Guides

5 Steps to Analyze Voice of Employee (VoE) Data Effectively

Nov 03, 2026 - 7 minute read
5 Steps to Analyze Voice of Employee (VoE) Data Effectively

Most organizations already ask their employees what they think. They run an annual engagement survey, send shorter pulse surveys in between, ask new hires about onboarding and ask leavers why they are going. What many lack is a reliable way to turn those answers, especially the written comments, into findings that leaders trust and act on.

This guide is for HR and people-analytics teams that want to analyze this feedback properly. It covers what Voice of Employee means, how to protect the anonymity that makes honest feedback possible and five steps from collection to closing the loop. The steps work with a spreadsheet; the last part shows where a dedicated analysis platform helps.

What Voice of Employee Means

Voice of Employee (VoE) is the practice of systematically collecting and analyzing what employees say about their work, their managers, their teams and the organization. It includes ratings, such as an engagement index or an eNPS question, and the open-ended answers, interview notes and comments that explain them.

VoE is broader than a single engagement survey. A survey measures how people feel at one point in time; a VoE program connects several sources over time, so an organization can see whether a concern raised in exit interviews also appears in pulse comments, and whether it fades after something changes.

Why It Matters

Employees see problems before most dashboards do. A broken handover between two teams, a policy that works on paper but not on the night shift or a manager who has stopped holding one-to-ones shows up in what employees say long before it shows up in turnover figures. Listening systematically gives HR a chance to act while the problem is still small.

Feedback also explains the numbers. An engagement score that drops in one location tells you something has changed; the comments from that location tell you what.

Listening affects whether people feel heard, too. Asking for feedback and then visibly doing nothing with it teaches employees that their answers do not matter, while a program that reports back and follows through keeps response rates and candor high in the next survey.

Protecting Anonymity and Trust

Employee feedback differs from customer feedback in one important way: the people who give it work for the organization, and their managers may read the results. If employees believe their comments can be traced back to them, they stop answering or stop being candid, so protecting anonymity is the condition for having useful data at all.

Start by removing names and identifying details before anyone analyzes the data. That means more than deleting the name and email columns: open-ended answers often contain colleagues' names, job titles or descriptions that identify a person, such as "as the only nurse on the weekend schedule", and these need to be removed or generalized too.

Do not report results for groups too small to keep individuals anonymous. If a department has four people, its average score and its comments effectively belong to those four people. A common safeguard is a minimum group size below which results are not shown, with smaller groups merged, for example by combining tenure bands or locations, until every reported group meets it. Apply it to every cut of the data, including combinations such as department and tenure together.

Be clear with employees about who will see the raw comments, who will see only summaries and how long the data is kept. Use the right word, too: an anonymous survey collects no identity at all, while a confidential survey records identity but restricts who can see it.

Data protection law applies as well. Under the GDPR, information that can still be linked to a person, including data pseudonymized by replacing names with codes, remains personal data; only data that has been irreversibly anonymized falls outside the regulation, as the European Commission explains. Involve your data protection officer when you decide what to collect, where it is stored and which tools process it. Where policy requires employee data to stay inside the organization, the analysis itself can also run in-house, an option covered in the Kimola section below.

Five Steps to Analyze Employee Feedback

Step 1: Collect Feedback From the Right Sources

Different sources answer different questions, so a VoE program usually combines several. An annual engagement survey gives a broad, comparable picture across the organization. Shorter pulse surveys, with a handful of questions every month or quarter, show whether things are moving between annual surveys and let you follow up on a specific topic.

Employee Net Promoter Score (eNPS) is a common pulse question. It asks employees how likely they are to recommend the organization as a place to work, on a scale from 0 to 10. Following the standard NPS method, those who answer 9 or 10 are Promoters, 7 or 8 are Passives and 0 to 6 are Detractors, and eNPS is the percentage of Promoters minus the percentage of Detractors, giving a score between −100 and 100. As Bain notes, answers must be kept confidential to encourage honest feedback. A score on its own says little, so pair it with an open question such as "What is the main reason for your score?"

Onboarding surveys capture how new hires experience their first weeks while the memory is fresh. Exit surveys and exit interviews explain why people leave, and stay interviews, conversations with current employees about what keeps them, catch the same issues earlier. Internal feedback channels such as a suggestion form or an HR feedback inbox add a continuous stream between surveys.

