Text analytics—Improving the use case for unstructured text

"With this application," Halper says, "the users come up with sentiment scores on the executives, and the application even lets them drill down to the original articles the information was culled from and present pie charts that tell them the percent of comments that were negative, positive and neutral."

E-discovery

E-discovery is an application that was slow to start but now is coming on strong. When a company is sued, in the discovery phase they must present all information in their possession related to the subject of the suit and within a given time period. If it’s late in presenting materials or the materials are incomplete, it’s subject to fines or loss of the suit.

Companies used to need high-priced lawyers to comb the company archives for relevant documents. That was expensive and slow. Now with text analytics, they can extract much of the germane information in a fraction of the time by analyzing all relevant unstructured information in the company.

Hidden benefits

Text analytics offers valuable benefits in multiple applications and industries. Some of them might not be apparent at first glance because they are not related to straight dollar ROIs. For instance, with voice of the customer applications, companies get to know their customers better and can service them according to their stated needs. Better customer service, of course, makes companies more competitive and creates lower churn, which increases the level of purchases overall.

Cost

Because of the sophistication of the capabilities, text analytics starts at about $100,000 for an installed application and can run over $1 million, depending on the scope of the install, says Grimes. That would place it within the grasp of G2000s. The good news for smaller companies is there are software as a service (SaaS) versions that are quite affordable.

Grimes says, "We are at a stage where adoption in the general business world is being driven by end user business analysts, marketing folks involved with customer satisfaction, rather than by IT, and those people are maybe a bit impatient and willing to work with a hosted solution rather than go through IT." 

JetBlue "hears" the voice of the customer

JetBlue aspires to be a "value" airline with fares lower than the larger carriers and a better customer experience than budget airlines. To achieve that status, it offers more legroom in coach, DIRECTV and XM radio for each seat, leather upholstery and other amenities. It also demands topnotch service from its staff.

To help guarantee the best customer experience, it relies on solicited feedback from customers in the form of surveys and unsolicited feedback in the form of e-mail comments submitted to the SpeakUp link at its JetBlue.com Web site.

According to Bryan Jeppsen, analyst, Customer Feedback, in the average month, his department receives about 15,000 e-mails and 40,000 surveys. Of course, all the information in the e-mails is unstructured information. But the answer to one question on the surveys is also unstructured text.

"The question," says Jeppsen, "is, ‘On the scale of one to 10, are you likely to recommend JetBlue to your friends and colleagues?’ Customers are also asked to explain their answer." Those answers are examined to indicate customer loyalty. About 40 percent of the 40,000 surveys have answers to that question, says Jeppsen.

Before using text analytics, the five employees in Jeppsen’s department had to manually categorize, analyze and respond to all unstructured text from those sources to determine the "voice of the customer." Needless to say, the task was nearly impossible, and the staff was overwhelmed.

Because Jeppsen knew he was missing important feedback that could affect whether a customer flew with the airlines again, he adopted Attensity’s text analytics software to automate the process. Now, he says, "I simply load the raw data file into Attensity, and it automatically categorizes all the information and transforms the unstructured text into structured text."

Structured text from the survey indicates the plane’s route, day of flight, customer’s seat, plane’s tail number and the crew working that flight. If a customer indicates, for example, that a flight attendant was rude, the software would reveal what seats were affected and the flight attendants on the particular flight. Then JetBlue could take action to fix the problem and maybe give those customers free mileage to keep their business.

After the analysis is finished, Jeppsen says, he can export the transformed file into his Business Objects’ business intelligence tool so JetBlue executives can query the data, run reports and use it for decision support about long-term strategic activities of the company.