Employee Mood Analytics: Don’t Just Measure the Symptom, Measure the Cause!
In this post, I share my thoughts on the article, “Employee mood measurement trends” by Tom Haak, which amongst other things, describes three main means of measuring employee mood: traditional surveys, simple daily feedback tools, and passive data mining of employee online communications (emails, Slack, Yammer etc.).
I share my assessment of each method. Furthermore, I discuss why passive communications mining is likely to generate data that is unrepresentative of employee mood. Instead, it is more suited for analyzing supervisor effectiveness, which is a leading indicator and arguably the most important determinant of employee mood (the symptom).
Finally, I opine that analyzing supervisor effectiveness through communications data mining could be combined with traditional employee mood surveys to generate actionable insights to improve overall employee performance.
Introduction
You wake up on the wrong side of the bed, feeling terrible about the fight you had with your partner the previous night. There’s no time to discuss, with little time to spare till your first meeting of the day. You dash out and within minutes, you’re at work, settled in front of your laptop. That’s when you see the email from your supervisor critiquing the slides you sent the previous night – the slides you worked on till late, prompting the fight with your partner. The critiques are polite but do nothing to mask the passive aggression written in-between the lines. There’s no acknowledgement of what’s good about the slides. Just bad, bad, bad! You feel gutted. Still you respond, “Thank you so much for your kind feedback.” The day only gets worse from then on, with you making atypical mistakes throughout.
Like you, every employee’s moods, emotions, and overall dispositions have an impact on job performance, decision making, creativity, turnover, teamwork, negotiations and leadership. It therefore comes as no surprise that an “affective revolution” has occurred over the last 30 years as academics and managers alike have come to realize that employees’ emotions are integral to what happens in an organization, says Sigal Barsade, Wharton professor and co-author of “Why Does Affect Matter in Organizations?”.
Taking this into account, companies began by instituting annual or bi-annual lengthy surveys to measure employee mood. These surveys afforded employees the golden benefit of anonymity but seemed too infrequent to credibly move the needle. This resulted in the emergence of pulse surveys which are shorter in length but administered more frequently. Then came the more recent rise of data analytics, enabling the passive scraping and analysis of data from employee digital communication platforms. Today, all three methods are being used to measure employee mood as a factor of productivity.
My Thoughts
Of the three, I believe that pulse surveys are most effective. While it shares the critical feature of anonymity with its lengthy traditional counterpart, it is far less tedious and may be more representative of employee mood, given its shorter length and higher frequency. Ever felt that sinking feeling when you click the ‘next’ button on a survey, hoping it’s the end, but only to find that you have 10 more pages to go? If you’re like most people, you’ll end up blitzing through the remaining questions and this effectively reduces the survey to a box-checking exercise, thereby yielding unrepresentative data. Pulse surveys are therefore increasing in popularity, according to Qualtrics. However, because determining employee mood may require analysis of a broader set of issues than a pulse survey can cover, traditional bi-annual surveys are still very common. To see other reasons why surveys remain a popular choice for determining employee engagement, see this HBR article.
On the other hand, passive data mining also has its appeal. Privacy issues aside, it surmounts the survey’s limitation of being a snapshot in time. It can be analyzed real-time and its proponents suggest that it is more objective than active data. However, it is important to ask what the data is representative of. If the goal is to measure employee mood, I’d argue that such data is most likely unrepresentative of reality. Take the prior illustrative scenario. The response, “thank you for your kind feedback” doesn’t shed any light on the emotions behind them. If anything, going by the words alone, one can infer that the feedback was received well, even though this isn’t the case. Little wonder, as a quick google search on email etiquette yields a plethora of articles which advise people to be polite and seek to not stir too many waters through their emails. In general, it is incredibly difficult to ascertain the intended tone of an email, and how it will emotionally be received by the reader.
Furthermore, the issue is not resolved by mining data from less formal channels like Slack. Besides the fact that workplace etiquette does not suddenly disappear when using these tools, modern day colloquialism, especially between peers, is often filled with hyperbole. Ever notice someone type LOL, without as much as a half-smile on their face? Or the excessive use of exclamation marks to appear enthusiastic, despite the writer’s bored countenance? These behaviors are further aggravated by workplace pressures to boost one’s likeability. While it may be possible to train AI algorithms to filter off such misleading data, questions persist about the precision and accuracy with which such tools can control for them.
However, these limitations can be surmounted, if the focus of analysis is on a key determinant of employee mood (the cause), as opposed to the mood itself (the symptom). According to the article under review, the most important driver of employee engagement seems to be supervisor behavior. Consider Google’s “8 habits of highly effective managers” and Sujan Patel’s 7 deadly sins of manager-employee communication, lists resulting from extensive research on the topic. It is possible to analyze the mined data to determine how effective supervisors are by measuring them against standards like these. Refer to my initial example. Whether or not the supervisor’s email is polite, it is possible to assess whether the critique contained therein was constructive and actionable. While such analysis may still not be able to accurately assess how the emails are received by the employee, the results can be combined with those from periodic pulse surveys to paint a fuller picture of employee mood. More importantly, actionable feedback can be generated and provided to below-average supervisors accordingly. But wouldn’t supervisors become more cautious in their communications once they learn of this? one may ask. Since the focus of measurement is on managerial effectiveness as opposed to tone, it won’t matter. At worst, it’ll cause them to communicate better and leverage habits that improve their overall effectiveness, which I can’t say is a bad thing.
Conclusion
To conclude, rather than analyze for employee mood alone, analytics of employee communications can be focused on improving supervisor effectiveness. This will in turn improve employee mood, which can be determined by existing surveys.
Interesting!!!
I am a true believer of identifying the root causes rather than implementing initiatives which addresses symptoms. I think this principle should also be implemented when building analytical models/ algorithms for which we use to make people decisions. Data scientists should to go beyond searching for surface correlations between variables and the outcome and search for the causes of the surface variables in order to make accurate predictive models.
I agree, Zubby. I felt similarly about today’s Hitachi case. Capturing employee ‘happiness’ or ‘mood’ seems to be measuring the symptom and not the cause. If I am an executive, I’d want to know beyond what my employee moods are, and get to why my employee’s moods are what they are! Great walk through of the different tools available to managers – it’s easy to forget that simple things like pulse surveys can help capture that data – it doesn’t require an expensive analytics program!
This is very interesting Zubby! I agree with Dan that in light of the recent Hitachi case, this seems like a problem that a lot of companies are trying to solve. As in the case discussion, I am curious to understand the bias in self reporting bias if the pulse survey is used as all employees will report what they want to show they are feeling rather than their true feelings. I wonder how important employee mood is in achieving organizational goals as it is hard to make everyone happy!