Employer brand research: staff survey methodology
Why a hypothesis is needed, not just a questionnaire
The first mistake of most surveys is launching them without a question they should answer. A proper methodology starts with a hypothesis: for example, 'employee willingness to recommend the company (eNPS) is related to the actual number of referral responses' or 'the gap between internal perception and external reviews on Glassdoor indicates a problem in onboarding.' A hypothesis defines which metrics to collect and how to interpret them later.
It is important to distinguish two layers of data that are often confused. Perception — the overall perception of the company as an employer — changes slowly, and if the reputation is already damaged, it 'sticks' for a long time even after real improvements. Drivers — specific factors such as salary, career growth, quality of leadership, and workload — move faster and respond to changes almost immediately. If you measure only perception, you will see that 'it is bad,' but you will not understand what exactly needs to be fixed.
What sample size is needed for the results to be believable
Sample size is not an abstract number from a statistics textbook, but a specific percentage of responses below which the data is considered unrepresentative.
| Type of survey | Company size | Acceptable % of responses |
|---|---|---|
| Census survey (full coverage) | Large company | 72–88% |
| Pulse survey (short, frequent) | Large company | 55–81% |
| Any type | 5000+ employees | 65–80% |
A value below the lower threshold is a signal that the sample is skewed: most likely, the most loyal or most dissatisfied responded, while the 'silent majority' remained silent. This is exactly the response bias mentioned below.
Even with a high response rate, the sample can still be skewed in composition — for example, mostly women from one region with 5+ years of experience responded. To correct this, the data is weighted (weighting) by key characteristics: gender, age, seniority, area. After weighting, an efficiency indicator is calculated — a measure from 0 to 100, reflecting how well the sample matches the actual personnel structure. For instance, in a study of a Portuguese company with 16 801 respondents, the efficiency indicator was 83,6% — this is considered a good level of representativeness. Aim for 80%+ as a working minimum.
How to structure a questionnaire to avoid just getting a 'mean score'
A questionnaire with 40 questions will almost certainly yield a low response rate and respondent fatigue. The optimum is 10–15 questions, unless the goal is a deep academic study, but rather a working tool for decision-making. For quantitative data, use one scale throughout the entire questionnaire, do not mix formats: either Likert 1–5, or NPS 0–10.
Essential blocks without which an HR brand assessment is incomplete:
- Willingness to recommend the company as an employer (eNPS) — the basic loyalty metric
- Assessment of key drivers: salary and benefits, work-life balance, job stability, team atmosphere, fairness (equity) — these are consistently top 5 factors in large-scale studies such as Randstad Employer Brand Research
- Evaluating the direct manager separately from the company's evaluation — otherwise, a poor manager 'taints' the brand of the entire organization in the respondent's eyes
- An open-ended question without predetermined answers is the only source of quality data that does not fit into a scale
- A demographic block for subsequent weighting (not for de-anonymization)
For scale comparison: Randstad surveys more than 170 000 respondents in 6400+ companies, the interview takes 14 minutes — this is a benchmark at the upper limit of the acceptable survey length.
Internal and external HR brands are measured differently
An employee survey and candidate perception research are two different methodologies, not one questionnaire with a different audience.
| Parameter | Internal HR brand (employees) | External HR brand (candidates, market) |
|---|---|---|
| Data source | Direct staff survey, census/pulse | At scale, applicant surveys, labor markets |
| Key metric | eNPS, Employee Engagement Index | Employer Brand Index, recognition and attractiveness |
| What it measures | Real work experience inside | Expectations and reputation before contacting the company |
| Additional sources | Exit interviews, stay interviews | Glassdoor, review sites, social networks, NLP analysis of mentions |
The gap between these two pictures is the most valuable finding of the study. If employees inside rate the company 8 out of 10, while external reviews on Glassdoor lean towards 3, this is not a contradiction, but a diagnosis: the problem is either in the dismissal process (resentful former employees write reviews), or in the gap between real experience and promises during hiring.
