Voice of Employee Analytics: Step-by-Step Guide

  15 m, 17 s

Voice of Employee, People Analytics and Proving the Value of Human Resources

Retaining an employee is 21% cheaper than replacing one. But employee turnover is at a record-high. Human Resources departments are the frontline for this battle. Yet many departments lack the resources or data-driven approach they need to make a real impact. According to Mike West, the author of People Analytics for Dummies and the father of the “people analytics” movement in HR, this is because “it is not clear how to relate HR actions to business impact.”

British not-for-profit consultancy, Investors in People, says that real business value is realized when Human Resources is deployed effectively:

“If employees are able to speak up about what’s important to them and what’s achievable, HR can design incentive schemes that better orient employees towards meeting organizational goals.”

Investors in People

In this article we lay out the process for a straightforward Voice of Employee analytics program. We’ll demonstrate how a data scientist, workforce analyst or other HR professional can gather, prepare, analyze and interpret employee reviews and other feedback data to increase employee retention, reduce hiring costs and improve workforce productivity. And we’ll show how you can do this without any experience in coding or data science.

We start with the macro-level by investigating discussion topics and themes in a set of Glassdoor reviews, as well as their associated sentiment scores. Next, we dig into the details. We identify areas for organizational improvement, discover their root causes, and then we make actionable recommendations that further organizational goals.


Finding and Sourcing Employee Experience Feedback

Preparing Employee Feedback Data For Analysis

Creating a Configuration and Running an Analysis

Visualizing High-Level Discussion Topics and Themes

Identifying Areas for Organizational Improvement

Tracing Problems Back to Their Root Causes

Examining Individual Reviews to Gain Context and Humanize the Data

Proving the Value of Human Resources

Quick note: We used the Lexalytics Intelligence Platform to create analytics dashboards for this article. But you can apply the general processes outlined here to any employee feedback analytics tool.

Finding and Sourcing Employee Experience Feedback

As with any data analytics program, Voice of Employee is focused on answering questions. As data analysts, our question for this project is: How can we reduce people related constraints and further business goals?

The best data for people analytics comes from the employees themselves. This natural language feedback is known as “the voice of employee.” For this article, we gathered 6,000 employee reviews from Boeing, the aerospace/defense corporation.

To gather this data, we used the web scraping program WebHarvy to scrape Glassdoor, an employer-review platform. You can find an easy, step-by-step tutorial on how to do this here. WebHarvy is a point-and-click application. At $11 a month, it’s an affordable and easy tool for most business use cases.

Preparing Employee Feedback Data For Analysis

After gathering our employee feedback data, we export it into Microsoft Excel for cleaning.

Voice of Employee data sets usually contain at least three columns: ID, text, and date. An ID is merely an arbitrary number representing a row in the data set. The machine uses the ID to index each row of data. Text is the natural language document (like a Glassdoor review, survey response, or Slack message) that you want to analyze. The date indicates when the employee created the document.

Of course, your data may include many more columns beyond these items. These additional columns are called metadata: data about data. For voice of employee, metadata may consist of the manager on duty, the office location, the department or team an employee belongs to, or anything else you can think of that’s relevant.

Boeing voice of employee data set in Excel

Often a data set is stored as a .csv in Microsoft Excel; this can be directly uploaded to an NLP dashboard, like Semantria Storage & Visualization

For our analysis, we’ve opted to use eight columns: ID, date, title, text, rating, recommendation, outlook, and CEO. Title is the title of the review; rating equates to the number of stars the employee used to rank their employment at Boeing; recommendation is whether or not the employee recommends working at Boeing; outlook is their overall outlook for the company; and CEO is the employee’s opinion of the CEO of Boeing.

Voice of employee represented in two pie charts measuring the impact of negative management on an employee outlook of the company

You can use metadata to segment your data for layers of analysis; for example, we segmented our Boeing Glassdoor data set to see how negative interactions with management affect an employee’s likelihood to recommend the company

Next up is data preparation. Preparing a data set for analysis involves purging irregularities that might cause errors with your analytics tool. For example, you can check random data samples for misspellings or erroneous characters.

We prepared our data by simply scrolling through the data set and looking for irregularities. Then we fixed or deleted irregular data using the Microsoft Excel replace feature.

Once you’ve fixed or removed any problematic data, you’re ready to configure and run an analysis.

Creating a Configuration and Running an Analysis

When you (a human) talk with your friend (another human), you use context to gain deeper understanding. For example, you interpret their facial expressions, tone of voice, and past history. These elements help you understand the deeper meaning behind the explicit text.

When you talk to a machine (such as writing a review or posting a social comment), that machine doesn’t have that learned context (because it’s a machine). In order to account for this, data analysts tell the machine how to interpret specific words and phrases in a larger context. This process of adjusting for context is called tuning – and you create a configuration to “remember” your adjustments.

Remember: the goal of Voice of Employee analytics is to understand what employees are talking about, how they feel and why they feel that way. But without a proper configuration, the insights you generate will be questionable at best, and downright misleading at worst. 

