pm-transversal-analysis
Skill1 starsChanged 42 days agoMITView on GitHub
Cross-phase skill for **qualitative synthesis, quantitative analysis, triangulation of quali+quant evidence, and parsing of videos/audio/transcripts** into product insight. Invoke whenever the user hands over interview transcripts, user-test recordings, call notes, survey free-text, support tickets, session replays, analytics queries, or funnel/cohort data — and expects insight, not just a summary. Trigger on "sintetize essas entrevistas", "triangule quali e quant", "analyse these transcripts", "cohort analysis", "funnel interpretation", "batch synthesis", "fecha a tarefa via MCP". Produces thematic syntheses with evidence quotes, cohort/funnel interpretations, and triangulated findings that make the team act differently. For interviews at the very start of a discovery (problem still soft, JTBD undefined), use `pm-phase-discover` first to scope the research, then chain to this skill for synthesis — this skill assumes the upstream framing is set.
In this repository65 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- anti-slopSkill
- data-science-analystSkill
- humanize-deliverablesSkill
- humanizerSkill
- inference-disciplineSkill
- pm-archetype-aiSkill
- pm-archetype-enterpriseSkill
- pm-archetype-growthSkill
- pm-archetype-platformSkill
- pm-phase-defineSkill
- pm-phase-deliverSkill
- pm-phase-developSkill
- pm-phase-discoverSkill
- pm-prioritization-regua-comumSkill
- pm-product-senseSkill
- pm-storytellingSkill
- pm-transversal-analysisSkill
- pm-transversal-commsSkill
- pm-transversal-docsSkill
- pm-transversal-stakeholderSkill
- repo-doctorSkill
- anti-slopSkill
- data-science-analystSkill
- humanize-deliverablesSkill
- humanizerSkill
- inference-disciplineSkill
- pm-archetype-aiSkill
- pm-archetype-enterpriseSkill
- pm-archetype-growthSkill
- pm-archetype-platformSkill
- pm-phase-defineSkill
- pm-phase-deliverSkill
- pm-phase-developSkill
- pm-phase-discoverSkill
- pm-prioritization-regua-comumSkill
- pm-product-senseSkill
- pm-storytellingSkill
- pm-transversal-commsSkill
- pm-transversal-docsSkill
- pm-transversal-stakeholderSkill
- repo-doctorSkill
- umbertoSkill
- anti-slopSkill
- data-science-analystSkill
- humanize-deliverablesSkill
- humanizerSkill
- inference-disciplineSkill
- pm-archetype-aiSkill
- pm-archetype-enterpriseSkill
- pm-archetype-growthSkill
- pm-archetype-platformSkill
- pm-phase-defineSkill
- pm-phase-deliverSkill
- pm-phase-developSkill
- pm-phase-discoverSkill
- pm-prioritization-regua-comumSkill
- pm-product-senseSkill
- pm-storytellingSkill
- The first 60 shown.
The file
Files
PM Transversal — Qualitative + Quantitative + Media analysis
Transversal skill. The analytical layer that sits under every phase. Discovery without synthesis is anecdote; analytics without qualitative grounding is confident guessing.
Progressive loading
Load this SKILL.md first. For large or specialized tasks, use references/progressive-loading.md to choose the narrowest supporting reference before reading more.
When to use this skill
Invoke when:
- interview transcripts, user-test recordings, or session replays need to be turned into insight
- survey free-text, support tickets, NPS comments, or in-product feedback must be coded and synthesised
- a funnel, retention, cohort, or segmentation analysis needs interpretation (not just numbers)
- quali and quant evidence conflict and must be reconciled (triangulation)
- a video/audio recording or its transcript is the main input (user test, sales call, exec interview)
- an "analysis" is needed but nobody has said whether it's quali, quant, or both
Skip for pure dashboard consumption (no synthesis needed) — that stays in pm-phase-deliver/references/product-analytics.md.
Trigger phrases: "sintetize essas entrevistas", "analyse these transcripts", "assista esse vídeo", "extraia insights", "coding temático", "triangule quali e quant", "cohort analysis", "funnel interpretation", "qualitative themes", "review session replays".
For interviews at the very start of a discovery (problem still soft, JTBD undefined, segments still in flux), use pm-phase-discover first to scope the research and only then chain to this skill for synthesis — this skill assumes the upstream framing is set and the work is "what does the evidence mean for product?".
Prime directive
Evidence → insight → decision. Quotes alone are colour. Numbers alone are noise. The expert move is synthesis that changes behaviour: "because of X (quali evidence) and Y (quant evidence), we should Z."
Core sub-skills
1. Qualitative synthesis (interviews, transcripts, feedback)
Turn raw text/audio/video data into themes with evidence. Open coding → axial coding → thematic synthesis. Distinguish user pain from user request.
Outputs: theme map, coded excerpts with timestamps/locators, insight repository entry (problem, evidence, segment, implication), opportunity list ranked by evidence strength + frequency.
Anti-patterns: cherry-picking quotes, no sampling discipline, confusing request with pain, "analysis paralysis" vs acting on clear themes, n=1 generalisation.
→ Deep-dive: references/qualitative-synthesis.md
2. Quantitative analysis (funnels, cohorts, retention, segments)
Interpret behavioural data structurally — not just "it went up". Understand what the funnel/cohort means for the hypothesis.
Outputs: funnel breakdown, cohort retention narrative, segment comparison, statistical framing of the claim (CI, practical significance, guardrails), decision implication.
Anti-patterns: dashboard worship, no segments, correlation-as-cause, p-hacking, reporting averages that hide segment harm.
