USA 
United Kingdom
For every stride forward HR has made in its ability to listen to its employees, there is still an inescapable frustration from workers that their voices are not heard.
In a LeadershipIQ study of over 27,000 executives, employees, and managers, less than a quarter of employees say their leadership always encourages and recognizes feedback or suggestions for improvement. Indeed, whilst we have moved beyond the annual engagement survey of old to more sophisticated forms of employee listening, frustrations over compensation, recognition, wellbeing, flexibility, and other aspects of the employee experience are ever-present.
AI has already played a significant role in advancing capabilities for analyzing employee feedback. AI-driven algorithms allow HR teams to rapidly process huge data sets of employee feedback data, saving employee listening professionals from slaving over Excel spreadsheets or survey platforms. Every employee listening tool worth its salt will now offer some type of tool that uses AI to visualize trends in such data and even provide recommendations.
Processing qualitative employee feedback through open-text comments, however, is a different story. Open comments allow employees to share their opinions unshackled from Likert scales or forced ranking surveys. This freedom creates an outstanding opportunity to truly get to grips with employee sentiment, feedback, and even specific recommendations – but also the challenge of analyzing and drawing actions from hundreds, thousands, or tens of thousands of employees.
Has AI technology come far enough along, in the past year in particular, to support with this process? Or is software that promises to summarize answers to open-ended employee feedback comments too good to be true?
AI for sentiment analysis: Real or overhyped?
Open text sentiment analysis represents a significant frontier for HR. Historically it has been an extremely challenging and manual process for employee listening teams or HR professionals to review survey comments, codify them by topic, and attempt to draw themes from the text. Matthew Castillo Ph.D., Head of Employee Listening at Wholefoods, argues that qualitative data analysis is a time- and resource-intensive process. “Annual engagement surveys at large organizations can result in hundreds of thousands of open-text comments, which would take a team of analysts several months to sift through and derive themes,” he adds. It is no doubt this human limitation that means some feedback and suggestions are overlooked, rather than ignored, given the scale of the task at hand, hence employees become frustrated, feeling their voices are not heard.
AI and ML-enabled technology has made this process far easier. “I leverage AI as a starting point when reviewing a large volume of comments,” says Bryan Vermes, Director of Employee Experience and Communications at Mimecast. “It can be useful for identifying key topics or themes in what is being shared, especially as we’re all busier than ever.”
Neal Quinn, Manager of Product Management at Qualtrics, echoes that this process can be incredibly time-consuming. “But AI can accelerate the process and draw key insights across different employee segments to surface the most impactful areas for leaders to address,” he continues.
Castillo also agrees that AI is a useful starting point, adding that “AI can reduce the amount of work required to summarize comments and even provide verbatim examples of comments that map to major themes.”
There are limitations, and AI will only get you so far
No doubt then, there has been significant growth in the sophistication of sentiment analysis thanks to AI. However, there are still limitations with open comment analysis that still apply.
Before we come onto AI capabilities, there may be organizational barriers to implementation. “It will depend on organizations' comfortability with inputting potentially proprietary data into an AI tool,” explains Castillo. This, he argues, will depend on company policies and legal input from the company.
Moreover, AI can still struggle to understand content and nuances in language, particularly when meaning is implicit, ironic, or ambiguous. And particularly when raising frustrations or complaints, many employees will no doubt turn to irony or sarcasm, two of AI’s major shortfalls.
There are also difficulties in drawing out sentiment analysis when multiple topics are involved – and as matters regarding the employee experience and engagement surveys themselves typically cover a range of matters – performance, pay, reward, mobility, skills development, and so on, this can present limitations for the ability of AI-enabled sentiment analysis to draw out comprehensive and accurate themes from open text feedback. A further drawback comes with text in multiple languages, due to nuances in grammar, expressions, and vocabulary. This will be an issue for any major employer home to employees with more than one native language.
AI also falls short of delivering truly effective actions. Any sentiment analysis has to be supported by manager or leader involvement to ensure any actions taken are meaningful and effective. “The challenge for leaders is effectively synthesizing those comments into actionable insights,” says Quinn, adding that organizations should use AI as a tool to support leaders “by personalizing recommendations for their teams with the highest potential to drive improvements."
Vermes also argues for the need for human involvement to add ensure insights turn into meaningful action. “I advise that leaders still take the time to review the comments themselves and have discussions with their teams,” he notes. “AI isn’t able to replicate the experiences we’ve had within our organizations, and that’s the context you’ll need to plan effective actions.”
Castillo agrees, suggesting AI is limited in its ability to ensure text-based feedback is analyzed and acted on as it misses contextual information. “AI can provide key themes and recommended actions based on employee feedback, but the latter will likely be "off-the-shelf" recommendations with a one-size-fits-all approach that does not account for an organization's specific needs,” he argues. “It will be up to the leader to determine what actions will be most impactful for their area of responsibility that are within their ability to control. AI may be able to track the actions a leader takes based on feedback but will be unable to determine if they successfully effected change or if other organizational or internal and external factors led to changes in employee sentiment.”
Castillo recommends that companies work with internal analysts who have the business acumen to make the recommended actions more impactful than what AI could provide.
There’s no doubt that AI represents a truly exciting step forward for open comment sentiment analysis in the world of employee feedback. However, it must be used as a tool to augment, rather than replace, work from managers and leaders to drive action based on qualitative employee feedback.