How not to get distorted answers
Response bias is the main enemy of any internal survey, and fighting it must be done at the procedural level, not post-factum.
- Guarantee technical anonymity: survey through a third-party service, not corporate email, without linking to an account
- Do not collect detailed demographics in small teams — if in a team of 4 people you ask for gender, age and experience, anonymity practically disappears
- Separate the role of HR as the research requester and HR as the recipient of open comments — some employees will not answer honestly if they know their manager will read the comments
- Publish aggregated results to everyone, not just top management — this reduces the feeling that the survey is just for show, and increases sincerity next time
- Cross-check self-reported data with behavioral: if eNPS is high, but actual referral rate is low — somewhere people answered socially desirably
How often to research and what to do with the numbers next
A full census survey makes sense to conduct 1–2 times per year, at the same time each time — this allows comparing dynamics without seasonal distortions. Short pulse surveys on 3–5 questions are appropriate between major waves, which capture sharp changes faster.
Interpretation of eNPS is not binary: a result above 50 is considered excellent, the range 10–30 is average and acceptable, and a negative value requires urgent intervention, not 'monitoring next quarter'. But eNPS by itself, without breakdown by drivers, is a useless number: it says 'something is wrong', but does not say what to fix.
The true value of the methodology lies in mixed methods (mixed methods): quantitative surveys from 10–15 (or up to 55 for in-depth research) questions are supplemented with analysis of open-ended textual data — employee comments, reviews on Glassdoor — through NLP. Such comparison reveals inconsistencies that are not visible in average scores: for example, the 'salary' driver is rated as average, but in textual comments, salary is mentioned in 70% of negative reviews — meaning the scale underestimates the true urgency of the problem.
And the final, most frequent failure of the methodology — the results are collected, but not tied to the owners of the solutions. Every identified driver with a low rating must be assigned a responsible person, a specific action, and a deadline that is verified in the next wave of the survey. Without this feedback loop, any, even perfectly constructed, research will remain a beautiful presentation that changes nothing.
Why a hypothesis is needed, not just a questionnaire
The first mistake of most surveys is launching them without a question they should answer. A proper methodology starts with a hypothesis: for example, 'employee willingness to recommend the company (eNPS) is related to the actual number of referral responses' or 'the gap between internal perception and external reviews on Glassdoor indicates a problem in onboarding.' A hypothesis defines which metrics to collect and how to interpret them later.
It is important to distinguish two layers of data that are often confused. Perception — the overall perception of the company as an employer — changes slowly, and if the reputation is already damaged, it 'sticks' for a long time even after real improvements. Drivers — specific factors such as salary, career growth, quality of leadership, and workload — move faster and respond to changes almost immediately. If you measure only perception, you will see that 'it is bad,' but you will not understand what exactly needs to be fixed.
What sample size is needed for the results to be believable
Sample size is not an abstract number from a statistics textbook, but a specific percentage of responses below which the data is considered unrepresentative.
| Type of survey | Company size | Acceptable % of responses |
|---|---|---|
| Census survey (full coverage) | Large company | 72–88% |
| Pulse survey (short, frequent) | Large company | 55–81% |
| Any type | 5000+ employees | 65–80% |
A value below the lower threshold is a signal that the sample is skewed: most likely, the most loyal or most dissatisfied responded, while the 'silent majority' remained silent. This is exactly the response bias mentioned below.
Even with a high response rate, the sample can still be skewed in composition — for example, mostly women from one region with 5+ years of experience responded. To correct this, the data is weighted (weighting) by key characteristics: gender, age, seniority, area. After weighting, an efficiency indicator is calculated — a measure from 0 to 100, reflecting how well the sample matches the actual personnel structure. For instance, in a study of a Portuguese company with 16 801 respondents, the efficiency indicator was 83,6% — this is considered a good level of representativeness. Aim for 80%+ as a working minimum.
How to structure a questionnaire to avoid just getting a 'mean score'
A questionnaire with 40 questions will almost certainly yield a low response rate and respondent fatigue. The optimum is 10–15 questions, unless the goal is a deep academic study, but rather a working tool for decision-making. For quantitative data, use one scale throughout the entire questionnaire, do not mix formats: either Likert 1–5, or NPS 0–10.
