By Joe Mullich
Published: February 4, 2013
Last Updated: February 26, 2013
Companies have always wanted to know what their customers really think about their products, services, and their overall brand. Sometimes customers are more than willing to tell them, such as when they vent to service reps at a call center. More and more, however, much of this conversation takes place online on blogs, Facebook, Twitter, and other social media sites.
It’s one thing to get a sense of how people feel if you’re only reading five blogs or a few tweets. Making sense of thousands and thousands of comments pouring in from all sorts of social media in a timely way is another matter. That’s where customer sentiment analysis comes in. This fast-growing branch of technology attempts to turn all those tweets, comments, compliments, and rants into actionable insight.
The Voice of the Customer
For all the attention being directed at customer sentiment analysis, the concept of deriving insight from customers’ feelings isn’t new. A decade ago, companies like American Honda, Sub-Zero, and Whirlpool analyzed frequently-repeated words in warranty claims and service data, identifying potential problems and taking early action before they became widespread. Leslie Ament, vice president, research and client advisory of the Hypanthia Research Group, considers customer sentiment analysis to be a branch of voice of the customer initiatives.
There are more than 100 social analytics vendors that produce technology to decipher the tsunami of public opinion. Some of these tools analyze only one distinct type of information, such as Twitter and/or Facebook, blogs, RSS feeds, or certain media outlets. The most sophisticated tools, Ament says, have the ability to “drink from the fire hose” and analyze a vast array of online content.
While online comments have drawn the most attention, they aren’t the only places that use sentiment analysis. For example, sentiment analysis tools can be applied to the information that comes into a call center. In this case, the tools would analyze email messages or the transcripts of telephone calls.
Currently, customer sentiment analysis is primarily used by marketing departments. However, it can be leveraged for a wide range of purposes. Companies can:
- Identify how people feel about their brand in general.
- Track the impact of marketing efforts.
- Help develop ideas for new products and services.
- Identify public reactions towards specific events, such as product launches and keynote addresses by key executives.
- Determine the probability that a product or service will be purchased.
- Develop early warning systems that identify issues that are bubbling under the surface.
- Understand how the brand/product/service compares to the competition.
“With the advent of social media and the proliferation of reviews, ratings, recommendations, feedback and other forms of online expression, online opinion has turned into a kind of virtual currency for businesses looking to market their products, identify new opportunities and manage their reputations,” says Venkat Viswanathan, the CEO and founder of LatentView, a data analytics company.
How Sentiment Analysis Works
Customer sentiment analysis is a complex, multi-step process. The various tools work in somewhat different ways. Typically, they rely on natural language processing, which turns text like tweets and blogs into a format that computers can understand. And then artificial intelligence applies algorithms against that data.
A basic task in sentiment analysisis classifying the polarity of a given piece of material, which means determining whether the expressed opinion is positive, negative or neutral. “Now there are beyond-polarity solutions, which look at emotional categories — for instance, angry, happy, sad, frustrated, satisfied,” says analyst Seth Grimes who runs the Sentiment Analysis Symposium.
Customer sentiment is rarely a completely automatic process. Depending on the specific tool, a human analyst may need to set up categories for the type of insight you are trying to find and then train the software on how to classify comments based on the categories.
The information is presented in terms of metrics, trend analysis, and/or key performance indicators on dashboards or data visualization tools.
Why Is It So Hard?
Customer sentiment analysis is an inexact science. Theoretically, the accuracy of customer sentiment analysis is how well it agrees with human judgment. However, a group of humans can read the same piece of material and not agree whether the opinions are positive or negative.
Human speech is difficult to analyze. The software tools might not support every language or dialect. An ironic or sarcastic comment can be misinterpreted. For such reasons, customer sentiment analysis has many false negatives. For example, the word “crying” in a Facebook post might be considered to be a negative comment because the software doesn’t realize the phrase “crying with joy” is enthusiastic praise. A compounding sentiment – “I love my cellphone but hate my carrier” – can be difficult for the software to decode.
Customer sentiment analysis is often said to be successful if it agrees with human thought at least 80 percent of the time, but there are lots of debates on how to best measure it. Grimes says the accuracy level necessarily depends on what business problem you are trying to solve. “If you are trying to get the mood of people in response to a new marketing campaign or a presidential campaign, you don’t need to get into the nitty-gritty,” he says. “In the case of counter-terrorism, you want 100 percent accuracy.”
There are other potential issues. The comments coming in from, say, Twitter may not be representative of the entire customer base, so a small subset of vocal customers can skew your view of what true public opinion is.
An Invitation to More Customer Research
Even when customer opinion is accurately measured, that information doesn’t explain why the customer feels that way. “Social intelligence is not root cause analysis,” Ament says.” You can tell the reaction to something is positive, negative, or neutral, but you have to figure out the reason. It tends to be high-level trending information.”
For these reasons, sentiment analysis is usually a starting point that requires a company to dig deeper. For example, if an analysis shows that 11 percent of posts mention the word “problem” in conjunction with your product, that should be a warning to investigate. Usually the results from customer sentiment analysis have to be carefully studied, and sometimes merged with data from other sources, such as customer data, to understand what the “problem” might be.
Joe Mullich, a freelance writer based in Los Angeles, can be reached at email@example.com.