Companies such as JetBlue have used it to mine e-mails and other customer feedback for insight.
There's nothing like an ice storm and angry customers to focus the mind. JetBlue Airways went through a trial by ice in February 2007, when a storm grounded flights at New York's Kennedy Airport and left its passengers stranded on the runway for hours. During the crisis and its aftermath, the volume of e-mail to the company soared from 400 a day to 15,000. Personnel couldn't possibly read and report on all the messages in a timely way--but the company needed to respond.
Not long before the storm, JetBlue had begun to explore text analytics, intrigued by new technology that promised real linguistic analysis that could mine for concepts, not just do the simplistic "counting words" approach. In the heat of the crisis, one text analytics vendor, Attensity, analyzed all the e-mail messages JetBlue had received and produced a report in two days. The report proved critical to the development of JetBlue's customer bill of rights, says Bryan Jeppsen, a JetBlue senior analyst.
The airline went on to integrate Attensity with other customer analysis. It uses Satmetrix's Net Promoter metrics, which gives JetBlue a score based on the percentage of customers classified as "detractors," "neutrals," or "promoters." That score is useful because it's clear, simple feedback everyone in the company understands. But "executives want to know why the score has moved up or down," Jeppsen says. "Was it because planes left on time or were late? Was it the average fare? The only ones who really know are the customers." By using Attensity to match specific comments and comment patterns with structured data such as arrival times, fares, crews, and gates, the airline thinks it can solve problems more rapidly.
Customers today aren't just customers--they're influencers and social networkers. Across the Web at any hour, they're sharing observations about your company's products and services, and those of your competitors. Customers amplify their single voices when they blog, write online comments and reviews, and participate in communities such as Facebook and Twitter. These new modes of customer behavior make it essential for companies to move beyond traditional ways of gathering, analyzing, and acting on customer information.
Customers also continue to reach out in conventional ways, such as feedback surveys and letters to the company. But the big difference today is that those surveys are often digital and the "letters" are e-mail--meaning it's in a digital form that should be easier to analyze.
In most organizations, transaction data is still the raw material of customer intelligence, and advances in the depth, breadth, and timeliness of transaction data analysis can help companies deliver competitive advantages. However, what's firing the imagination in many organizations these days is the potential to apply analytics to search results, text, and social network content to better understand customers and predict their behavior.
Analyzing Unstructured Content
This effort doesn't replace the work companies do to analyze structured data. Bringing structured data that measure hard facts such as transactions together with unstructured information such as e-mails and surveys that measure sentiment is essential to understanding customers. Of course, textual information, including forms, letters, survey responses, and warranty cards, has long been part of customer data sets. But it's the digitizing of this information that has interest in text analytics on the rise.
Text analytics, like data mining, is an umbrella term that covers a range of techniques and practices, including natural language processing, text mining, relationship extraction, classification and tagging, visualization, modeling, and predictive analysis. Compared with structured data analysis, text analytics is by nature less precise and complete; "good enough" is often the rule. Thus, text analytics can be most valuable when used in tandem with structured data analysis, particularly when an organization wants to combine or correlate customer predictors found in data and text.
Choice Hotels and Gaylord Hotels are both using text analytics software from Clarabridge to quickly make sense of thousands of customer satisfaction surveys gathered each day. The software spots positive and negative comments that can then be correlated with specific hotels, facilities, services, rooms, and employee shifts. The feedback drives immediate customer service response, with outbound calls or letters to acknowledge and apologize for problems and perhaps offer discounts to win over upset customers. More important, chain and facility managers track trends to spot problems and best practices. For example, Gaylord Hotels found out from using text analytics of post-visit surveys that noisy rooms weren't the most common problem, but they were among the most serious because they were most correlated to a "wouldn't return" or "wouldn't recommend" response.
Gaylord Hotels used to hire a company to sift through surveys, and it took eight weeks to get aggregated results. Using Clarabridge software, Gaylord now obtains verbatim results and trend information overnight, so it can immediately address service problems and make "rescue" calls or send e-mail to disaffected customers.
Rosetta Stone, the provider of technology-based language-learning product, uses IBM SPSS text analytics software to analyze answers to open-ended questions in surveys of current and potential customers. Along with text, the participants provide structured data in the surveys, such as identification information and product purchasing codes. Rosetta Stone uses the resulting insights to drive decisions on advertising, marketing, and product development as well as strategic planning. The company also uses the text analytics software to analyze online comments and reviews, including those about competing products, letting it size up its product strengths and weaknesses, says Nino Ninov, Rosetta Stone's VP of strategic research and analysis.
Know Your Customer "Sentiment"
Demand for text analytics has raised the profile of specialized vendors such as Attensity, Clarabridge, and Overtone, which do trended and basic root-cause analysis of why customers are commenting as they are. SAS Institute, IBM SPSS, SAP (Inxight), and Tibco (Insightful) offer tools for analyzing text for predictive insights.
One application of text analytics getting a lot of attention is "sentiment analysis," which lets companies discover positive and negative comment patterns in social media, customer reviews, and other sources. Lexicons, word extraction, pattern matching, and other tools and approaches are used to develop the knowledge. Just as Gaylord Hotels learned about noisy rooms, organizations can use sentiment analysis to find hidden factors that may be having a big impact on customer loyalty and churn.
Lexalytics, Nstein, and Teragram (a division of SAS) have text mining specialized for sentiment analysis, which often is applied to call centers. Verint Systems, for example, offers applications that analyze recorded calls for changes in customers' voice volume or the frequency with which they use certain words and phrases. The idea is to detect which issues are important and which have dropped off the radar.
Text analytics is maturing just in time to catch the "text tsunami" rolling across the Internet in the form of social networking, communities, messaging, e-mail, blogs, online reviews, and more-interactive company Web sites. And companies can get a view into a much larger universe of potential customers, not just those who have responded to offers.
A new role for customer intelligence in the era of social networks is to let companies discover and measure what's being said over the Internet that could influence customers. Query-oriented information discovery such as focus groups and surveys must take a backseat to behavior analysis based on observing people's activities and views online, and the network effect of that. Forrester Research has defined "listening platforms" as a distinct category that includes vendors of tools and services for brand monitoring and management, such as Nielsen Online (BuzzMetrics), TNS Cymfony, and Visible Technologies.
Listening can't be entirely passive, however. Visible's TruCast is an example of an application that helps companies both listen, discovering important and potentially receptive influencers within communities, and then understand when to act by joining the conversation, such as to counter negative impressions.
Text analytics offers the promise of helping companies understand much more about what their customers are doing before and after a sale. Organizations need to know what current and potential customers do when they are in the "consideration" stage, before they make a transaction, and after, when they express satisfaction or gripe. Analysis of data from search, social networking, and collaborative services helps fill knowledge gaps.
Again, though, that customer intelligence is most complete when companies have an integrated view of both structured data and unstructured information. Companies that already have a sound strategy for using business intelligence, data visualization, data mining, and data warehousing to wring customer insight from transactional data have a great base on which to add these new insights.
David Stodder runs Perceptive Information Strategies and is a research fellow for Ventana Research.