How to Analyse Employee Engagement Survey Results and Act on Them
The short answer
To analyse engagement survey results, read response rates first, then the spread of scores by group before the headline, then each item's favourable share with its margin of error and trend. Use driver analysis to find low scores that move with engagement, test them in focus groups, check turnover by group, then agree two or three priorities and tell everyone what you heard and will do within six weeks.
- Favourable % is a share of answers, not of people
- Read the spread by group before the headline, and never report groups under five
- Treat changes smaller than about 1.4 times the margin of error as no clear change
- Correlation shows where to look, not what causes what
- Measure turnover on average headcount, and look at regretted and first-year shares
- Send 'you said, we did' within six weeks
What do engagement survey scores actually mean?
Most engagement surveys use a five-point agree scale, from 1 strongly disagree to 5 strongly agree. The favourable score for an item is the share of answers that were 4 or 5. Neutral is 3, unfavourable is 1 or 2. Decide these definitions before anyone sees results and keep them year to year, or your trend means nothing.
An engagement index is usually the favourable share across a small set of outcome items, such as pride in the organisation, willingness to recommend it as a place to work, intent to stay and going beyond what you would elsewhere. An index of 60% means 60% of answers to those items were favourable. It does not mean 60% of people are engaged.
Be careful with comparisons to published figures. Gallup's State of the Global Workplace 2026 report puts global employee engagement at 20% in 2025. That figure comes from Gallup's own items and method, so it cannot be set against your favourable percentage. Your own trend on unchanged items is the most reliable comparison you have.
| Measure | How it is calculated | Watch out for |
|---|---|---|
| Response rate | Responses / people invited | Low response in one group may itself be the finding |
| Favourable % | Answers of 4 or 5 / all answers to the item | Not a share of people |
| Engagement index | Favourable answers across the outcome items / all answers to them | Not comparable with other providers' measures |
| Margin of error | About 1.96 x square root of p(1 - p) / n, with a finite population correction | Small groups swing a lot |
| Correlation (r) | How an item moves with each person's engagement score | A clue, not a cause |
| Voluntary turnover | Voluntary leavers / average headcount | Year-end headcount distorts growing and shrinking teams |
In what order should you read the results?
The order matters because the first number people see becomes the story. Read response first, by site, function and other groups. Then read the spread: which groups sit well below the organisation? Only then read the headline index and the items, with last time's figure beside each one. Comments come last, read alongside the scores, because a theme with many comments is loud but not necessarily important.
Treat any group below your minimum group size as unknown, not as low or high. Five is a common floor. Apply it to cross-cuts such as function by site as well, and never show figures that would let someone work out a hidden group by subtraction.
- Response rates by group
- Spread: groups well below the organisation
- Suppressed groups named as unknown
- Items with margin of error and trend
- Driver analysis
- Comment themes beside the scores
How big does a change need to be before it is real?
Every survey result is an estimate, and the margin of error tells you roughly how far it could move by chance. For a whole-company item with 296 answers from 400 people, the margin is about plus or minus 2.8 points once you correct for the size of the organisation. When you compare two surveys, a change smaller than about 1.4 times the margin is not clear evidence of change, so say 'about the same'.
Small groups swing much more. In a team of eight, one person changing their answer moves a score by about 12 points. That is why ranking small teams on scores is meaningless, and why you should use words like 'about' and look for a trend before reacting to a single result.
What is driver analysis, and how should you use it?
Driver analysis asks which items move together with engagement in your data. For each person, take the average of their answers to the engagement outcome items. Then calculate the correlation between each driver item and that score. A rough reading used in survey work is that about 0.1 is weak, about 0.3 moderate and 0.5 or more strong.
Plot each driver item by favourable % across and correlation up, and split both at the median. The idea adapts importance-performance analysis, which John Martilla and John James described in the Journal of Marketing in 1977. You get four quadrants: Priority (low score, strong link), Maintain strengths (high score, strong link), Monitor (low score, weak link) and Lower priority (high score, weak link).
Two cautions matter. Correlation is not cause: engaged people may answer everything more positively, so test Priority items in focus groups before spending money. And a low score on speaking up, respect or safety needs action whatever its correlation says.
How does turnover fit into the picture?
A survey tells you how people feel; leaver data tells you what they did. Calculate voluntary turnover as voluntary leavers in the period divided by average headcount over the period. Then add two shares of voluntary leavers: regretted (people you would have wanted to keep) and first-year (people who left within a year of joining). A high first-year share points at recruitment and onboarding; a high regretted share in one group points at a retention problem worth a plan of its own.
Decide once whether retirements count as voluntary and keep the rule. The US Bureau of Labor Statistics, in its job openings and labour turnover survey, counts retirements separately from quits, and many organisations do the same. Read exit interview reasons together with their coverage: if only half of leavers gave an interview, treat the reasons as a sample.
If you estimate the cost of turnover, build it from visible parts (recruitment, hiring time, onboarding, lost productivity while someone new ramps up, and the vacancy) and label it as an estimate every time you quote it.