Collect only from channels employees knowingly use to give feedback. Monitoring employees' social media or internal chat for sentiment raises serious consent and privacy concerns and erodes the trust a VoE program depends on.

Step 2: Prepare the Data

Analysis is much easier when every source ends up in the same shape: one row per response, with the open-ended answer in one column and the context in the others. Keep the columns that let you compare groups, such as department, location, tenure band, job level and survey wave or date. Keep the scores too, since a comment means more when you know it came with an eNPS of 2 or of 10.

Remove personal data at this stage, before the file is shared for analysis. Delete names, email addresses and employee IDs, replace exact values with bands ("2 to 5 years" instead of a start date) and merge categories below your minimum group size, so that every group in the file is safe to report. If you combine several waves, make sure each column means the same thing in all of them; a department renamed between two surveys will otherwise look like a real change.

Step 3: Find Themes and Sentiment

With a few hundred comments, a careful manual process works. Read a sample of answers without labeling anything, then write a short list of codes that describe what people actually talk about, such as workload, recognition, career growth, communication from leadership or tools and equipment. Apply the codes to every answer, adding new ones when a comment does not fit and merging codes that turn out to mean the same thing.

Expect one answer to carry more than one theme. "My manager is supportive, but the workload since the reorganization is not sustainable" is positive about management and negative about workload. Coding it as a single "mixed" comment loses both findings, so record each theme with its own sentiment. Once every answer is coded, count how often each theme appears and how often it is positive or negative.

Step 4: Ask Deeper Questions and Compare Groups

Counting themes tells you what employees mention. The next step is to ask more specific questions of the same comments: what creates difficulty in people's work, what they expect to be added or changed, what they say is behind a problem and what they value and want to keep. A complaint about workload reads differently once you know whether employees attribute it to understaffing, unclear priorities or a new process.

Then compare groups. A theme that appears in 15% of all comments may appear in 40% of comments from one location, or it may have doubled since the last pulse. These differences are usually where the most useful findings are, but they need care. Percentages from small groups are unstable: in a group of 12 people, two comments move a theme by almost 17 points. Before reporting a difference, check that both groups are reasonably large and that the gap is bigger than you would expect by chance, using a significance test such as a chi-square test if your team has the skills.

Step 5: Close the Loop With Employees

Share the main findings with employees, at a level of detail that keeps the anonymity you promised. Say what the organization will do, who owns each action and when employees can expect to see a change. Say as clearly what will not change and why; a reasoned no is better received than silence.

Then measure again. Ask about the same topic in the next pulse survey and compare the results with the previous wave. If a theme becomes less frequent or less negative after a change, you have evidence that the action worked. If not, the comments will usually tell you why.

Where Manual Analysis Stops Working

The process above works for a few hundred comments from one survey. With thousands of comments in several languages every quarter, codes drift as different people apply them, and comparing every theme across every department, location, tenure band and wave creates hundreds of combinations that no one can test by hand.

General chat assistants are good at summarizing a few dozen comments. Classifying thousands of records consistently, testing whether differences between groups are statistically significant, getting the same result from the same data every time and showing the source of every figure and quote require dedicated infrastructure. With employee feedback there is also the question of where the data goes, since pasting comments into a general tool may conflict with what employees were promised.

Analyzing Employee Feedback in Kimola

Kimola is an AI platform that turns customer feedback into enterprise-grade market research. The same analysis applies to employee feedback from engagement surveys, internal reviews and exit interviews, starting from the file your survey tool or HRIS exports.

Uploading the File

You upload the prepared file, an Excel or CSV export with column names in the first row, from Home. You choose the column that holds the open-ended answers and, optionally, a Date column such as the survey wave date. Every other column, such as department, location, tenure band or eNPS score, is kept and becomes a filter and a comparison group in the Report. The columns you keep in Step 2 are the groups Kimola can compare, and the personal data you remove never reaches the analysis. A Report analyzes one text column, so if a survey has several open-ended questions, create a separate Report for each important one.

