Running eNPS in a 15-person company without destroying trust

Why engagement surveys fail in small teams

Engagement surveys have a terrible reputation at small companies, and it's earned. The failure is always one of two things — usually both.

"Anonymous" is a fiction. At 15 people, everyone recognizes everyone's writing style, knows who was unhappy about the release weekend, and can guess which "department" filter contains exactly two people. The first time a founder reads a critical comment and asks around about who wrote it, the survey is dead — from then on you're collecting polite fiction.

Nothing changes afterwards. The team spends ten minutes being honest, the results disappear into a founder's drawer, and the next survey arrives three months later with the same questions. People aren't cynical about surveys — they're cynical about surveys that go nowhere. Response rates sag, the honest people stop answering first, and the score drifts up while the actual mood drifts down.

Both failures are process problems, not tooling problems. The fix is a set of rules you commit to before the first survey: anonymity rules that hold at small scale, and a follow-up ritual you actually run. eNPS happens to be a good vehicle because it's short, comparable over time, and hard to over-engineer.

eNPS in plain words

eNPS (employee Net Promoter Score) is one question — "How likely are you to recommend working here to a friend?" — answered 0–10, usually with a free-text "why?" underneath.

Scoring:

• 9–10 = promoters. Would genuinely recommend the place, and probably already do.
• 7–8 = passives. Fine, not enthusiastic. They count in the denominator but not in the score.
• 0–6 = detractors. Anything from mildly disappointed to actively interviewing.

Score = % promoters − % detractors, so it runs from −100 (everyone a detractor) to +100 (everyone a promoter).

Now the small-team arithmetic most advice ignores: with 15 respondents, each person is worth about 7 points of score. One engineer having a bad sprint and dropping from 8 to 6 moves the company number by 7 points; one person swinging from promoter to detractor moves it 13. That isn't noise in your culture — it's the resolution limit of the instrument. Which is why at this size the score is a trend to watch, not a number to hit, and why the free-text answers are the real payload: fifteen honest sentences about what would make people recommend the company are worth more than the score they arrived with.

The anonymity rules that hold at 15 people

Anonymity at small scale has to be engineered — and the rules stated to the team up front, because trust in the mechanism is the actual product here.

• A minimum-response threshold before anyone sees anything. If only a few people answered, showing results at all points a finger at them. The correct behavior is brutal: below the threshold, nobody — including the owner — sees a score, a comment, or a distribution. An "insufficient responses" screen is the feature working.
• No per-person drilldown, ever. Not for the owner, not for HR, not "just this once" after a worrying comment. The moment one answer gets traced to one person, every future answer is written for an audience.
• No slicing into tiny segments. A per-team breakdown where a "team" is two people is a de-anonymization tool with a dashboard. At 15 people, read results at company level, full stop.
• Themes, not verbatims. Reading raw quotes aloud in a retro invites style-matching. Summaries — "workload came up repeatedly", "three comments mention unclear ownership" — carry the signal without pointing at an author.
• Don't over-survey. Monthly or quarterly at this size. A weekly pulse in a 15-person company is exhausting, and with a sample this small, weekly noise impersonates trend. Set the cadence per survey and announce it up front.

What to do with the results — the part that decides everything

The survey earns or loses the team's trust in the week after it closes. A simple ritual, every cycle:

1. Share results with the whole team within a week. The score, the response rate, and the comment themes — as summaries, never quotes. Withholding results signals the data is being used on people rather than for them.
2. Pick one or two things, not five. Choose the theme you can visibly move before the next survey. A small fixed thing beats an ambitious ignored list.
3. Say what you're NOT going to act on, and why. "Several comments mention compensation; the review is planned for January, not before" builds more trust than silence. People can live with a no; they can't live with a void.
4. Close the loop next cycle. Open the next survey's results with "last time you said X, we did Y". That sentence is the entire reason response rates hold.

And the rule above the ritual: the score is a thermometer, not a KPI. The moment eNPS lands in someone's bonus target, it stops measuring engagement and starts measuring the ability to campaign for nines. Watch the trend, watch the response rate — a falling one is itself a result: people stopped believing the exercise — and treat any sharp drop as a conversation starter, not a scoreboard.

How Helia HR does this

The Engagement pack ($2/employee/mo — see pricing) builds the rules above into the mechanism, so anonymity is a property of the system rather than a promise:

• Anonymous pulse surveys with the eNPS question plus free text — designed so answers are never attributed to a person.
• A minimum-response threshold: until enough people have answered, no results are shown to anyone — no score, no comments. In a small team this is the load-bearing feature.
• eNPS scoring and trend over time, so you manage direction across cycles instead of panicking about one jumpy number.
• AI theming of comments — free-text answers are summarized into topics without exposing individual answers; you read "recurring theme: project switching", not recognizable prose.
• Response-rate tracking, because participation is the health metric of the survey itself.
• Per-survey frequency settings — pick the cadence survey by survey and keep it predictable.

The survey tells you something is off. Finding out what — with a name attached and consent given — is a job for a 1-on-1.

FAQ

What's a good eNPS for a 15-person company?

Resist benchmarking. Published averages come from companies large enough that one person is a rounding error; at 15 people one person is 7 points. A useful reading instead: negative and falling — act now; positive with a high response rate and honest comments — healthy. Compare yourself to your own last three surveys, not to an industry table.

How often should we run it?

Quarterly is the safe default at this size; monthly if you're actively fixing something and want to know whether it's landing. The real constraint: never open the next survey before you've closed the loop on the last one.

Can the founder see who wrote what?

In Helia — no, and that's the point. Results are anonymous, gated by the minimum-response threshold, and comments surface as summarized themes rather than attributable prose. If a specific person is clearly struggling, the tool for that is a 1-on-1, not survey forensics.

What if only four people respond?

Then nobody sees results — the threshold doing its job. Treat low participation as the finding itself: the team doesn't yet believe the survey is safe or useful. Fix that first, usually by running the follow-up ritual visibly, and the response rate recovers before the score does.