User Feedback Synthesizer

Raw feedback arrives as noise: support tickets shouting, survey comments mumbling, sales notes paraphrasing, and one forwarded email from a founder outweighing fifty quiet datapoints. This skill runs the pile through a disciplined synthesis — atomize, tag, cluster, weigh, rank — so what comes out is a short list of evidenced problem themes with counts, severity, and quotes, not a mood.

When to use this skill

• A backlog of tickets, reviews, survey verbatims, or call notes needs to inform a decision
• Post-launch listening: two weeks of reactions need shaping into a keep, fix, or revert view
• Different teams cite "what users are saying" and are somehow all correct and all contradictory
• A recurring monthly synthesis feeds the roadmap and quality reviews

Workflow

1. Atomize everything into single-observation units. One unit is one user expressing one thing; a ticket complaining about three problems becomes three units. Keep the verbatim text, source, date, and whatever segment data exists — plan, tenure, company size.
2. Strip the solution wrapper off feature requests. "Add an export button" is recorded as the underlying need — cannot get data out to their reporting workflow — with the requested solution noted alongside. Users are experts on their problems and volunteers on your product design.
3. Cluster bottom-up, not into predefined buckets. Group units that describe the same underlying problem and let categories emerge and merge. Predefined buckets find only what you already believed.
4. Name each theme as a problem statement, not a noun. "New admins cannot find where sharing permissions live" — never "permissions". A theme that resists a problem statement is two themes, or none.
5. Count and weigh. For each theme: unit count, distinct users, share by segment and by channel, and a severity read — blocks core work, degrades daily use, annoys, or cosmetic. Weight severity and strategic-segment concentration above raw count: ten enterprise admins blocked outranks forty free-tier grumbles, and saying so explicitly is the analyst's job.
6. Attach two or three representative quotes per theme, chosen for typicality rather than vividness, each with source and segment. Quotes carry conviction that counts cannot.
7. Rank and recommend. Order themes by weighted importance. Give each a recommended motion — investigate, fix, redesign, communicate (some problems are documentation problems), or monitor — and note contradicting evidence wherever it exists.

Output format

A synthesis memo containing: a method line (sources, date range, unit count); the theme table — problem statement, units, distinct users, severity, trend versus last period, key segments; quotes per theme; ranked recommendations; and a weak-signals footer for small-count items worth watching that do not yet justify action.

Guardrails

• Never let one channel dominate silently: report the channel mix and flag any theme living in only one channel
• Distinct users, not units, is the headline count — one prolific complainer is one datapoint
• Frequency measures detectability, not importance; severity and segment weighting are mandatory
• Silence about a flow users abandon is not satisfaction; note where the feedback channels cannot see
• Keep an audit trail from every theme back to its units so any claim can be spot-checked in seconds

Worked example: one theme row

Theme: "Finance users cannot reconcile exported totals with the dashboard." Units: 23. Distinct users: 17 across 11 accounts, 9 of them on the business tier. Severity: degrades daily use, with two units at blocks-core-work. Trend: doubled since the reporting release. Recommended motion: fix — and communicate, because three units show users distrusting correct numbers. Quote: "the export says one thing, the screen says another, so now I check both by hand every Friday."