Social networking, blogs and other forms of Internet content are generating a tidal wave of opinions on subjects topical and obscure. Usually, the vast majority of these opinions are ignored, dismissed as having little value. But some argue these opinions, if aggregated, sifted and analyzed correctly, could offer valuable insights into the public's perception and mood on individual companies, people and products.
Seth Grimes, founder of consulting firm Alta Plana, says that so-called "sentiment analysis" offered by several technology companies involves sophisticated algorithms intended to listen to customers and understand what they're saying and in what context. "For brand and reputational management this is a real growth area," he says. In the case of financial services "there's a lot of opinions about sales and bonuses, and an organization might want to factor that into how they present themselves."
Larry Levy, CEO of Jodange, a firm that offers sentiment analysis technology, says "a bank is a business and should always be looking at how they're being perceived in the marketplace...who is saying what about your brand, competitors, executives, products and marketing. You need to be monitoring social media as well as traditional media and those lines are blurring." Levy says Jodange can assess the "buzz" around a company, monitor every mention, and identify the key opinion holders.
For instance, to assess "buzz" Jodange crunches mountains of data to determine an aggregate "opinion momentum." As an experiment, Jodange analyzed all the mentions of Countrywide over two years, drawing on hundreds of thousands of mentions in blogs, tweets, news articles, TV and radio. The experiment took only a day, and the result was that Jodange could correctly predict the direction of Countrywide's stock the next day 70 percent of the time.
Jodange can also isolate major influencers to see, for instance, what publications have the most effect on certain stocks. It found that when the Web magazine "Seeking Alpha" opined on Countrywide, the stock reliably popped in one direction or another, but that wasn't the case with other stocks, such as Fannie Mae. Knowing the key influencers in the Web-osphere could help a company know who to talk to, he says.
As promising as these applications seem, there's still plenty of skepticism around sentiment analysis. How, for instance, can an algorithm divine all the subtleties of language? Consider the sentence "I just love it when my bank charges me $30 in overdraft protection fees to buy a $2 cup of coffee." Algorithms don't get sarcasm. "I'm skeptical," says Marc DeCastro, a senior analyst at Financial Insights. "It's mind boggling what there is to monitor."
But Jeff Catlin, CEO of Lexalytics, another provider of sentiment analysis technology, says people shouldn't focus on individual instances where the technology fails. The point is to crunch so much data that you get an aggregate direction. "If we're right 75 to 80 percent of the time, we don't care about any single story." Several hedge funds already use Lexalytics to gauge the direction of stocks, he says. Also, Cartesian Trading, a company that provides data to day traders, makes daily recommendations on about a dozen stocks using the technology.
The Financial Times also uses Lexalytics as part of the technology behind its Newssift service. Anyone visiting the site can type in the name of a bank or bank executive and the engine serves up a pie chart with a sentiment breakdown: positive, negative and neutral sentiment. On September 14, for instance, a Bank of America search generated 5,801 sources of sentiment from blogs, online news, newspapers, magazines, TV, radio and research. The Newssift chart showed 30 percent were positive, 29 percent negative and 41 percent neutral, with the company trending neutral after an up and down 2009.
Today, most financial institutions not involved in trading use sentiment analysis to better understand internal customer data. SAS has one large U.S. bank client that uses the SAS sentiment analysis manager to assess customer calls, emails and faxes pertaining to its loan modification program. By assessing the language used by the customer in each exchange, the bank can filter out those who don't qualify for a loan modification, and to rank which customers it must attend to first in order to prevent attrition. "It's a way to handle the sheer volume of inquiries they're getting," says Manya Mayes, chief text mining strategist at SAS. SAS says its sentiment analysis manager is the industry's first system combining a statistical method for computing reviews as well as a rules-based approach that lets brand managers evaluate certain specific terms and syntaxes.
Another example of a financial institution using sentiment analysis is the Navy Federal Credit Union, the largest credit union with $40 billion in assets, 3.3 million members and 179 branches worldwide. Software from SPSS helps analyze the unstructured data in the comment fields of the credit union's frequent customer surveys. In September, Nucleus Research gave Navy Federal Credit Union a 2009 Technology ROI award, saying that the credit union recorded a 1,531 percent return on its technological investment in a two-month period, with the annual benefit equating to almost $1.5 million.
Ron Shevlin, a senior analyst at Aite, says banks would be wise to consider these sentiment analysis tools only within a broader strategic context. All banks should be developing "a sense and response competency," he argues, and any sentiment analysis tool must serve the strategic goal of creating this competency.