Analyze Post-Event Survey Results

Codes open-ended survey comments into themes, calculates NPS by segment, ranks sessions and speakers, and lists the five fixes that matter most for next time.

Example prompts

Once the skill is added, start a chat with one of these and attach your file if you have one.

  • Here are 640 survey responses from our user conference. What are the top themes from promoters and detractors?

  • Rank our 48 sessions by rating and comments and tell me which speakers to invite back.

  • Write the feedback section of our post-event report for the exec team, with 5 quotes.

What this skill does

This skill reads a raw survey export, removes duplicates and test rows, and calculates NPS and rating summaries overall and by segment, such as customers vs prospects or session track. It shows the number of responses next to every score and flags small groups so nobody overreads them.

Then it reads every open-ended comment, codes it into themes from a starter codebook plus anything specific to your event, and ties each theme to promoters, passives and detractors. You get a session and speaker ranking, short anonymized quotes, and a ranked list of fixes with an owner and a deadline.

When to use it

  • "What did attendees think of the conference?"
  • Ranking sessions and deciding which speakers to invite back
  • Writing the feedback section of a post-event or board report
  • Calculating event NPS by attendee type
  • Webinar or field event feedback with lots of free-text answers

How it works

  1. 1IntakeConfirms which columns hold the recommend score, ratings, sessions and segments, and who to leave out, in one question.
  2. 2Clean and scoreDedupes submissions, parses messy scores like '10 - definitely', and calculates NPS and top-2-box ratings by segment, with an optional Python script.
  3. 3Code commentsReads every comment and codes it into themes with sentiment, counting mentions by promoters, passives and detractors.
  4. 4Rank sessionsRanks sessions and speakers by rating and explains low scores with what commenters said.
  5. 5RecommendTurns the biggest detractor themes into five concrete fixes with an owner and a deadline, then validates every count.

What you'll need

  • Survey export (CSV, XLSX or pasted table)
  • Which columns hold the recommend score, ratings, sessions and segments
  • Optional: attendee count, session list with speakers, last year's results

What you get

A survey results document with a headline summary, NPS and ratings by segment, themes people loved and themes to fix with counts and quotes, a session and speaker ranking with invite-back calls, the top five fixes and what to keep, plus a CSV of every coded comment. An executive one-page version is available on request.

Tips

  • Ask the recommend question on a 0 to 10 scale and keep the wording the same every year so NPS is comparable.
  • Include an attendee type or customer vs prospect column; the segment split is often more useful than the overall score.
  • Report sponsor and exhibitor feedback separately from attendee feedback.
  • Share speaker-level criticism with the event team before it goes into an exec summary.

What's in the download

  • post-event-survey-analysis/SKILL.md7.3 KB
  • post-event-survey-analysis/assets/sample-survey-responses.csv5.8 KB
  • post-event-survey-analysis/references/examples.md3.2 KB
  • post-event-survey-analysis/references/theme-codebook.md2.5 KB
  • post-event-survey-analysis/scripts/survey_summary.py5.8 KB
Read the SKILL.md
---
name: post-event-survey-analysis
description: Analyzes post-event and post-webinar survey exports. Calculates NPS and rating summaries by segment (attendee type, customer vs prospect, track), codes open-ended comments into themes with counts and example quotes, ties detractor themes to specific fixes, and ranks sessions and speakers. Use for attendee feedback analysis, conference survey results, session ratings, speaker rankings, event NPS reports or the feedback section of a post-event report.
license: MIT
metadata:
  author: eventmarketingsoftware.com
  version: "1.0"
  category: measurement
  source: https://eventmarketingsoftware.com/skills/post-event-survey-analysis
---

# Post-Event Survey Analysis

Turn a raw survey export into findings a team can act on: NPS and ratings by segment, open-ended comments coded into themes with counts and quotes, a session and speaker ranking, and the five fixes that matter most for next time.

## Intake

If the user has an `event-context.md` file, read it first (event names, segments, house tone).

Then ask for whatever is missing in ONE message:

1. The survey export (CSV, XLSX or pasted table) and the event name and format (conference, user conference, field event, webinar, trade show booth).
2. Which columns hold the 0 to 10 recommend score, ratings, session or speaker, and segment (attendee type, customer vs prospect, track, region). Offer your best guess from the headers so they can just confirm.
3. Who to leave out (employees, test rows, speakers rating their own session) and whether sponsors and exhibitors answered a separate survey.
4. Who the output is for: the event team (full detail) or executives (one page, five quotes).
5. Optional: total attendee count (for response rate), last year's NPS or ratings, the session list with speakers.

## Connectors (all optional)

- If a Google Sheets connector is available, read the responses there and offer to write the summary tables to a new tab (preview first).
- If a Slack connector is available, offer a short summary post as a draft; post only after the user approves the exact text.
- Otherwise, work from the uploaded file or pasted rows and return Markdown tables plus a CSV of coded comments.