Essential blocks without which an HR brand assessment is incomplete:
- Willingness to recommend the company as an employer (eNPS) — the basic loyalty metric
- Assessment of key drivers: salary and benefits, work-life balance, job stability, team atmosphere, fairness (equity) — these are consistently top 5 factors in large-scale studies such as Randstad Employer Brand Research
- Evaluating the direct manager separately from the company's evaluation — otherwise, a poor manager 'taints' the brand of the entire organization in the respondent's eyes
- An open-ended question without predetermined answers is the only source of quality data that does not fit into a scale
- A demographic block for subsequent weighting (not for de-anonymization)
For scale comparison: Randstad surveys more than 170 000 respondents in 6400+ companies, the interview takes 14 minutes — this is a benchmark at the upper limit of the acceptable survey length.
Internal and external HR brands are measured differently
An employee survey and candidate perception research are two different methodologies, not one questionnaire with a different audience.
| Parameter | Internal HR brand (employees) | External HR brand (candidates, market) |
|---|---|---|
| Data source | Direct staff survey, census/pulse | At scale, applicant surveys, labor markets |
| Key metric | eNPS, Employee Engagement Index | Employer Brand Index, recognition and attractiveness |
| What it measures | Real work experience inside | Expectations and reputation before contacting the company |
| Additional sources | Exit interviews, stay interviews | Glassdoor, review sites, social networks, NLP analysis of mentions |
The gap between these two pictures is the most valuable finding of the study. If employees inside rate the company 8 out of 10, while external reviews on Glassdoor lean towards 3, this is not a contradiction, but a diagnosis: the problem is either in the dismissal process (resentful former employees write reviews), or in the gap between real experience and promises during hiring.
How not to get distorted answers
Response bias is the main enemy of any internal survey, and fighting it must be done at the procedural level, not post-factum.
- Guarantee technical anonymity: survey through a third-party service, not corporate email, without linking to an account
- Do not collect detailed demographics in small teams — if in a team of 4 people you ask for gender, age and experience, anonymity practically disappears
- Separate the role of HR as the research requester and HR as the recipient of open comments — some employees will not answer honestly if they know their manager will read the comments
- Publish aggregated results to everyone, not just top management — this reduces the feeling that the survey is just for show, and increases sincerity next time
- Cross-check self-reported data with behavioral: if eNPS is high, but actual referral rate is low — somewhere people answered socially desirably
How often to research and what to do with the numbers next
A full census survey makes sense to conduct 1–2 times per year, at the same time each time — this allows comparing dynamics without seasonal distortions. Short pulse surveys on 3–5 questions are appropriate between major waves, which capture sharp changes faster.
Interpretation of eNPS is not binary: a result above 50 is considered excellent, the range 10–30 is average and acceptable, and a negative value requires urgent intervention, not 'monitoring next quarter'. But eNPS by itself, without breakdown by drivers, is a useless number: it says 'something is wrong', but does not say what to fix.
The true value of the methodology lies in mixed methods (mixed methods): quantitative surveys from 10–15 (or up to 55 for in-depth research) questions are supplemented with analysis of open-ended textual data — employee comments, reviews on Glassdoor — through NLP. Such comparison reveals inconsistencies that are not visible in average scores: for example, the 'salary' driver is rated as average, but in textual comments, salary is mentioned in 70% of negative reviews — meaning the scale underestimates the true urgency of the problem.
And the final, most frequent failure of the methodology — the results are collected, but not tied to the owners of the solutions. Every identified driver with a low rating must be assigned a responsible person, a specific action, and a deadline that is verified in the next wave of the survey. Without this feedback loop, any, even perfectly constructed, research will remain a beautiful presentation that changes nothing.
Recent articles
MegaFon, BMW, VTB and more 240+ companies. See how it looks in the showcase reel:
Our branding agency is approached to order:
Branding agency services