What the Analysis Does

Themes from the data itself. Kimola finds themes without a predefined list of categories and names each one after what employees actually say, so "workload" can appear as specific themes such as overtime after the reorganization or unclear priorities. Each theme gets its own sentiment through aspect-based sentiment, so a comment that praises a manager and criticizes workload counts as positive on one and negative on the other.

Research questions asked of every comment. Dimensions ask the questions from Step 4 of every piece of feedback. Pain Points identify what goes wrong or creates difficulty, Unmet Needs surface what employees expect to be improved, added or changed, Underlying Causes explain the conditions that lead to a situation and Motivations show what employees value. Dimensions can be added when the Report is created or afterwards.

Differences that are statistically real. Kimola crosses every theme and Dimension with the columns in your file and with time, and tests each combination. Only statistically significant findings become Signals, for example that Pain Points about scheduling are mentioned far more often in one location than in the rest, or that a theme has become more frequent since the previous wave. If nothing meets the definition, the Report has no Signals, which means nothing stood out. Kimola's tests decide whether a difference is real, not whether a group is large enough to protect anonymity, so apply your minimum group size in the file before uploading.

Scores that account for volume. On the Performance screen, Satisfaction and Dissatisfaction Scores rank themes by how positive or negative employees are about them, taking the number of comments into account along with the share. A theme with 5 comments that are all positive does not outrank one with 50 comments of which 43 are positive.

Recommendations checked against the evidence. From the Actions screen, Kimola writes an Action Plan of up to 12 actions, each tested against the feedback, with its figures and quotes checked and a confidence level attached. For HR, this is a starting point for Step 5, not a substitute for deciding what to commit to.

Recurring Surveys and In-House Deployment

Two capabilities are available only in the Enterprise package. For recurring pulse surveys, a Space keeps the program in one place: each new survey file is classified with the same Funnel, the Space's set of custom models and Dimensions, so results stay comparable from wave to wave, and Signals run again after every new file.

For organizations whose policies require employee data to stay in-house, Kimola On-Premises runs on the organization's own infrastructure under a license. Feedback, analysis results and the AI models stay within that environment, and no data is sent to external AI services. It does not include Feeds or third-party integrations, so data is uploaded as a file.

What Kimola Does Not Do

Kimola does not design or send surveys, run interviews or manage an engagement program, and it does not make a dataset anonymous for you: removing personal data and applying a minimum group size remain the responsibility of the team that prepares the file. It does not predict who will leave. If a survey produces a few dozen comments, reading them carefully may be all the analysis you need; Kimola is most useful with hundreds or thousands of comments and several groups or waves to compare. Kimola has a free Starter plan with no credit card required, and paid plans start at $49 per month, as listed on the pricing page.

Frequently Asked Questions

What Is eNPS and How Is It Calculated?

eNPS asks employees how likely they are to recommend the organization as a place to work, from 0 to 10. It is the percentage of Promoters (9 or 10) minus the percentage of Detractors (0 to 6), so it ranges from −100 to 100. It is most useful tracked over time and paired with an open question about the reason for the score.

How Often Should You Run Pulse Surveys?

There is no single correct frequency. Monthly or quarterly pulse surveys alongside an annual engagement survey are common, and Bain notes that eNPS is most commonly asked monthly. The practical limit is follow-through: run surveys only as often as you can analyze the results, report back and act, because frequent surveys with no visible response lower participation.

Are Public Employer Reviews Useful for Internal Research?

Public reviews on employer review sites can add an outside view, particularly of how candidates and former employees perceive the organization. They are not a substitute for internal feedback: reviewers are self-selected, their role and tenure are often unknown and they cannot be connected to departments or survey waves. Use them as context, not as the basis for decisions about specific teams.

Conclusion

Analyzing Voice of Employee data well depends as much on trust as on technique. Anonymity, clear communication and visible follow-through are part of the analysis, not extras around it.

Most organizations already hold the data they need in survey exports, pulse results and exit interview notes. With personal data removed and the department, location, tenure and wave columns kept, that data can be uploaded to Kimola and turned into enterprise-grade market research about the workforce, without designing a new study. You can create a free account to try it with your own file.

Subscribe to Our Newsletter!

Stay up-to-date with the latest product features and company news.

Get Started for Free!

Turn customer feedback into structured market research in 30+ languages—no AI training needed.

Create a Free Account No credit card · No commitment