| name | strategic-voc-synthesis |
| description | Validate discovery hypotheses against Voice-of-Customer signal across tickets, reviews, churn reasons, surveys, and sales feedback. Produces a synthesis that confirms, contradicts, or qualifies a hypothesis with triangulated VoC evidence. Use to pressure-test a discovery finding before committing it to the roadmap. |
| argument-hint | [the discovery hypothesis to validate against VoC] |
| user-invocable | true |
| allowed-tools | Read, Write, Edit, Bash, Glob, Grep |
Validate $ARGUMENTS against Voice-of-Customer signal.
This is the product manager's VoC lens. In a distributed-VoC model, several roles each hold their own
distinct view of the customer's voice — support hears tickets, customer-success hears churn and health, the
ux-researcher hears usability sessions, GTM hears win/loss. Those lenses overlap on purpose: different
roles surface different signals from the same customer base, and the conflict between them is informative,
not a problem to reconcile away. The PM's lens is specifically about whether a discovery hypothesis holds
up against the breadth of customer signal before it earns a place on the roadmap.
Follow every step. The output is a triangulated verdict on the hypothesis, with the conflicting lenses
surfaced rather than smoothed over.
Step 1: State the hypothesis precisely
Write the discovery hypothesis as a falsifiable claim about a customer need or behaviour, usually carried
over from interview synthesis (/product-manager:synthesise-interviews): "Mid-market accounts churn
primarily because onboarding takes more than two weeks." Vague hypotheses can't be validated. This step is
complete when the hypothesis is a single falsifiable statement.
Step 2: Pull each VoC source and tag its lens
Gather signal from every available source, and record which role's lens it comes from:
| Source | Lens / owner | What it tells you | Bias to watch |
|---|
| Support tickets | Support | Frequency and acuteness of friction | Skews to users who complain |
| Churn / health signal | Customer success | Why accounts leave or stall | Post-hoc rationalisation |
| Reviews (G2, Capterra, app stores) | (no single owner) | Comparative sentiment, competitor gaps | Self-selecting; reviewers go to complain or rave |
| Surveys (NPS/CSAT/CES) | Product / CS | Trend and segment cuts | Stated, not behavioural |
| Win/loss | GTM | Buyer-side decision drivers | Sales-reported reasons differ from buyer-reported |
| Discovery interviews | PM | Depth on the specific need | Small N |
For each source, record what it says about the hypothesis. This step is complete when each available source
has been pulled and tagged with its lens.
Step 3: Triangulate and weight by evidence tier
Don't treat all sources equally. Weight regulatory/behavioural and independent signal above self-reported
and subject-supplied. A claim supported by one source is single-sourced — label it. A claim supported
across two or more independent lenses is triangulated. Behavioural signal (what accounts actually did)
outweighs stated signal (what they said in a survey). This step is complete when each piece of support is
tiered and single-source claims are labelled.
Step 4: Surface the conflicting lenses
Where two lenses disagree — sales says price, the interviews say onboarding; tickets say one thing, churn
data another — surface the conflict explicitly. Do not average it away or pick the convenient one. The
disagreement usually means the hypothesis is partial, segment-specific, or that one lens is closer to the
real driver. Name which lens you trust more for this hypothesis and why. This step is complete when every
cross-lens conflict is stated, not hidden.
Step 5: Verdict
Render one of three verdicts with the evidence behind it:
- Confirmed — triangulated across independent lenses, behavioural signal present. Ready for the roadmap
- Qualified — true for a segment or under conditions; state the boundary
- Contradicted — VoC doesn't support it; send back to discovery or drop
This step is complete when the verdict is stated with a confidence rating (0-4) and the evidence behind it.
Rules
- Triangulate or label single-source. A hypothesis riding on one VoC source is not validated. Mark it
[single source] and seek a second lens.
- Behaviour over stated intent. What accounts did outweighs what they said in a survey. Weight
accordingly.
- Surface conflict, don't reconcile it. When lenses disagree, the disagreement is the finding. Don't
pick the lens that confirms your plan.
- Watch self-selecting samples. Review sites and complaint channels skew. Note the skew rather than
treating raw sentiment as representative.
- Consult the other lens-holders. This is the PM's view, not the only view. Where the hypothesis touches
pricing, churn, or positioning, consult GTM, customer-success, or the ux-researcher for their lens.
- Don't manufacture a verdict. If the signal is genuinely thin, the verdict is "needs more discovery",
not a forced confirm.
Output Format
Write to docs/product/voc-synthesis-[hypothesis-slug].md:
# VoC synthesis: [hypothesis]
**Hypothesis (falsifiable):** ...
## Signal by source
| Source | Lens | What it says | Tier | Single-source? |
|--------|------|--------------|------|----------------|
| ... | ... | ... | T1-T5 | yes/no |
## Cross-lens conflicts
- [Lens A] says X; [Lens B] says Y → [which is closer for this hypothesis, and why]
## Verdict
**[Confirmed / Qualified / Contradicted]** — confidence [0-4]
[evidence summary; if Qualified, state the segment boundary]
## Lenses to consult further
- [role] for [aspect]
Related Skills
/product-manager:synthesise-interviews — produces the hypotheses this skill validates./product-manager:define-icp — VoC signal sharpens the ICP's behavioural criteria./product-manager:write-roadmap — only confirmed or qualified hypotheses earn a roadmap place.
