What All That Chatter Is Really Saying

Text Analytics Can Turn Customer Feedback Into More-Meaningful Insight

By Published on .

YORK, Pa. (AdAge.com) -- When Rebecca Gillan walks through the lobby of a Starwood hotel, she doesn't want to hear the guests talking about "good" or even "very good" service. That's because she knows "good" is only worth a six or seven when a guest fills out a customer service survey. She'd rather hear superlatives such as "excellent," "outstanding" or even "cool," because that's where the nines and 10s are.
Tom H.C. Anderson, managing partner, Anderson Analytics
Tom H.C. Anderson, managing partner, Anderson Analytics

While that seems to make sense, Ms. Gillan, VP-global market research for Starwood Hotel & Resorts, is going on more than gut instinct. She's got the text analytics to prove it.

Text analytics -- a general term for the mining and interpretation of written words -- has been used for more than two decades, most notably by the defense industry as far back as the Cold War to read into the word choices and text of, say, a speech written by Russian leader Mikhail Gorbachev.

Listening to the boss
However, today it is marketers who are increasingly turning to text analytics to mine information from the mountains of customer data they've accrued from customer-service surveys, e-mails, online forums, hosted feedback sites and user-generated blogs. It's a way to listen to the boss -- customers -- and hone marketing based upon those insights. "Any company research is what a company thinks is important to them at that time," said Robin Korman, VP-loyalty marketing at Starwood. "But in the blog atmosphere, participants decide what's important to them."

Starwood hired Anderson Analytics for several projects, including the one that culled through the more than 1 million customer-feedback surveys -- 250,000 with comments in 20 different languages -- to find that "good" vs. "outstanding" insight.

"You can have someone read through 100 comments, and they will likely overstate the importance of some concepts, understate the importance of some concepts and totally miss other things," said Tom H.C. Anderson, managing partner, Anderson Analytics. "For instance, if one person in 100 mentioned something, it would be missed. But if in 100,000 responses, 1% of people say the same thing, it could be noticed as important, like a new trend that's developing or something wrong with a product that's just starting to surface."

That core value of text analytics -- plucking qualitative "aha's" from hundreds of piles of unstructured text -- is complemented by other marketing tactics, including competitive-intelligence gathering, real numeric tallies of positive and negative comments, short-term-goal checking and long-term-capital-spending value.

Harvesting comments
Starwood, for instance, in another Anderson text-analytics study of frequent-traveler website Flyer Talk, discovered that its guests discussed beds and showers more favorably than other hotels, while competitor Hilton's guests more often discussed food and health clubs positively. That validated the "tens of millions" spent on new beds in Starwood hotels, , Ms. Gillan said, while also giving them new areas to work on. Starwood is now focusing on healthful foods. And those insights all were delivered at a much lower cost than commissioned research.

In recent work for Unilever's Dove brand and its Pro-Age marketing campaign, Anderson went digging for consumer insight on Dove's own message boards, coding the text content against 43 different psychological attributes. Anderson found the vast majority of women who posted comments appreciated the realness of using older nude models. But they also discovered other common sentiments. For instance, most women over 50 strongly dislike the concept of "perfection" in beauty images. They also often talked about their mothers, grandmothers and daughters with concern about their portrayal in media. In fact, two in 10 women expressed real anger at how other advertisers portray women.

"Text analytics is a new methodology for us, and we were very pleased with the results and the depth of insight," said Catherine Cardoso, associate insights manager at Unilever, in a statement. "The results were helpful beyond understanding reactions to our campaign. We also gained an understanding of what motivates people on discussion boards, which issues are most important to women in our target group, and how to create better products and messaging for them."

More data
A surge in meaningful data online to mine -- and the ability of smarter, faster computers and programs to decipher it -- is one of the reasons text analytics is finally catching on with marketers.

"There is all this information online, and now you have an imperative to do something with it," said Seth Grimes, a consultant and technology analyst with Alta Plana. "Text analytics has grown hand in hand with the explosion in social media and online publishing."

For in fact, while the blogosphere and social networks have so far not proved great advertising media, text analytics offers the potential to make them stronger marketing vehicles. "A lot of marketers are paying a lot of attention to trying to figure out how to leverage social networking, for instance," Mr. Anderson said. "This type of analysis is an answer in how to unlock the value of those social networks."

The lessons Starwood learned

Good is not great. Guests who claim a good stay are only moderately satisfied. Guests who use superlatives to describe their stays are much happier and more likely to rate the hotel as a nine or 10.
  • Lurking works. The Starwood Lurker, also openly known as Starwood employee William Sanders, has a positive effect on customers on the website FlyerTalk. After Mr. Sanders interacted with guests on the message board, Anderson recorded spikes in psychological attitudes such as "respect for authority."

  • Ah, heavenly beds. Starwood's investment in great beds, such as Westin's Heavenly Bed, paid off in positive guest recognition.

  • Rates matter -- or do they? The most frequently commented-on term was "rate." However, further digging suggested there were almost an equal number of positive and negative comments. For instance, when one customer would complain about rising rates, another might jump in and defend it by explaining rising costs.
Most Popular
In this article: