Engagement Survey

You find out how people really feel about working here, protect their anonymity while doing it, and turn the results into two or three actions that leadership visibly takes. A survey nobody acts on is worse than none: it teaches people that speaking up is pointless. So the standard is a short, well-designed survey; honest, statistically sensible analysis that does not over-read small numbers; and a plan with owners and dates.

When to use

• "Run an engagement survey for the team."
• "Set up a quarterly pulse."
• "Analyse these survey results and tell me what matters."
• "What's our eNPS?"
• "People seem unhappy, how do we find out why?"

What you need

• The purpose: a baseline, a check after a change, or tracking over time. Who will see the results and who commits to act.
• Headcount and the groups you might report by (team, location, tenure band). Each reported group must meet the minimum group size.
• Previous survey items and results, if any, so you can compare like with like.
• A survey tool the company already uses (for example its forms tool or HR system). Access through a connected app or the agent's own browser signed in by the owner. Do not sign up for a paid survey service. If nothing is available, ask with clarify.
• The owner's privacy notice for employee data.

Method

1. Set the anonymity rules first. Choose a minimum reporting group (5 is a common minimum; use a higher number for sensitive topics or small companies) and never show results, including comments, for any group below it.
   • Decide whether the survey is anonymous (no identity collected) or confidential (identity held by HR but never reported).
   • Tell employees which, honestly.
   • Collect the fewest demographic fields possible: every extra field makes people identifiable in small teams.
2. Choose the items. Use or adapt the item set in references/survey-items.md: an engagement index of four items (pride, recommend, intent to stay, motivation), driver items across clarity, manager, growth, recognition, workload, belonging, voice, confidence in leadership, and tools; one eNPS item; two open questions. 20 to 30 items for an annual survey, 5 to 10 for a pulse.
   • Use a five-point agreement scale.
   • Do not copy proprietary survey items from commercial vendors.
3. Write the invitation (draft for the owner). Why we are asking, how long it takes (aim for under 10 minutes), the anonymity rule in plain words, the closing date, and when results and actions will be shared. Plan one reminder midway and one the day before closing.
4. Run it. Open for 7 to 10 working days.
   • Track the response rate daily by group without looking at answers.
   • As a working rule, treat any result below about a 50 per cent response rate as directional only, and say so.
5. Analyse. With execute_code and the code in references/scoring-code.md:
   • Favourability per item: per cent agree plus strongly agree; also show neutral and unfavourable.
   • Engagement index: mean favourability of the four engagement items.
   • eNPS: per cent scoring 9 to 10 minus per cent scoring 0 to 6, on the 0 to 10 item. Range minus 100 to plus 100.
   • Driver analysis: correlate each driver item with the engagement index across respondents (Spearman). Plot importance (correlation) against performance (favourability). High importance with low favourability is where to act.
   • Change since last time: only for identical items. With small groups, differences under about 10 points are usually noise; say so rather than celebrate or panic.
6. Code the comments. Read every comment.
   • Assign each to one or more themes, count them, and pick a few representative quotes that cannot identify the writer (remove names, specific events, distinctive phrases).
   • Report themes for groups at or above the minimum size only.
7. Report. A two-page summary: response rate, engagement index and eNPS with context, top three strengths, top three priority areas (from the driver analysis and comments), and what you are not concluding because numbers are too small. Use show_card for the headline numbers.
8. Plan actions with leadership. Pick two or three focus areas, not ten.
   • For each: what we heard, what we will do, owner, date, how we will know it worked.
   • Teams can choose one local action of their own.
9. Close the loop. Draft a "you said, we did" message for the owner to send within about three weeks of closing. Schedule a short pulse in three to six months on the focus areas with cronjob.

Judgement calls

• Under about 15 people: anonymity is fragile. Collect no demographic fields, and consider confidential conversations run by someone neutral alongside or instead of a survey.
• Leaders ask for raw comments: share coded themes and safe quotes only, within the minimum group rule. The anonymity promise outranks curiosity.
• A comment reports harassment, a safety risk or illegal conduct: you cannot follow up with an anonymous writer, so raise the issue with HR, who can act on it (for example by reminding everyone of reporting routes, or investigating if it is specific enough).
• Results are worse than leaders expected: present them plainly. Do not re-run the survey hoping for better numbers.
• Pulses running faster than visible action: slow the cadence down.

Output

• survey-<period>/questionnaire.md and the invitation and reminder drafts.
• survey-<period>/results.md: headline numbers, item table, driver chart, themes, caveats.
• survey-<period>/actions.md: focus areas with owners and dates, and the you-said-we-did draft.

Headline card layout:

<Survey name>: <n> of <headcount> responded (<rate>%)
Engagement favourable <x>% (last time <y>%, same items) · eNPS <z>
Strengths: <item>, <item>
Priority areas: <item>, <item>, <item>
Caveats: <small groups, low response, items changed>

Checks before you finish

• No result or comment is shown for any group below the minimum size.
• Comments quoted cannot identify anyone.
• Response rate is reported with every result, and small-sample caveats are stated.
• Trend comparisons use identical items only.
• There are at most three focus areas, each with an owner and a date.
• Nothing has been sent to employees without the owner's approval.

Pitfalls

• Breaking anonymity by slicing. "Women in the Lisbon sales team" may be one person. Respect the minimum group size, including when combining filters.
• Hunting for the commenter. Never try to work out who wrote what, and refuse if asked.
• Over-reading small changes. A 4-point drop in a team of 12 means nothing. Say so.
• Survey and silence. Without a visible response, participation and trust fall next time.
• Too many actions. Ten initiatives become zero. Choose two or three.
• Using results against people. Survey data must never feed individual performance or pay decisions.

See also: team-rituals-and-recognition for actions on belonging and recognition, and one-to-one-meetings for manager-level follow-up.