Friday, November 7, 2014

You're Being Watched: Online Stores Know Your Habits

Online Shopping
Booking travel or buying something online? There's a good chance you aren't paying the same price as the next guy.

New research reveals what some buyers have suspected and some companies even confirm: Online travel sites and retailers use information about customers to alter the items that show up in online stores. Occasionally, stores even show different prices to different users.
The algorithms are so opaque that researchers aren't yet sure how companies use consumer information to alter search results, and even developers themselves may not fully grasp the complexity of the methods.

"We're building revenue-maximizing robots," said study researcher David Lazer, a professor of political science and computer science at Northeastern University in Boston, said of the algorithms.

Personalized shopping

Given the type of information that retail companies can glean about consumers — from their search and purchase histories to their Web browsers and operating systems — it's no surprise that sellers would try to personalize search results, Lazer told Live Science. In many cases, personalization is arguably helpful. If Google knows that a particular user is more interested in bike tours than bus tours, for example, the search engine can make travel-related results more useful to that person.

In some cases, though, personalization has its downside. In 2000, a public outcry forced online retailer Amazon to back off from experiments in which users were offered different prices for DVDs. And in 2012, the travel website Orbitz experimented with showing more-expensive hotel rooms at the top of search results for Mac users than for PC users, based on findings that Mac users were more likely to spend money on pricier hotels.

Sniffing out price discrimination

Offering different users different prices for the same good is called price discrimination. Showing users different goods at varying levels of expense is called price steering.
It might seem straightforward to tell if a website is engaging in these practices, but it's actually quite complex,said study researcher Christo Wilson, a computer scientist at Northeastern University in Boston. Inventory can change quickly, altering prices and results, and searches from different computers might be run through different data centers, he said. 

To test whether websites were using consumers' information to alter prices or results, Wilson, Lazer and their colleagues had to come up with a way to control for as many of those unrelated factors as possible. They first tried a real-world experiment, recruiting people online through the site Mechanical Turk to run a code through their browsers that would automatically search retail and travel sites for certain terms.

"That lets us see, OK, these sites are actually personalizing [results] for users," Wilson said. "But it doesn't tell us why they're getting personalized results."

To find out, the research team set up fake accounts, varying single factors like the browser or the search history. Every search in each test ran simultaneously to prevent changes in inventory from affecting the findings. All machines had hard-coded IP addresses and were running through a data center on campus to prevent geographic differences from altering the results.
Even two identical searches run at the same time can return slightly different results because of random and uncontrollable factors, however. To control for this, the researchers ran identical searches in identical conditions twice. The differences between the two results could be dismissed as noise in the data. If two searches differed more than this baseline level of noise, the researchers could assume the search results had been deliberately skewed.

How your data is used

Of the 16 sites tested in the study, nine showed evidence of personalization based on user data, the researchers report in a new paper that they will present next month at the 2014 Internet Measurement Conference in Vancouver, Canada.

The researchers weren't able to tell what was driving the personalization in every case, but their experiments revealed the reasons on some sites. Travel sites Cheaptickets and Orbitz offered people with registered accounts slightly lower prices, an example of price discrimination (though one that people might expect as a benefit of membership, Lazer said). Priceline, another discount travel site, showed differing results based on user history. Retail giant Sears appeared to offer search results in varying orders for different users, though it wasn't clear why.
In a statement, an Orbitz representative said that account members do receive discounts, but that the programs are transparent and membership is free, making the arrangement a valuable deal for consumers. (Cheaptickets and Orbitz are owned by the same parent company.)
The experiment showed that Travel sites Hotels.com and Expedia randomly assign users to different groups, some of which see more expensive hotels first in the search results, a process called A/B testing.

In a statement, a spokesman for Expedia said that the company does not manipulate prices, but instead sorts results that they expect customers will prefer, based on trends seen in previous searches. The company does engage in A/B testing, the statement said, but only in an effort to improve customer service. 

"Presenting different booking paths and options to different customers allows us to determine which features customers appreciate most," the spokesman said. 

Skewing search results can help sites manage their inventory, Wilson said. But that's not the whole story.

"If it were just trying to spread people out [among the hotels], you could spread them out randomly," he said. "It's likely that this is an experiment to see if these people who they are showing the more expensive stuff to are more likely to book those hotels."

Travelocity offered cheaper prices to people searching from iPhone versus Android devices or desktop computers, but "the weirdest one is probably Home Depot," Wilson said. The home improvement store offers 24 search results per page for desktop users, and each item on page 1 costs an average of $100. Mobile users see 48 completely different results per page, with an average price of $230 per item.

"It's inexplicable," Wilson said.

A spokesman for Home Depot said that a number of factors go into the results users see on the company's search page, but that "we don't even have the capability to intentionally steer someone to a product.

"None of it is intentional, and none of the results are specific to any device or operating system," said the spokesman, Stephen Holmes. To understand why the researchers' results suggested otherwise would require "a lot of deep research," he said.   

The future of personalization

Based on their findings, the researchers say they now want to get a better idea of what user data drives search results. Theoretically, a system could differentiate between a comparison shopper and a compulsive buyer, and offer the two consumers different prices, Lazer said. Another goal is to study Facebook and its personalized News Feed, Wilson added.
For deal-hunting consumers, the new study offers frustratingly little in terms of advice. Algorithms change rapidly, Wilson said, so "Any concrete advice I give you today might not work tomorrow."

The best you can do, he said, is to search using a private or incognito browser window on your desktop, then repeat the search on your mobile device. "And if you're really paranoid, you should phone a friend or a relative and have them search to see if there is anything different," Wilson said.

It's likely that personalization will only become more ubiquitous in the future, Wilson said.
"The piles of data that companies have are just going to keep getting bigger," he said, "and there is a lot of pressure to try to monetize that."

No comments:

Post a Comment