Reading helps; measuring tells you what to work on. These AI-graded assessments on AssessAll pair with this topic:
How do you turn results into action in six weeks?
The work that matters most happens after the analysis. A simple sequence works in organisations of most sizes: leaders commit to two or three organisation-wide priorities with owners; managers receive results for their own group if five or more people answered; each manager holds a 60-minute team conversation; each team picks one to three actions with an owner, a date and a measure; and everyone hears 'you said, we did' within six weeks of the survey closing. Track actions monthly and repeat the priority items word for word in a pulse.
Support managers rather than ranking them. A team's score describes its experience in one period. Tie managers' pay to it and you teach the organisation to manage the number instead of the work. Say what you will not do this year, and why: it shows people you listened even when the answer was no.
What should the leadership results briefing include?
Hold it within two weeks of the survey closing and give it 90 minutes. Open with the rules for reading results: spread before headline, suppressed groups are unknown, and changes inside the margin are 'about the same'. Then show the response rate by group, the headline index with its trend and margin, the groups well below the organisation, the priority matrix, the top comment themes and turnover by group.
End with decisions rather than discussion: two or three organisation-wide priorities, an owner and a measure for each, and an honest list of what leaders will not do this year and why. Agree the dates for manager packs, team conversations and 'you said, we did' before anyone leaves the room. If the meeting produces only a slide, the six weeks after it will produce nothing.
- Rules for reading results
- Response and spread by group
- Headline with trend and margin
- Priority matrix and comment themes
- Turnover by group, with costs labelled as estimates
- Two or three priorities, owners, measures and what you will not do
How do stay conversations help retention?
Exit interviews tell you why people left. Stay conversations ask why people stay and what might make them leave, while there is still time to act. A manager holds a private, planned conversation of about 30 minutes with each person in the team, separate from performance reviews, built around a few open questions: what keeps you here, what are you learning, what would make your job better, what might tempt you to leave, and how can I support you.
Offer them to everyone rather than only the people you think might leave: targeting signals suspicion and is often wrong. The manager agrees one or two follow-ups and comes back within two weeks, because a conversation with no follow-up does more harm than none. Pay questions go through the normal pay process rather than an on-the-spot counter-offer, and HR receives themes without names, which feed a retention plan built around roles and groups, never a list of individuals.
How do you protect confidentiality and use AI safely?
Say exactly what you promise. An anonymous survey collects no identifiers; a confidential one keeps a link that only the provider holds. Publish who sees what, the small-group rule, how comments are read and how a comment that discloses harm will be handled. Regulators such as the UK Information Commissioner's Office publish guidance on employee data and monitoring, and in some countries works councils must agree to staff questionnaires, so check local rules and consult representatives where they exist.
AI assistants can draft items, remove identifying details from comments and group anonymised comments into themes. Use only tools your organisation has approved, paste only aggregated or anonymised data, never ask a tool to identify a writer or score individuals for 'flight risk', and let people make every decision. If you want to test your own judgement with people data, the AssessAll People & Performance Data Judgment assessment is one option, and a Jobulary individual development plan can turn the gaps into a plan.
The Engagement and Retention Kit from Bodhih Training puts this whole method into working files: a workbook that suppresses small groups and runs driver and turnover analysis, a turnover calculator, a listening planner, scripts and templates for every step.
Make your next survey lead somewhere
The Engagement and Retention Kit gives you the method, the workbooks, the forms, the scripts and the templates for the whole listening year. Open the Start Here map and your first hour is planned.
More from the Bodhih family
Questions people ask next
What is a good response rate for an engagement survey?
There is no single right number. Compare with your own previous surveys and look at the spread between groups: a group with much lower response than others may doubt that answering is safe or useful, and that is a finding in itself.
What does 'favourable' mean in survey results?
On a five-point agree scale, favourable usually means an answer of 4 (agree) or 5 (strongly agree). The favourable percentage is the share of answers to an item that were 4 or 5.
Why should groups under five be hidden?
Small groups make it possible to work out who said what, and their scores swing so much that they are hard to read. Hiding them protects people and keeps answers honest next time.
Can I run driver analysis for each team?
Only for groups with enough responses, often 30 or more. Below that, correlations jump around. For small teams, use favourable scores and a team conversation instead.
Should benchmarks drive our priorities?
Use them with care. External benchmarks often use different items, scales and samples. Your own trend on unchanged items, and the spread between your own groups, are usually more useful.
What is the difference between regretted turnover and voluntary turnover?
Voluntary turnover counts everyone who chose to leave. Regretted turnover counts only the voluntary leavers you would have wanted to keep. Agree the definition of regretted once and decide it at the time someone leaves.
How long should an engagement survey be?
A full census of around 25 to 40 items takes about ten to fifteen minutes. A pulse of five to ten items takes about three. Every extra item costs time and goodwill.
What if we cannot act on what people said?
Say so, with the reason, and offer a smaller step if you can. An honest 'not this year, because' does far less damage than silence.