Tuning is a critical step in any analytics project that involves natural language data.  But it can be hard work to build a configuration from scratch. That’s why some natural language processing vendors provide pre-built configurations, like the Human Resources Industry Pack from Lexalytics. Pre-built configs let you focus on running analyses, instead of building configurations.

Lexalytics software being used to tune a people analytics configuration

Without a proper analysis configuration, the insights you generate will be questionable or misleading.  Lexalytics offers a sleek graphical user interface for creating and tuning configurations.

Nonetheless, it’s important to understand that a pre-built configuration is a foundation for you to build on. For example, we are using Lexalytics’ Human Resources Industry Pack for our analysis of Boeing employee reviews. But we may also want to tune the configuration for Boeing specific entities, products, or programs.

Different vendors handle tuning in their own way. Lexalytics uses a point-and-click graphical dashboard that lets you take a pre-built industry pack and then further tune it to improve accuracy. The process is quick, intuitive, and code-free.

Visualizing High-Level Discussion Topics and Themes in Employee Reviews

So far, we’ve gathered, cleaned, configured and analyzed our employee review data. Now we’re ready to create visual analytics dashboards that will help us understand what’s in our data. To do this, we’ll use a product in the Lexalytics Intelligence Platform called Semantria Storage & Visualization, a business intelligence platform that gives you the tools to gather, analyze, and visualize text data.

Large data sets can appear overwhelming, especially when they’re full of messy natural language. Because of this, we recommend you start high-level. Instead of diving into the weeds, “zoom out” to observe relationships between entities and themes.

Visualizations that represent collocation help analysts zoom out on a data set.  In text analytics, collocation shows how frequently two concepts appear next to each other in a set.

This macro approach helps you tell powerful stories about the employer’s brand health, and then identify the root causes of organizational issues. HR professionals may want to begin visualizing their data through a widget like the sunburst chart, which you can find in Semantria Storage & Visualization.

A sunburst chart displaying voice of employee data

A sunburst chart represents the collocation of concepts within a data set; Semantria users can drill into segments of the sunburst chart for more context

In this case, we’re able to immediately ascertain that some Boeing employees express concern that ageism plays a role in the corporate pay structure. For a workforce analyst tasked with improving employee retention, an insight like this can make the difference between success and failure. More on that later.

In the meantime, with just one click we can see what else is being discussed in relationship to “ageism.” Or, we can drill into the underlying documents and hear the voices that support this claim.

Top concepts mentioned in the Boeing voice of employee data set, represented as a column chart

Here, we use the sentiment polarity visualization to illustrate Boeing as an employer. We can see that most of our documents (read: Glassdoor reviews) weighted as “very positive.” This means that employees mainly use positive language in their reviews.

But remember the goal of our project: to provide actionable insights drawn from employee feedback that can either improve retention or reinforce organizational cohesion.

It’s nice to know that employees are generally happy – but we need to go deeper. To do that, we filter the visualization to show only reviews that express somewhat negative or very negative sentiment (representing upset or dissatisfied employees).

Identifying Areas for Organizational Improvement

To generate useful insights, we’ll focus on areas with a high volume of negative sentiment. These documents demand our attention because they represent a threat to employee retention. But the same reviews can also tell us how to reduce employee turnover.

Our analysis of Boeing’s Glassdoor reviews indicates that the primary areas negatively impacting employee engagement and productivity are layoffs, technology, and pay structure.A column chart showing topics filtered by negative sentiment

Tracing Problems Back to Their Root Causes

Now that we’ve identified areas of employee dissatisfaction, we need to trace their root causes and provide our management team with specific actions they can take to address those issues.

Filtering on the topic of “Budget-Layoffs,” we see the sentiment of employees who are upset with layoffs within Boeing. This visualization reveals that the topic “Budget-Layoffs” ranks very poorly (-1.25) in reviews where employees complained about layoffs or budget issues.

A bar chart illustrating the average sentiment score for several topic in the Boeing voice of employee data set

Of course, this insight isn’t exactly news-worthy. And sometimes a review site like Glassdoor serves as platform for ex-employees to vent their frustration.  But an obvious trend can lead to valuable insights when you dig into the data. So, let’s dig in.

Examining Individual Reviews to Gain Context and Humanize the Data

Next we examine individual documents (Glassdoor reviews) to gain context and identify how this data impacts the organization as a whole. To do this, we segment our analysis by the “Layoff-Budget” topic. Our analysis reveals major, company-wide issues. For example:

“There is too much influx with employment. Employees live in fear of layoffs and too many politics with the company.”

A constantly-changing workforce means less productivity, and more errors by the stressed-out, distracted workers who remain. At an aerospace and defense company like Boeing, thisA cartoon of a stressed out worker can be disastrous.

Indeed, as we examine more reviews, we continue to see a trend of fear and confusion amongst Boeing employees:

“Lack of team-working due to continual competitive environment and fears of layoff.”

One employee points to an organizational flaw that may be responsible for the trend in the first place:

“They ramp up hiring WAY more than they should, only to end up going through layoffs down the road.”

To recap: we found that the topic “Budget-Layoffs” had a very negative sentiment score. But rather than accepting this as obvious and then moving on, we asked ourselves why. By digging into individual reviews, we revealed major issues with the way Boeing manages its workforce.