→ Deep-dive: references/quantitative-analysis.md
3. Triangulation (quali + quant)
Cross-reference qualitative themes with quantitative signals to strengthen or challenge a conclusion. If quali says "checkout is confusing" and quant says "drop-off at step 3 is 40% higher than benchmark", the insight is stronger and the fix is targeted. If they conflict, investigate the gap.
Outputs: triangulation memo (claim → quali evidence → quant evidence → confidence → action), conflict log (where they disagree and why), sharpened hypothesis for next experiment.
Anti-patterns: using one type to validate the other without checking assumptions, over-weighting quant because it feels objective, ignoring quali because it feels soft.
→ Deep-dive: references/triangulation.md
4. Media and transcript parsing (video, audio, transcripts)
Process videos/audio files/transcripts (interviews, user tests, sales calls, session replays with narration) to extract insight. Use timestamps as locators; preserve raw signal; synthesise in a second pass. If a transcript is not available, request one or use available tooling to generate one before analysis.
Outputs: timestamped highlights, coded excerpts, thematic synthesis (§1), key-quote list, follow-up-question list for next research session.
Anti-patterns: "I watched the video and vibed with it", summarising before coding, losing the raw reference, treating transcripts as ground truth without checking against audio on load-bearing quotes.
→ Deep-dive: references/media-transcript-parsing.md
5. Batch interview synthesis (5-10 transcripts)
The many-transcript case: fix one shared codebook first, give every transcript its own fixed-schema excerpt log (one worker per transcript where the harness offers subagents, one file at a time otherwise), then merge by participants, never by quotes, with a counter-evidence column and a saturation check. Raw transcripts stay in raw-evidence/; worker lines marked inferred stay inferred in the memo.
Outputs: frozen codebook, one excerpt log per transcript, merge table, synthesis memo at discovery/<topic>/synthesis.md, ranked themes in insights.md.
Anti-patterns: no codebook before fan-out, counting quotes, the lead reading raw transcripts, names in the memo, worker inference promoted by paraphrase.
→ Deep-dive: references/batch-interview-synthesis.md
6. Connector task recipes (Jira and analytics MCP)
Close a routine PM task end to end through MCP connectors: launch retro from tickets, feature adoption from analytics, behaviour-split retention. Each recipe fixes the question, the tools by suffix name, the discovery step before the query, the query shape, the output with numerator, denominator, window (with its time zone) and source link, completeness (results paged to the end), the unit of analysis and its deduplication key, where it persists, and what stays TBD because no tool returned it.
Outputs: the number with its link and query text persisted in project memory; TBD claims named with an owner.
Anti-patterns: stating a number no tool returned, percentage without denominator or window, a count from the first page, a cohort defined with information from the window it is measured in, adoption read as retention, ephemeral result without the query.
→ Deep-dive: references/connector-task-recipes.md
Workflow
- Load the raw data — transcripts, recordings, dashboards, exports. If media is provided, confirm transcript is available or request one; use Read on provided text files. When present,
.ai/memory/org/personas.mdsupplies the segment labels the coding scheme should reuse. For five or more transcripts followreferences/batch-interview-synthesis.md: shared codebook first, one excerpt log per transcript, merge by participants. - Classify the ask — quali, quant, triangulation, or media-first?
- Sampling discipline — for quali, check sample size and segment coverage. For quant, check cohort definitions and exclusion rules.
- Code then synthesise — for quali, resist the urge to summarise before coding. For quant, define the hypothesis before slicing.
- Triangulate — always ask "what does the other type of evidence say about this claim?"
- State confidence and implication — every insight ends with "confidence: low/med/high; implication: X".
- Persist — ranked themes →
.ai/memory/projects/<slug>/insights.md(locators only); per-topic memo →discovery/<topic>/synthesis.md; connector results →analytics/<topic>-<date>.mdwith the query text; experiment plan if needed →experiments.md; raw signal stays inraw-evidence/(PII, never copied into memory or chat).
Output contract
## [Synthesis / analysis title]
### Inputs
- raw sources + sample size + time range
### Method
- how analysis was done (coding scheme, cohort logic, triangulation protocol)
### Findings (ranked by evidence strength)
1. [Claim]
- Qualitative evidence: [theme, N quotes, timestamps/locators]
- Quantitative evidence: [metric, cohort, magnitude, CI if available]
- Confidence: low/med/high
- Implication: [what to do differently]
2. ...
### Conflicts or gaps
### Recommended next action
### Files / raw artefacts preservedIntegration
pm-phase-discover— feeds research synthesis, segment models, opportunity ranking.pm-phase-deliver— interprets post-launch metrics, A/B results with quali context.- Transversais:
pm-transversal-docs(synthesis published in Confluence insight repo),pm-transversal-stakeholder(exec memo citing triangulated evidence). - MCP tools: when analytics (PostHog) or Jira connectors are available, this skill queries them directly for quantitative inputs (tool names by suffix, such as
query-runandsearchJiraIssuesUsingJql; themcp__<server>__prefix varies by environment) followingreferences/connector-task-recipes.md: numbers only from tool results, source links in the output, query text and decision persisted. When the team uses Zoom/Gong/Otter, transcript files come in via Read.
Communication modes follow ../../docs/patterns/COMMUNICATION_MODES.md. Per-skill: Lean (default) is ranked findings + implications + confidence; Standard is the full synthesis memo with method + findings + quotes; Caveman is the top 3 findings in 2 lines each.
Success criteria
- insights change team behaviour (new experiment, scope change, deprecation)
- raw signal is preserved and retrievable
- quali and quant are triangulated by default, not on request
- video/transcript analysis produces timestamped evidence, not vibes
- the team stops confusing "I talked to a customer" with "I have evidence"
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

