How Sentiment-Analysis Software Helps Companies Assess Employee Attitudes - And Take Effective Action

In an age when human capital and the intellectual capital it generates often comprises much of the value created by organizations, it is increasingly important for those enterprises and their managers to get a more proactive handle on the attitudes of those who are the source of future success.

It is not surprising that technology and the data it provides contribute significantly to this process. What may be more curious - and beneficial - is the degree to which organizations that employ these means are building it into their strategic planning processes, as the following article explains. JL


Rachel King reports in the Wall Street Journal:

Sentiment-analysis software gauge(s) how employees feel about everything from diversity to their prospects for promotion. Such tools let managers automatically sort through hundreds or thousands of comments to get a reading of where management can make changes to improve the likelihood that employees will remain enthusiastic about the company and stay there.
In a competitive hiring market, many companies are trying to figure out how to better retain employees. And for many, the first step is figuring out how workers feel about the company.
Big data offers them a chance to do just that.
A host of companies, including Intel Corp. INTC -0.42 % , Twitter Inc. TWTR 2.93 % and International Business Machines Corp. IBM -0.86 % , are using so-called sentiment-analysis software to gauge how employees feel about everything from diversity efforts to their prospects for promotion. Such tools let human-resources managers analyze text such as internal comments on blog posts or responses to open-ended questions on employee feedback surveys.
The idea is to be able to automatically sort through hundreds or thousands of comments to get a reading of where management can make changes to improve the likelihood that employees will remain enthusiastic about the company and ultimately stay there.

Going deeper

While attitude surveys and internal blogs for employee discussions have existed for years, the data they produce are often unwieldy for companies to manage. Similarly, even when employees speak directly about an issue, there can be subtle signals—in their language, for instance—that simple readings won’t pick up on.
Intel, for instance, asks its 106,000 employees to complete an annual survey that gauges the company’s organizational health. The survey looks at a broad range of metrics from strategy to employee engagement. Some of the questions are designed to reveal how employees feel about the company. Workers are asked whether they’re proud to work at Intel and whether they hope to continue working there for five more years.
About 50,000 comments are generated by the survey, says Richard Taylor, senior vice president and director of human resources for Intel, and “we’ve gotten to a point where we have a hell of a lot of data and not necessarily that much knowledge.”
To glean more from its data, Intel earlier this year decided to try a product from San Francisco-based Kanjoya Inc. that uses language-processing and machine-learning algorithms to decipher emotions from text. The program analyzes the language workers use in open-ended survey questions and in blog posts to determine their underlying emotions, including frustration, disappointment or anger.

Put to the test

The software was quickly put to the test analyzing a set of internal blogs where employees could comment on diversity-related topics, including a $300 million, five-year diversity plan for the U.S. workforce announced at the beginning of the year by CEO Brian Krzanich.
While the blog comments were positive overall, the software identified agitation among employees who responded to a post by the company in June announcing it would pay double the employee-referral bonus if an employee referred a minority job candidate or a veteran and that person was hired.
Some who posted comments wondered if this was reverse discrimination or if Intel was breaking the law, says Mr. Taylor, who adds that while the employees were clear in what they were saying, the company used the sentiment-analysis software to drill deeper into what other motivations might lie behind their comments.
What the software revealed, Mr. Taylor says, was that ultimately people were expressing frustration and fear based on a misunderstanding—a wrong impression that their own jobs were at risk.
“It couldn’t be further from the truth,” Mr. Taylor says.
What the software also showed Intel, he says, was that the company needed to better communicate with employees going forward.
Some who posted comments wondered if this was reverse discrimination or if Intel was breaking the law, says Mr. Taylor, who adds that while the employees were clear in what they were saying, the company used the sentiment-analysis software to drill deeper into what other motivations might lie behind their comments.
What the software revealed, Mr. Taylor says, was that ultimately people were expressing frustration and fear based on a misunderstanding—a wrong impression that their own jobs were at risk.
“It couldn’t be further from the truth,” Mr. Taylor says.
What the software also showed Intel, he says, was that the company needed to better communicate with employees going forward.

Pulse at Twitter

Tech hiring in the San Francisco Bay Area is especially challenging because so many new companies are competing for workers there. “Making sure that we know what employees expect out of their experience at Twitter and the degree to which we’re living up to those expectations is incredibly important to us,” says Shane McCauley, director of people systems and analytics at Twitter.
To that end, Twitter this year redesigned its employee-pulse survey to include more open-ended questions, and started using Kanjoya to help it take the temperature of its workforce. Previously, the survey was given to employees twice a year and only had one or two open-ended questions. After moving to Kanjoya, Twitter began to give the survey to one-sixth of the employees each month, so there’s continual feedback.
There now are open-ended questions in every part of the survey. Kanjoya’s automated tools make it easier to analyze the text-based responses, says Subhadra Dutta, people scientist at Twitter. It’s better, she says, than traditional engagement surveys that contain a standard set of questions.
“By the middle of the survey, people start getting bored and hitting 3, 3, 3, and then you have a data set that is so highly neutral that it is hard to do anything,” Dr. Dutta says of traditional surveys. With the new survey, executives also love seeing the results of the qualitative questions, which in the past were much harder to analyze, she says.
One concern of companies conducting these kinds of analyses is the privacy of employee communications. Some experts say analytics software should only be applied to communications where an employee has no expectation of privacy. Most companies only use employee sentiment analysis in certain contexts, such as feedback surveys or postings on internal corporate blogs.
“We’re only going to do it where it’s very clearly been an employee statement in a known public forum where they know their stuff is being looked at,” says Intel’s Mr. Taylor. On internal Intel blogs, for example, employees must attach their real names to comments. In the company’s eyes, such posts thus are acceptable for analysis. Employee emails, by contrast, are private, says Mr. Taylor. The company won’t analyze those.
“We would lose the trust of our employees if we did that,” he says. “That would be the worst thing.”

Getting social

Some companies are analyzing social media to determine employee sentiment. IBM began a project known as Enterprise Social Pulse several years ago to look at employee posts on an internal social-networking site. The data is first rendered anonymous and then analyzed for sentiment. Subjects that employees talk about most are weighted most heavily. This has served the company as a barometer for issues that employees may have strong feelings about.
The company learned some valuable lessons along the way—that sarcasm and humor can be difficult for an algorithm to detect, for example, and that taking the pulse of employees should only be considered a first step. Action may be necessary to respond to issues that it identifies.