Boeing should re-evaluate their hiring needs, stabilize their workforce, and communicate more effectively. This will improve their profitability through better employee retention, lower hiring costs, and higher productivity. Not heeding this advice will likely hurt Boeing. Indeed, they already face $1.6 billion in pretax accounting losses due to retiring baby boomers.

This is a great case study in how Human Resources departments can use People Analytics and Voice of Employee to uncover big organizational issues and then trace the root causes into individual comments that provide vital supporting evidence.

Finding Another Issue and Tracing its Root Cause: Boeing Employees Want Better Technology

Now with our process in place, let’s examine other causes of negative employee sentiment. We’ll look at the topic “Employees-Quality,” which has a strongly negative sentiment score of -1.23.

A bar chart representing concepts related to the topic employee-quality

When we examine individual reviews, we find an unexpected cause:

“100 year old company and feels like it.”

In fact, this sentiment can be found through many of Boeing’s employee reviews. Looking deeper, we find that a major cause of employee complaints is their facility technology.

“Technology is archaic.”

The trend in negative sentiment continues, even disparaging the “old” digital applications still being used 2019.

“Lots of old applications still in production and by old I mean OLD and ugly, ColdFusion, etc.”

How might a workforce analyst or other HR professional use this insight? By bringing a visual, data-driven report on strategic capital expenditure to their management team, backed up with explicit comments drawn from real employee reviews.

In this case, IT might be compelled to design and implement a program to improve or replace critically outmoded software and technology. This serves to increase productivity overall while reducing employee churn.

And remember: In the age of People Analytics, Human Resources is no longer limited to hiring and firing. HR — once viewed as purely reactive — is using data science to take on proactive initiatives. By sharing the results of your People Analytics program with other teams, you will become a valued ally in the drive to improve profitability across the company.

Examining Pay Structure: How to Improve Retention

Our final visualization reveals how employees feel about the topics of pay and incentives. Let’s filter the data for the three wage-related topics: “Pay-Structure-Quality,” “Pay-Structure-Fairer,” and “Employees-Incentives-Morale.” These topics have average sentiment scores of –1.07, -0.96 and -1.89 respectively.A breakdown of Boeing employee opinion of compensation, represented as a stacked bar chart

Money is always a touchy subject in the employee/employer relationship. But drilling down into individual reviews uncovers  a troubling trend in Boeing’s compensation structure.

“They don’t promote or pay for performance but for time served”

Indeed, many negative reviews point to Boeing’s pay structure as a cause for anger, especially among young or newer employees.

“Poor starting pay. I would not recommend it to those just starting their careers”

“Big age gap between workers. Difficult to change anything.”

“Not a good place for young people to grow.”

This crisis among younger employees and applicants poses a real challenge for a company like Boeing. The aviation consultancy Leeham Co warns that Boeing risks losing thousands of engineering, technician, and touch-labor personnel to retirement over the next five to 10 years, with a limited talent pool for replacements. In fact SPEEA, the engineers’ union, reports that 29% of Boeing’s engineering workforce is over 55.

If Boeing’s HR team actively listens to their Voice of Employee, they’ll spot this trend and help the company’s management team avert a potential catastrophe. For example, Boeing can make reasonable changes to their compensation and incentive programs for younger and newer employees. Without a proper VoE analytics program, however, Boeing could face a major talent crisis.

Proving the Value of Human Resources by Transforming Data Into Recommendations

The abundance of data can be overwhelming unless it’s put into a clear and actionable plan for your organization. As an HR professional, you should use people analytics and voice of employee to bridge the gap between employee data, your organization’s business objectives, and the goals of your employees themselves.

The insights generated by your VoE program will be multi-layered. They will be as high-level as a general overview of employee morale across the company, all the way down to a granular look at how tech-savvy your sales team is and if they need new laptops.

We know that Human Resources departments are frequently mischaracterized as not contributing to business revenue. This misconception will be broken when you deploy a People Analytics program powered by Voice of Employee. Your Human Resources team will operate as the nexus between employee relations, workforce optimization, and overall profitability.

Analyzing Employee Reviews: What Did We Accomplish?

By beginning with a high-level view of topics and themes and then digging down into individual reviews, our analysis of Boeing employee reviews brought us:

  1. Clear areas of organizational improvement
  2. Detailed context, root causes, and supporting evidence
  3. Specific recommendations that their HR team can act on to increase employee retention and improve workforce productivity

This analysis serves as a roadmap for HR departments and data analysts alike. Follow these steps and the information laid out in our People Analytics and Voice of Employee: What, Why and How article to prove the value of workforce analytics while building a powerhouse People Analytics program in your organization.A cartoon woman on an airplane, representing a happy Boeing end user

Want to explore Semantria Storage & Visualization for yourself? You can play with it for free here.

Want to get your hands on the data set we used for this article? Contact us. We’ll send you a fully-prepared data set of Glassdoor reviews for the 5 largest insurance companies and the 5 largest aerospace/defense firms.

Categories: Analysis, Insights, Voice of Employee