kentoh - Fotolia
Calling data "the oil of the 21st century" has become a widely repeated maxim -- and customer data is seen as particularly high-grade stuff. Add in customer-related forms of big data -- social media posts, emails, surveys, customer service call transcripts -- and many organizations are sitting on vast reserves of valuable information. But the oil analogy only goes so far: It isn't enough to pump, process and refine customer data in order to extract its business value. Until someone takes action on the data, it's nothing more than an untapped well of 1s, 0s and text.
That's where customer data analysis comes in. And increasingly, organizations looking for a competitive edge are investing in multifaceted analytics programs designed to turn their customer-data crude oil into business fuel that can power their efforts to both attract new customers and keep the ones they already have happy.
For example, eBay Inc. analyzes a combination of sales records from its customer database and website activity data it gathers on users to create targeted marketing campaigns and personalize the homepage and other screens on its online auction site for individual visitors. Speaking at the 2014 Big Data Innovation Summit in Boston, Vadim Kutsyy, formerly eBay's head data scientist for big data applications and now the leader of a data research lab at the San Jose, Calif., company, said the analytics program has helped drive increased business on the site.
In addition, eBay is using the findings to try to avoid showing site users a constant stream of ads and available products that they aren't interested in -- a potential source of frustration and irritation for customers. Ensuring that people have a positive experience on the site -- and keep coming back -- is one of Kutsyy's top analytics priorities: As part of the program, he said, "I ask whether or not our customers see a benefit from having their data with us."
Kutsyy listed a variety of data management platforms and programming languages that eBay uses to power the analytics initiative -- for example, Hadoop, a Teradata data warehouse, and MySQL and Cassandra databases. But he said one of the keys to getting customer analytics right is to not get hung up on -- or religious about -- the underlying technology. Whatever tools an organization chooses need to serve the end goal of identifying what customers want and delivering it to them. Customers "don't care if we're on Hadoop or Teradata, or if we write [applications] in Java or Python," Kutsyy said. "They care that we make their lives better."
Customer analytics keeps the money coming in
Netflix Inc. has also made customer data analytics a key component of efforts to personalize its online streaming-media service for users and ensure that people are satisfied with the service -- and keep paying their monthly fees. Nirmal Govind, director of streaming science and algorithms at Netflix, said at the big data conference that he collects and analyzes data about everything users of the service do, including the movies they watch, how long their viewing sessions last and what kind of Internet connections they have. "We have a lot of data on how members consume content, what they like," he said. "And all that data can be used to improve the member's experience."
Like eBay, Netflix uses a mix of technologies as part of the analytics program, among them Teradata, Cassandra, the open source Apache Hive data warehouse software and Tableau's data visualization tools. After collecting and preparing data, Govind's team runs a variety of algorithms against it to determine things like what movies and shows it should recommend to users and what resolution videos should be streamed at based on their Internet bandwidth. Since Netflix began planning and producing its own original content in 2011, the Los Gatos, Calif., company has also mined data on customer likes and dislikes to help it decide which shows to make.
There have been challenges along the way. For example, Govind said that getting the recommendation engine right was difficult -- the recommendations are based on general characteristics, but it can be hard to accurately predict what a particular person will want to watch. To help fine-tune the engine, his team does a lot of A/B testing, giving multiple sets of users recommendations that are based on different predictive models and then tracking how much time each group spends streaming the recommended content.
Data analysts at payroll and human resources services provider Paychex Inc. are focusing on another aspect of customer data analysis: using churn models to identify customers that might be on the verge of taking their business elsewhere. In an interview at the 2014 Predictive Analytics World conference in Boston, Philip O'Brien, risk analytics manager at Paychex, said he and his team use data on company size, payment history and customer service interactions to build models designed to point to corporate clients that could be looking to leave the fold and others that are likely satisfied. Early on, the analytics team found that 21% of the customer service money earmarked to cover discounts on contract renewals was being spent on customers that probably would have stayed with Paychex regardless.
O'Brien said the Rochester, N.Y., company has since implemented a set of prescribed approaches for dealing with customers according to the characteristics highlighted by the churn models. But getting business managers to embrace analytical findings remains one of his biggest challenges. In the past, Paychex had more of a "shoot-from-the-hip" culture, he said. "When you're dealing with people who are used to using their intuition, you have to show them the benefit [analytics can add]." To help get people on board, he's leading an internal education campaign on the business value of customer analytics. Media articles on big data's potential benefits have also helped raise analytics awareness, he said.
Customer data danger: Knowing too much
There's a potential danger, though: going too far. Knowing when to stop is an important part of analyzing customer data. Businesses might have mountains of data, but using that information cavalierly could make customers uncomfortable -- and drive them away.
Andy McNalis, senior manager of big data and enterprise data warehouse administration, operations and development at Sears Holdings Corp., said at Predictive Analytics World that the Hoffman Estates, Ill., retailer analyzes customers' Web browsing history, past purchases and demographic data to help set and modify product prices in stores and online. But McNalis said he and his colleagues have access to some customer data that they just won't touch. For example, most Sears locations have in-store Wi-Fi networks, and the company can see when a customer is using a network to check prices on a competitor's website. That could be a good opportunity to offer the customer a coupon, but doing so would show people that Sears is watching what they do online, which McNalis said is "bordering on the creepy factor."
Customer data analysis efforts also take some effort, he added -- and there's more to it than deploying systems and feeding data into them. Sears uses Hadoop clusters and a Teradata data warehouse as part of its customer analytics program; the analytics team writes algorithms in programming languages such as open source R and runs them using Hadoop-based data analytics and visualization tools from Datameer. But skilled hands are needed to shape those algorithms into something that will generate useful findings, and to assess and analyze what they do find. "People think you can dump a lot of data into these things and a pony is going to come out the other end -- or maybe with Hadoop an elephant," McNalis said. "It's not."
Read more about how CRM analytics can help you get to know your customers better
Get tips on how data discovery tools can open up new marketing opportunities
Find out how sports teams are using customer data analytics to boost their business operations