How Ebay uses big data to make you buy more

Online auction website Ebay uses big data for a number of functions, such as gauging the site's performance and for fraud detection. But one of the more interesting ways the company makes use of the plethora of data it collects is by using the information to make users buy more goods on the site.

With 180 million active buyers and sellers on Ebay, the website generates a lot of data. At any given point in time, there can be around 350 million items listed for sale, with over 250 million queries made per day through Ebay's auction search engine.

Speaking at the Big Data Summit in Sydney, Hugh Williams, Ebay's vice-president of experience, search, and platforms, said that the company typically holds 10 petabytes of raw data in its Hadoop clusters and Teradata installations.

Obviously, Ebay can't force users to buy every item that they come across, but the company takes advantage of big data to improve the chances of making a sale.

One of the ways it does this is through optimising its search engine and the results that come up by making tweaks based on customer behaviour patterns deciphered through the collected data.

"If you wound back the clock just a couple of years and used our search engine at Ebay, you might have found it was too literal," Williams said. "There were things you could express to the search engine that it would literally find, but didn't really understand the deep meaning of user intent that you had.”

"We've been on a journey to really make our search engine more intuitive."

For example, through using big data, Ebay found users looking to buy a Pilzlampe, a type of collectible German mushroom lamp, were more likely to make a purchase when they entered ”pilz lampe” in the Ebay search engine, as it would yield more results.

By simply putting a space in a word in the search engine, Ebay could improve the chances of a sale through the website.

With this kind of information, Ebay alters and rewrites the search queries made by users through its search engine, adding synonyms and alternative terms so that it can bring up more relevant results.

Not only that, Ebay uses big data to make predictions on whether a listed item will sell and how much it will sell for, which affects how high an item ranks on the auction site's search engine.

All of this can increase the likelihood of a user making a purchase.

But implementing factors to shape the way search queries are answered can be risky, according to Williams.

"It takes several months of engineering to implement a factor, and it's very high risk because we don't know at the time whether its actually going to be useful for our customers in helping them find items," he said. Which is why Ebay usually runs a number tests on the website for a sample group of users first to gauge the response.

Another challenge is taking the context of search queries into account. One example is if a user looks up “Geelong Cats”, Ebay's search engine may just take 'Cat' as the keyword and run it across the Pets category — not particularly useful when the user is searching for the sports team's merchandise.

"There are very subtle problems that can occur at our scale, so we need the likes of data scientists to investigate these issues," Williams said.