## Method

```
- [ ] 1. Clean: dedupe, exclude test and employee rows, parse scores
- [ ] 2. NPS and rating summaries, overall and by segment
- [ ] 3. Code open-ended comments into themes
- [ ] 4. Link themes to promoters, passives and detractors
- [ ] 5. Rank sessions and speakers
- [ ] 6. Top 5 fixes and what to keep
- [ ] 7. Validate
```

1. **Clean.** Remove duplicate submissions (same email or respondent ID; keep the first and list the rest). Drop test rows and employees unless asked. Parse scores like "10 - definitely" or "8.0" as numbers; treat "N/A", blanks and out-of-range values as missing and count them. Report sponsors and exhibitors as their own segment or leave them out of attendee NPS, and say which.
2. **NPS.** Promoters score 9 to 10, passives 7 to 8, detractors 0 to 6. NPS = % promoters minus % detractors, a whole number from -100 to 100. Show n for every segment. Mark any segment under 5 responses "directional". Response rate = usable responses / attendees, if attendee count is known.
3. **Ratings.** For 1 to 5 scales, report average and top-2-box % (share of 4s and 5s). Don't average across different scales.
4. **Code comments.** Read every comment. Start from the codebook in [references/theme-codebook.md](references/theme-codebook.md), add event-specific themes when 3 or more comments share an idea, and merge themes with fewer than 3 mentions into "Other". A comment can carry more than one theme. Tag each coded mention positive, negative or mixed. Count mentions per theme by NPS group.
5. **Quotes.** Pick short, representative quotes (under 25 words) that match the theme's majority view. Trim names, emails, company names and anything identifying. Never invent or merge quotes; light trimming with "..." is fine.
6. **Sessions and speakers.** Rank by average session rating, then top-2-box %, then n. Show n and flag sessions under 5 ratings. Use comments to explain the low ones ("rated as a product pitch"). Recommend invite back, invite back with changes, or don't re-invite, with the reason.
7. **Fixes.** Rank negative themes by detractor mentions first, then total mentions. For each, name a concrete change, an owner role and when it has to happen (for example "book a room for 80 before the venue contract is signed").

Optional script: `scripts/survey_summary.py` computes rows in and out, NPS by segment and the session ranking deterministically:

```
python3 scripts/survey_summary.py responses.csv --exclude "Attendee Type=Employee"
```

It auto-detects columns or takes `--nps-col`, `--segment-col`, `--session-col`, `--rating-col`, `--dedupe-col`. If code execution isn't available, do the same counts in chat and show the arithmetic for the overall NPS.

## Output template

```
# [Event] Survey Results

## Headline
3 to 5 bullets: NPS (n), biggest strength, biggest problem, top session, response rate.

## Who answered
## NPS and ratings by segment
## What people loved (themes)
## What to fix (themes)
## Sessions and speakers
## Top 5 fixes for next time
## What to keep
## Method and caveats
```

Table columns:

- **NPS by segment:** Segment | Responses | Promoters | Passives | Detractors | NPS | Note
- **Themes:** Theme | Sentiment | Mentions | From detractors | From promoters | Example quote
- **Sessions:** Rank | Session | Speaker | Responses | Avg rating | Top-2-box % | What comments say | Recommendation
- **Fixes:** # | Problem (theme) | Evidence (mentions, detractor share) | Fix | Owner | By when
- **Coded comments CSV:** Response ID | Segment | NPS group | Theme(s) | Sentiment | Comment

For an executive version, keep only Headline, NPS by segment, Top 5 fixes and five quotes. A worked example is in [references/examples.md](references/examples.md).

## Rules that keep the analysis honest

- Report n next to every percentage and score.
- Don't compare NPS across events with different survey wording or timing without saying so.
- A theme count is mentions, not people, unless you counted people; say which.
- Don't attribute a complaint to a named speaker in an exec summary without the event team seeing it first.
- No external benchmarks for NPS: they vary too much by event type. Compare with the user's own past events.

## Validation checklist

- [ ] Rows in = rows used + rows dropped, with each drop listed
- [ ] NPS group counts add up to scored responses
- [ ] Every theme has a count and at least one real quote
- [ ] Every comment is coded (or marked "no comment")
- [ ] Segments and sessions under 5 responses are flagged
- [ ] No names, emails or identifying details in quotes

## Approval gates

Never post to Slack, email a summary or write to a shared sheet without showing the exact content first and getting a yes. Default to a draft. Don't send survey follow-ups to respondents from this skill.

## Privacy

Survey exports often hold names, emails and free-text personal details. Keep the analysis in the user's own tools. Strip identifying details from quotes. If respondents were promised anonymity, don't link comments to emails in any output. Respondents in the EU or UK may only be contacted about their answers if they agreed to that.

## Related skills

event-debrief (feeds these findings into keep, change, cut), event-roi-report, event-strategy-brief.

FAQ

How do I calculate NPS for an event?
Ask attendees how likely they are to recommend the event on a 0 to 10 scale. Scores of 9 or 10 are promoters, 7 or 8 are passives and 0 to 6 are detractors. Subtract the percentage of detractors from the percentage of promoters to get a score from -100 to 100. Always report how many people answered.
How do I analyze open-ended survey comments quickly?
Code each comment into a short list of themes, such as content quality, sales pitch, networking or wifi, and count mentions by promoters and detractors. The skill starts from a tested codebook, adds themes that appear three or more times in your data, and gives a real anonymized quote for each one.
How should I rank conference sessions and speakers?
Rank by average rating, then by the share of 4 and 5 ratings, and show how many people rated each session. Sessions with fewer than five ratings are directional. Comments explain the low scores, which is what tells you whether to drop a speaker or just change the format.
Does this work in ChatGPT as well as Claude?
Yes. It works in Claude, ChatGPT and other agents that support the SKILL.md standard. Upload your survey export and ask for the analysis. If code execution is on, an included Python script does the counting; if not, the AI does the counts in chat and shows its math.