SKILL FILE

Insight Synthesiser with AI

Turn scattered customer quotes into strategic insights with patterns, themes, confidence ratings, and recommended actions. Lighter and faster than full multi-source synthesis.

Download Skill File ↓

What this skill file teaches Claude

Drop one markdown file into your repo. Claude Code learns how to run this entire workflow.

1

Top 3-5 headline insights

Forces synthesis down to a small set of decision-relevant insights. Each backed by evidence count, not vibes.

2

Quote tagging and clustering

Every quote tagged by topic, sentiment, persona match, and strength. Clusters surface patterns automatically.

3

Confidence ratings

Strong / medium / weak per insight based on quote volume and consistency. Honest about what needs more validation.

4

Surprises and contradictions

Pulls out the things you didn't expect to hear (often the headline) and the places quotes disagree (usually two segments).

5

Recommended action per insight

Each insight paired with a product / marketing / sales / research action. Stops insights dying in a memo nobody acts on.

6

Open questions list

The follow-up research the synthesis surfaced, so the next discovery round has a head start.

What you can build with this

Post-sprint synthesis

You ran a discovery sprint, collected 10 interviews, and need 3 headline insights for the Friday stakeholder readout.

Sales-call quote pile

Your AE has 50 call notes in Notion. Find the pattern, surface what marketing and product should know.

Support ticket review

Last quarter's support tickets reveal a pattern nobody noticed. Synthesise them to surface what should change in the product.

Survey free-text analysis

200 survey free-text responses, no time to read them all. Tag and cluster to surface the 3 things people are actually saying.

Get the full skill file

Everything above is 80% of the skill file. Download the complete version with full implementation details, agent prompts, and ready-to-run scripts.

Common questions

This one is lighter and faster, built for a single source (a pile of quotes). Research-synthesis-engine combines multiple sources (interviews + surveys + behavioural data + support) into one unified synthesis. Use this when you have quotes only and need answers by Friday. Use the other when you have a richer mix and time to do it properly.
15 to 20 is the minimum for meaningful pattern-finding. Below 10 you're probably reading individual stories, not synthesising. Above 100 the tagging gets long but the insights get sharper. The sweet spot is 30 to 60 quotes from a mix of sources.
Honest output. If the data only supports weak insights, the skill says so. A weak rating means "interesting hypothesis, needs more evidence" not "discard." The recommended next step is usually a focused follow-up research round to either strengthen or kill the hypothesis.
Yes. The synthesis works without attribution. You lose the ability to trace a quote back to a specific call, but the pattern detection still works. Try to keep at least the source type (interview / sales call / support ticket) so the confidence rating can weight evidence appropriately.
Either order. They produce different shaped output from the same quotes. Insight synthesiser gives you topical themes ("users abandon at second screen"). JTBD extractor gives you job-framed needs ("when planning my week, I want to..."). If you're doing both, run insights first for the topical map, then JTBD for the job framing.
Bad sign. Either the questions led to compliments (run `/interview-guide` next time) or you've selected for happy customers. The synthesiser will flag low signal-to-noise if a quote pile is mostly positive surface comments. Compliments rarely produce insights. Frustration, workarounds, and specific stories do.

Ready to automate?

SoloStack gives you every skill pre-installed — scraping, marketing, sales, CRM, and more. One repo. Every department.

Book a Call →