Listening to the Web: Effective Sentiment Analysis Makes All Departments Customer Centric

In an era when “green” and “sustainability” are feel-good buzzwords, it might be hard to imagine how anyone could be against an energy-saving light bulb.

Migros, Switzerland’s largest retailer, didn’t have to imagine, however. They detected the discontent by eavesdropping on conversations on the Web. With scanning tools that analyzed the Web for key words, Migros was able to identify a still-small movement called “Bulb Fiction,” which is trying to mobilize against a new light bulb the company is promoting. “Because of the identification of this movement we could, now, prepare a communication package ready to be launched in case ‘Bulb Fiction’ would become more vocal,” says Jann J. Hatz, vice president of corporate development for Migros.

Gazing at a tsunami of Facebook updates, tweets, and blogs, companies are pondering how to convert all this user-generated content into actionable information.

Customer sentiment analysis uses natural language processing, computational linguistics and text analytics to identify and extract subjective information from the web.  The tools offer dashboards and other visualization techniques that can reveal the volume and quality of the sentiment. In the case of Migros, the company scans for key words and judges the consumer’s conversation as “positive,” “negative,” “question,” or “comment.”

Jann J. Hatz of Migros

Customer sentiment analysis has already helped Migros avoid some touchy situations. For example, the company is sponsoring some open-air musical festivals. Online monitoring identified an upcoming PR risk because of complaints that a singer with supposedly homophobic lyrics was at one of those festivals.

“Knowing about the increasing number of comments on the topic we were able to proactively address the issue by setting up a conference among the interested parties to discuss the topic and work out a solution,” Hatz says. “In the end the festival organizers and the artist agreed to cancel the engagement and we got very positive feedback on all channels.”

Still an Inexact Science 
Anecdotes like this can make marketers and others whose livelihoods depend on the public’s mood giddy, but the reality is gauging the temperature of people on the Web isn’t easy. Customer sentiment analysis is an inexact science.   A sentiment analysis tool might judge a consumer’s comment about a product to be negative because his post included the word “crap.” However, the software might have missed the context: “Holy crap! This product is amazing!” As analyst Howard Dresner notes, “There are a lot of false positives.”

And even when the sentiment is accurately gauged, understanding what to do with it isn’t always easy. While sentiment analysis can indicate mood, it doesn’t reveal cause-and-effect.  “In sentiment analysis we can search for key words and count the incidences which already give some hints,” Hatz says. “The rest is educated interpretation.”

Consequently, he says Migros doesn’t systematically analyze customer sentiment in combination with sales figures (for example, matching public opinion with product sales). “The effort required is too big and that the customer sentiment data alone don’t give the full picture,” he says. “Therefore, not everything is analyzed.”

The Importance of Tangible Metrics 
Because of such challenges, Leslie Ament, vice president, research and client advisory of theHypatia Research Group, says  early adopters who are doing best with customer sentiment analysis pick a few key business initiatives that are directly tied to corporate goals and have tangible metrics.

She points to StubHub, a business unit of eBay that allows fans to sell tickets to concerts and other events to other fans.  The site facilitates the transaction, and guarantees the tickets will be valid, but doesn’t set the price. “For example, if a Cher concert sells out, the ticket sellers exploit supply-and-demand,” she explains. “StubHub is at the receiving end of a lot of negative comments, even though the problem is out of their direct control.”

As the volume of complaints rose, StubHub had one employee search them out one by one—but he could only respond to 20 a week. The company turned to software tools to find comments on LinkedIn, Facebook forums and other places. Then the comments were sorted by category of issue, such as return ticket or price issue, and judged to be  “positive,” “negative,” or “neutral.” This allowed an employee to respond to more than 200 conversations per week—a 10-fold increase in productivity. That was a slam-dunk for customer sentiment analysis, since the sites’ business model depended on people having faith in the service’s reliability.

The StubHub case, recounted in a new report by Ament, is not typical, however.  Ament‘s research indicates that more than half of companies aren’t even trying to measure the return on investment from their social analytics measures.  “Even the ones who are measuring ROI aren’t always measuring the right things,” she says.

Tips for Analyzing the Web’s Tea Leaves
Experts offer these points of advice for getting the most out of today’s first-generation customer sentiment analysis tools:

• Do use sentiment analysis to discover clues to consumers’ moods. But don’t assume any cause-and-effect relationship between online expressions and actual action without further analysis.
• Tie a sentiment analysis program to selected business initiatives that are tied to corporate goals.
• Remember that customer sentiment analysis is not for marketing alone. It should be integrated into your overall “voice of the customer” efforts, rather than be thought of as a standalone component.
• Understand your needs: Consumer-grade software might not be able to handle a large enough number of sources or volume for your purposes.
• Find ways to measure ROI of the efforts. Most companies don’t.

When All Departments Are Customer Service Departments
One reason is that customer sentiment analysis projects tend to be isolated in specific departments with their own agenda. Mitch Lieberman, vice president of strategy for CRM technology providerSword Ciboodle, points to the marketing department in an automobile manufacturer. “A  car that is advertised as getting 30 mpg,” he says. “People realize that the car actually only gets 15 mpg and begin to complain about it. Negative sentiment, for sure. What is marketing going to do? Giving comfort that 15 mpg is pretty good is not the solution—what else can marketing do in isolation?”

Instead, Lieberman says the consumer discontent should be directed to the right part of the organization (product, service, customer service) and then the resolution should be communicated back to the customer. He says companies at the extreme forward tip of customer sentiment analysis are figuring out ways for their customer service and marketing departments to work in conjunction with the information they unearth. But they have prickly issues like incompatible metrics—customer service is judged by customer retention, marketing by new customers. “People often talk about data silos, but operational metric silos are just as big an issue,” Lieberman says.

At this very early stage in the technology, it’s important to remember that customer sentiment analysis can be a useful tool when applied correctly, but it’s not a magic bullet or scientific research. “Social media is the new flavor of ice cream,”  Ament says. “It’s an important channel to understand customer feedback, but it’s not a panacea for every business problem.”

Joe Mullich, a freelance writer based in Los Angeles, can be reached at joe@joemullich.com. 

Source: Data Informed (http://goo.gl/9qZLw)