We run live AI coaching sessions for enterprise clients — typically 4 sessions per cohort, 20–50+ learners per session. After each one, learners fill in a short survey (engagement rating, usefulness rating, a couple of open-text questions).
The old way: export the data, calculate averages manually, write up a summary, copy-paste it somewhere. It took too long and happened inconsistently, which meant coaches and the product team weren’t getting timely feedback.
Here’s what I built in Rebel instead.
The workflow
**Trigger:** I open a conversation in Rebel and say something like:
“Analyze feedback for [client] cohort, session 4”
**What Rebel does from there:**
- **Pulls the data** — connects to our coaching platform via MCP, fetches all survey responses for that cohort and session, filters out staff responses and anyone who flagged they didn’t attend live
- **Calculates the scores** — engagement and usefulness averages (0–10 scale), handles edge cases like zero ratings and duplicate entries using a Python script it writes on the fly
- **Analyses qualitative themes** — reads through all the open-text responses and surfaces up to 5 positive themes and 5 improvement suggestions
- **Flags watch-outs** — only serious ones (inappropriate content, safety concerns, major tech failures) — not just pacing gripes
- **Extracts standout quotes** — verbatim, not paraphrased
- **Generates a formatted Slack message** — ready to post, with all the right emoji and structure
- **Saves it to a scratchpad file** for the record, then **shows me the output for approval**
- **Posts to feedback** after I confirm
The whole thing takes about 90 seconds in conversation. I just check the output and hit approve.
What the output looks like
Here’s a real example from a recent cohort — a large enterprise organization, 56 respondents:
```
*Live Session feedback*
*[Client] - Cohort 1*
*March 12, 2026 – Session 4: AI Competency and beyond*
*
Coach(es):* [Lead Coach]
*Responses:* 56
*
Average Scores*
• Engagement: 7.1
• Practical Usefulness: 7.0
*
Key Positive Themes*
• Peer sharing format — hearing how other teams across the organization are using AI in their daily workflows was highly valued
• Real-world examples of applications built during the program were inspiring and concrete
• Sessions created genuine curiosity and enthusiasm for applying AI across different functions
*
Suggested Improvements*
• Peer-sharing format not universally relevant — role diversity made some examples hard to apply
• Desire for more advanced/2.0-level content
• Corporate tool restrictions remain an issue — tools covered are blocked at many orgs
*
Standout Quotes*
“It was great to hear about how other teams across the organization are using AI in their daily workflow.”
“hearing about the applications that have been built”
```
The team gets this in Slack within minutes of the session ending. No spreadsheet ever opened.
What made it click
A few things I didn’t expect:
**The skill handles the messy bits automatically.** Things like filtering out people who said “I watched the recording” (they’re not giving feedback on the live session), handling duplicate names in the speakers list, and only flagging watchouts for genuinely serious issues — not just “please go slower.” I had to think about these rules once when building the skill, and now they just run every time.
**The scratchpad file is underrated.** Every run saves a clean markdown file. That means I can look back at any session across any cohort without digging through Slack. It also means if a coach wants to see their full cohort trend, it’s already there.
**Conversation as interface.** I don’t navigate to a tool or fill in a form. I just say which cohort and which session in plain language. Rebel figures out the rest. That’s the thing that actually changed my behavior — the barrier to running it is low enough that I actually do it.
We’ve now run this across 10+ cohorts. Happy to share the skill if anyone wants to adapt it for their own program feedback or survey workflows.
*Built with: Mindstone MCP connector (coaching platform data) + Slack MCP (posting) + Python scripts (data processing) + Rebel Skills*