Insight
Most companies learn how people feel from the people who resign.
Which is the worst possible sample, taken at the worst possible moment, from someone who has already decided and has nothing to gain by being precise.
A survey is the instrument for that question. An exit interview is a post-mortem.
What comes back
A distribution per question, not a mood score.
Questions can be a rating, a recommendation score, a single choice or several. When a survey closes, each question returns its average and its distribution — because an average of 3.8 with a long tail and an average of 3.8 clustered in the middle are different companies.
A single organisational happiness number is the thing everybody asks for and nobody can act on.
- Would you recommend working here?7.4Score 0–10
- My manager gives me useful feedback3.8Rating 1–5
- I have what I need to do my job4.1Rating 1–5
Who sees it, and who answers
Anonymous by permission, targeted by audience.
A survey is aimed at an audience rather than blasted at everyone, and it can allow anonymous responses — set per survey, not per person, so nobody has to decide in the moment whether their name is attached.
It moves from draft to published to closed. Responses arrive through the employee portal, where people already go for payslips and leave, which is the difference between a survey people answer and one that sits in an email.
- Anonymity is a property of the survey —
Decided when you design it, visible to whoever answers. A survey that is anonymous only if you tick something is a survey nobody trusts.
- Closed means closed —
Publishing and closing are explicit, so a response set has a boundary and a comparison next quarter means something.
- Participation is part of the result —
Audience size, how many answered, how many of those were anonymous, and the response rate — because a four-point average from nine per cent of a department is not a finding.
- Free text is counted, not displayed —
The summary says how many people answered in words and shows none of them, so an aggregate can never be read back to a person.
- It lives where employees already are —
In the same portal as payslips and leave, not a separate tool with its own login and its own reminder emails.
Being exact
It aggregates answers. It does not interpret prose.
Numeric and choice questions are aggregated. Free-text answers are stored and readable, and nothing scores them for tone — there is no sentiment model reading what your employees wrote about their manager.
That is deliberate. Sentiment scoring over free text is the feature most likely to be quoted confidently and be wrong, about the most sensitive material an HR team holds.
What an HR head asks
Mostly about trust in the instrument.
›Can people answer anonymously?
Yes, where the survey allows it — and that is set when the survey is designed, so whoever answers knows before they start.
›Do we get a single engagement score?
Per question, yes — a recommendation question returns its own headline score. What you do not get is one number composited across questions, because that is the number nobody can act on and everybody quotes.
›Can we break anonymous results down by department?
Yes — the answers carry the demographics without the identity. And any group too small to hide in is dropped from the breakdown rather than shown, with the report telling you how many slices were withheld.
›Does it analyse written comments?
No. They are stored and readable. Nothing scores them for tone.
›Can we survey one department only?
Yes — a survey has an audience rather than going to everybody.
›Can we compare this quarter to last?
Each survey is its own closed set with its own responses. Running the same questions again gives you a comparable set; there is no automatic trend line drawn across surveys for you.
- Publish a two-question survey to one department
- Answer it anonymously from the portal
- Close it and open the distribution, not just the average
- Find a question where the average hides the tail