of the largest purchases a consumer will make in their lifetime, second only the purchase of a home. Santa Monica, CA-based TrueCar Inc. has developed a proprietary methodology designed to help the three primary constituents in the car-buying process—the manufacturer, the dealer, and the consumer—complete a deal that is equitable for all parties.

John Williams, senior vice president of technology at TrueCar, says, "Dynamic pricing within automotive retail has been an ever-present market force; however, it is in many ways more complex and more foundational to the process than in other industries. It is often the case that two different shoppers can walk into the same store, on the same day, buy the exact same car, and pay completely different prices—sometimes as much as a 30% difference." The same pricing phenomenon may also be observed in other markets, but is amplified in a car purchase, where the price tag is in the tens of thousands of dollars.

According to Williams, "What’s especially fascinating about this inherent price variability in automotive retail is that it generally is inefficient for both the buyer and seller." TrueCar surveyed car buyers, asking how much profit they thought a dealer makes; consumers on average believed the average margin on a car is 19.7%. When asked what a fair profit margin would be, car buyers said 13.2%; in reality, the average dealer profit margin is closer to 3%–4%. "This shows that there is a huge opportunity to produce better outcomes for both buyers and sellers within the marketplace," says Williams.

The volume of data TrueCar must crunch is huge, requiring parallel computing and an enormous amount of storage. The company compares buyers’ needs with dealers’ existing inventories, adding in variables for what other buyers have paid, vehicle condition, demand for specific cars, and hundreds of other data elements. The result is a complex algorithm sitting on top of a massive data warehouse. TrueCar’s Williams said, "We built a multi-petabyte Hadoop cluster for $0.29/GB, a truly game-changing price point. Storing all the ambient data within a marketplace provides a powerful historical record to use for many dynamic pricing applications, including techniques like machine learning. Historical data is used to train machine learning models; the more history you can feed it, the smarter the algorithms are, and that yields a marketplace advantage."

New distributed computing solutions are changing both how data is stored and processed. Hadoop includes a highly scalable file system called HDFS that distributes data across many high-performance servers. Because those servers also have powerful CPUs and plenty of RAM, data crunching is handled by distributed programs that run on many cooperating servers at once. Techniques such as Map Reduce, originated by Google, are used to analyze huge amounts of data on Hadoop clusters that can contain thousands of servers. TrueCar uses this approach to model the automotive marketplace, creating giant lookup tables that contain every possible outcome to the many variables within the ecosystem. These lookup tables are then pushed into front-end databases and provide lightning-fast "real time" results when users begin adjusting options and exploring different possibilities. The next step in the evolution of these systems is real-time streaming of data and continuous data processing across the Hadoop cluster using technologies such as YARN and Storm, which manage resources for even faster data processing.

TrueCar has revolutionized the car-buying process; more than 2% of all vehicles sold in the U.S. are now through TrueCar.

Parking

Another unexpected market segment utilizing dynamic pricing is the parking industry. Up until a few years ago, companies that operated street parking essentially counted the number of coins in a meter to determine demand for parking; not only was this inefficient, but the operators also had no idea when parkers were coming or going, or how long they were parked. Municipalities worldwide have adopted electronic meters capable of not only accepting credit cards, but also able to set pricing based on space availability and time of day. Sensor-equipped meters are connected to a hub that can tell how many spaces are available on a street and adjust pricing accordingly. The goal is not to gouge consumers, but rather to optimize the price so there is always a space available—if the customer is willing to pay the price. Periods of lower demand mean lower prices, but when demand is high, customers should expect to pay more.

An important factor in dynamic pricing is the ability to develop "decent, well-performing scientific algorithms to determine sales prices based on accumulating sales data."

In the retail petroleum market, there is already a significant amount of variability in the price of gasoline at the pump, generally based on the price the gas station has negotiated with its supplier. This market has yet to embrace dynamic pricing at the pump, but it could be an area where we see some development. A customer pumping 20 gallons of gas, for example, might be offered preferred pricing when compared to another customer that only pumps five gallons. Discounts are already offered at some gas stations for customers that pay using cash rather than credit cards, and the next area of development in this market could take into account additional variables including the number of gallons pumped, the time of day, or even the weather (offering discounts on snowy days when demand might ordinarily be lower).

Start-Ups

Start-ups look at dynamic pricing as a relatively untapped market. Even in the travel industry, where dynamic pricing models have been in use since the 1980s, there are significant opportunities. At the November 2014 PhoCusWright Conference, a venture capital pitch-style event for travel companies, a number of presenters’ business ideas were built upon dynamic pricing models. Options Away, for example, was the first company to present at the conference, offering a unique take on airline ticket purchasing. The company’s product enables potential travelers to purchase options on ever-fluctuating plane tickets, locking in prices while they firm up travel plans. The company has already successfully partnered with leading ticket providers.

There is no lack of innovative ideas based on dynamic pricing. The success of dynamic pricing can be highly reliant on all of the variables in a given market. According to den Boer, success may depend on the ability to understand customer segments and customers’ willingness to pay different prices, as well as the ability to collect and interpret historical data. Another important factor is the ability to develop "decent, well-performing scientific algorithms to determine sales prices based on accumulating sales data."

The horizon for this growing field is wide open. There are already applications in a far-ranging number of fields, from credit card issuers determining customers’ interest rates based on their credit scores to the price riders pay for a car ride on Uber during peak hours. In the coming years, developers will continue to work on new and innovative ideas using dynamic pricing models, and it could alter the dynamics of an even broader range of industries.