Sales from mobile devices increased 101% in the first quarter compared to the same quarter last year for more than 350 retailer clients of ...
Big data, big opportunities
(Page 2 of 3)
Altogether, BeachMint, which doesn't reveal earnings but has raised more than $75 million since launching, spends around $100,000 annually on the Pentaho and Vertica software and services, Cohen says, plus monthly fees in the $10,000 range for the use of 10 to 15 servers from Amazon. But, Cohen says, "the cost is not that much when you consider the value it brings to the business. We're ROI-positive easily. I actually don't know how businesses operate without it."
For online classified listings site Cars.com, which sells advertising to car dealers nationwide, effectively analyzing lots of data is necessary to keep up with competitors, says Kevin C. Wyderka, director, data warehouse and business intelligence. Those competitors are mainly vendors that sell software to help dealers improve their web sites and online marketing, and they are increasingly turning to big data analytics to improve their products and services, Wyderka says. In response, last year Cars.com began switching from an Oracle Corp. data warehousing and analytics system to one from vendor Teradata Corp., which can process terabytes and larger amounts of information faster and at a much finer level, he says.
The largest database Cars.com stored with Oracle was about 350 gigabytes, and Cars.com added data to it at a rate of about one gigabyte per day, Wyderka says. Before, in order to process any business queries—for example asking how many customers viewed a particular car model then printed a map to the nearest dealer—Cars.com first needed to summarize the data, putting it in aggregate views per day, week or month, and then change it into a standard format, like a sequence of numbers, to feed into the Oracle reporting system, he says. Even then, it still took eight hours to return a requested report, he says.
In contrast, Teradata doesn't need data modified before running a query, Wyderka says. After five months of use, a single data set about one type of customer interaction with the site, such as product views, was one terabyte and growing at a rate of about 10 gigabytes per day, he says. When Cars.com queries that database, Teradata processes the request in one hour.
Wyderka says he's encouraged by initial tests of the new data-analytics technology. For example, Cars.com used Teradata to generate reports for all its 20,000 client dealers that show Cars.com search activity from shoppers at varying distances from their locations. That helps dealers to see that sometimes consumers are reading about cars from within a mile of their showrooms, perhaps sitting in a parking lot down the road and looking up dealer inventories and locations from a smartphone, he says.
The reports also revealed that consumers under age 30 rarely contact a dealer before visiting. Instead, they do most of their research online until they are nearly ready to buy, he says. That, he says, makes an argument for more advertising online versus on TV.
For customers like Cars.com that buy the hardware to create a Teradata data warehouse on their own premises, storage costs $32,000 per terabyte of data, a one-time fee that the company adjusts individually for clients as they add storage volume over time, Teradata says. Other Teradata services and support typically cost in the $10,000 range annually. Overall, that proved to be a more cost-effective investment for Cars.com than Oracle's competing Exadata system, Wyderka says. Oracle did not respond to a request for comment.
Books, media and music marketplace Alibris Inc., which had 2012 web sales of $96 million, has also recently begun adding big data technology to its e-commerce foundation. The retailer has more than 150 million items in its online marketplace, and it's hard to predict which type of customer might buy a particular book because some rare books have never been purchased at all, says Beryl Ness, director of e-commerce for Alibris Marketplace Services.
"Our recommendation engine couldn't deal with that lack of data," Ness says. Rather, it was designed to use a complete set of past purchasing data broken into subsets of similar customers to determine what a shopper fitting into that group would most likely buy, she says.
Aiming to improve results, last summer Alibris began testing technology from vendor Simularity that quickly processes millions of data points about customer activity, including items browsed, items purchased together, next item bought, wish list contents and social activity such as pins on Pinterest, tweets on Twitter or Likes on Facebook, to generate recommendations, Ness says. That way, if a customer Likes on Facebook a movie that no one has ever purchased from Alibris, the engine might still suggest that movie to him. That also helps expose more of the retailer's online catalog to more consumers.
The new recommendations initially showed a 2.9% incremental sales lift from suggested items, Ness says. Alibris then added the Simularity recommendations to thank-you pages and e-mails for order confirmations, shipping and follow-ups, with a plan to add them to marketing e-mails soon. The retailer expects a lift upwards of 5% in incremental sales once everything is in place and tuned, she says.
Altogether, the data Simularity processes is less than 100 gigabytes in size, smaller than what most retailers think of as "big data," Ness says. But, she says, for her, big data isn't really a matter of volume, it represents the ability to leverage "an explosion of new sources and types of data" all at once for a more useful result than old technology could provide, both in terms of relevance and speed.
To come up with the same recommendations as Alibris gets now using its previous engine would have required custom development costing $500,000 to $10 million, she says. With Simularity, Alibris pays well below the $90,000 to $120,000 other recommendation technology providers asked for development costs, she says.
"It levels the playing field a little bit as the price points come down," Ness says. "Big data technologies solve more complex problems at an investment level that now makes much more sense for business." Cracking those problems has been on his mind for two or three years, but she says adapting older technologies to solve them would have been both financially prohibitive, and less effective than what he can do with tools available today.