Sentiment analysis is being used to help HR discover what employees really think. It can help assess reactions to benefit changes or mine for problems in the workplace.
SPS Companies Inc., a steel and pipe supplier, is using an emerging technology to measure employee sentiment and get a better sense of how its employees really feel.
Corey Kephart, vice president of HR at Kansas-based SPS, wants the company's 600-plus employees to express themselves in surveys. One survey question prompted employees to write, in as many words as possible, "all of your thoughts, feelings and emotions about the person to whom you report."
"The more words that they can use and the more emotions that they can convey in the words, the more powerful, the more accurate" the employee sentiment analysis, Kephart said.
SPS is using the Ultimate Software UltiPro Perception tool, which assigns numerical value scores to assess whether a comment is positive, negative or neutral. Previously, SPS used online and paper multiple-choice surveys to assess employee sentiment. The surveys were followed by weeks of data analysis.
The employee sentiment software is giving near-instantaneous results. Employees answer questions that seek ratings, as well as open-ended questions.
Sentiment analysis, a new technology for HR, is trying to capture feeling, and it's not limited to surveys or open-ended questions.
The more words that they can use and the more emotions that they can convey in the words, the more powerful, the more accurate [the sentiment analysis].
Corey KephartVice president of HR, SPS Companies Inc.
KeenCorp, a company that focuses on HR engagement technology, uses another approach, "tension analysis," to monitor internal emails and chats to examine what's going on in the workplace. It may be able to flag a problem. For instance, if there is sexual misconduct in a department, the analysis may find normal patterns of engagement for the male population but what may be a "giant dip" in female engagement.
"We try to monitor when people self-consciously disconnect, and I think that is a really different way to monitor engagement," said Viktor Mirovic, KeenCorp's founder.
Tension analysis and sentiment analysis take different approaches but are based on broader technological developments.
"What these vendors dub 'artificial intelligence' rests on increasing capabilities in natural language processing and machine learning to produce predictive and prescriptive analytics that, for example, help organizations catch flight risk before a valuable employee leaves for another job elsewhere," said Brent Skinner, principal analyst at Nucleus Research.
Helen Poitevin, an analyst at Gartner, considers employee sentiment analysis an emerging space in the human capital management technology domain. "New techniques of natural language processing and more robust models continue to emerge, enabling higher accuracy rates and more pertinent insights," she said.
But Poitevin said users should view the market as "immature."
Poitevin doesn't believe sentiment analysis "has reached the degree of accuracy where you would see each and every sentiment tagging as correct. It remains useful and can provide useful insights, but some users may still be put off by the false positives or incorrect tagging of sentiment." She advises customers to test various solutions and monitor model performance over time.
These technologies fit under the broader category of emotion analytics. This is tech that can glean employee morale and feelings from verbal and nonverbal actions. The nonverbal technologies include video analysis. Emotion analytics can examine involuntary microexpressions, which are invisible to the eye but may capture a true reaction or feeling.
One vendor believes that its employee sentiment technology is mature. Armen Berjikly, senior director of strategy at Ultimate Software, said the company's employee sentiment tool was developed with the help of statisticians, mathematicians and psychologists, as well as machine learning and natural language processing specialists, to develop a process for classifying and summarizing open-ended comments immediately and accurately.
"Our emotion models are trained on millions of author-tagged documents, meaning our models have learned to identify emotions such as anger based on millions of real expressions by real people who were actually angry," Berjikly said.
At SPS, Kephart is pleased with the data he is getting. He believes it's 90% accurate based on his examinations.
The Perception tool's ability to sort results into different compare-and-contrast measurements, as well as display it on a heat map and present it on a dashboard, has gotten good feedback from upper-level management, Kephart said.
The speed and detail of the results mean that they can begin discussions quickly with managers, he said. They can give managers feedback "and pivot that into a conversation about what they could do to develop and grow as a manager into the future."
