--- name: signal-synthesis description: "Synthesize product signals from customer feedback, support tickets, and account data into a prioritized opportunity grid. Use when a PM says: 'analyze my data', 'what should we prioritize', 'run signal synthesis', 'identify top opportunities', 'what are customers telling us', or drops feedback/support/CRM data and asks for insights. Also use when anyone mentions 'weekly intelligence brief', 'opportunity signals', or 'product intelligence'. This skill processes raw data exports and produces a ranked, evidence-backed opportunity map — not a feature list." --- # Signal Synthesis Skill ## Template version — built for adaptation Transform raw customer data into a ranked opportunity grid that tells PMs where to focus discovery. ## When to Use A PM has dropped data files into the session and wants to know: what are the top opportunities this data is surfacing? ## Before You Start 1. Read `references/strategic-context.md` (bundled with this skill). It contains the current strategy, OKRs, roadmap exclusions, and product priority order. If the file is not found, ask the PM for: (a) what is already on the roadmap (to exclude), (b) product area priority order, (c) top 2–3 strategic objectives for this cycle. 2. Identify what data files are available. The ideal set is three sources: - **Customer feedback export** (CSV) — in-app feedback, NPS verbatim, user requests *[Meridian example: Intercom conversations export — look for Subject, Body, Customer Name, Account fields]* - **Support ticket export** (CSV) — customer-reported issues and feature requests *[Meridian example: Zendesk tickets — look for Subject, Description, Account Name, ARR, Priority fields]* - **Account and health data** (CSV or XLSX) — ARR, customer tier, health score, renewal dates *[Meridian example: Salesforce account export — look for Account Name, ARR, Health Score, Tier, Contract Renewal Date fields]* - Any strategy documents, OKR trackers, or roadmap slides for additional context 3. You can run with as few as one data source. Two is the minimum for meaningful signal convergence. Three is ideal. Tell the PM what you have and what is missing before you start. --- ## Step 1: Data Profiling (Do This First, Always) Before any analysis, profile each file: - Count total rows and unique accounts - Identify the source and type breakdown - Identify product area distribution if available - Identify date range of the data - Flag data quality issues: missing account names, blank descriptions, obvious outliers Share a brief summary with the PM before proceeding: "Here is what I am working with: X rows of feedback across Y unique accounts, Z support tickets covering [date range], etc." This builds trust and catches data problems early. --- ## Step 2: Theme Extraction Analyze the text content to identify recurring themes. Work at the initiative level, not the feature level — cluster related items into themes. Approach: - Read through the data systematically, not just keyword-matching. Customer language is inconsistent — "my numbers keep changing," "I can't trust the forecast," and "model keeps resetting" are likely the same theme. - Group sub-items under parent themes. A theme like "Forecasting Reliability" might contain: inaccurate predictions, manual override friction, stale data warnings, and sync delays between runs. - Tag each theme with the product area(s) it touches. - Preserve specific sub-items and raw customer quotes — these become the composing signals in the output. --- ## Step 3: Signal Strength Scoring (1–10) For each theme, calculate signal strength based on four factors: **Frequency**: How many individual data points mention this theme across all sources? - 1–5 mentions = low (1–3) - 6–20 mentions = moderate (4–6) - 21–50 mentions = high (7–8) - 50+ mentions = very high (9–10) **Account breadth**: How many unique accounts are affected? - 1–3 accounts = narrow (dampen score by 1–2 points) - 4–10 accounts = moderate (no adjustment) - 10+ accounts = broad (boost score by 1 point) **Source convergence**: Does the theme appear in multiple data sources? - Single source = no boost - Two sources = +1 - Three+ sources = +2 **ARR weight** (if account data is available): What is the total ARR of accounts raising this theme? - Under $100K total ARR = no boost - $100K–$500K total ARR = +1 - $500K+ total ARR = +2 (cap total signal score at 10) Assign an **evidence tier** for each theme: - **Anecdotal**: 1–2 sources, fewer than 10 data points, fewer than 5 accounts - **Emerging**: 2–3 sources, 10–30 data points OR significant ARR concentration - **Validated**: 3+ sources, 30+ data points, 10+ accounts, strong ARR weight --- ## Step 4: Strategic Alignment Scoring (1–10) For each theme, score against the current strategic context. **Financial lever fit** (0–4 points): *[Meridian's financial levers in 2026 priority order]* - Directly addresses retention and reduces churn = 4 - Drives expansion and upsell within existing accounts = 3 - Enables new logo conversion = 2 - Indirect or unclear connection to financial outcomes = 1 **Product priority alignment** (0–3 points): *[Meridian's product areas ranked by 2026 strategic priority]* - Forecasting (AI accuracy, model reliability, confidence) = 3 - Pipeline Management (health signals, deal risk, hygiene) = 3 - Onboarding and activation (time-to-value, setup completion) = 2 - Attribution (multi-touch, ROI reporting) = 1 - Platform and integrations (Salesforce sync, HubSpot, ecosystem) = 2 **Roadmap coherence** (0–3 points): *[Score against committed roadmap items in strategic-context.md]* - Directly supports a committed roadmap item (enables it, unblocks it, or the roadmap item underdelivers without this) = 3 - Adjacent to the roadmap direction = 2 - Orthogonal but valuable on its own = 1 - Contradicts or competes with committed roadmap = 0 Sum the three sub-scores for the Strategic Alignment score (max 10). **Roadmap exclusion check**: If a theme is essentially restating something already on the committed roadmap, exclude it from the opportunity list entirely. Call it out in the exclusions section with a note explaining why it was excluded. --- ## Step 5: Build the Output ### Part A: Opportunity Grid Create a summary table with these columns: - Opportunity number and title - Signal Strength (1–10) - Strategic Alignment (1–10) - Evidence tier (Anecdotal / Emerging / Validated) - Quadrant label: - High Signal + High Alignment (both 7 or above) = **"Act — pursue this cycle"** - High Signal + Lower Alignment (signal 7+, alignment below 7) = **"Investigate — strong signal, debate fit"** - Lower Signal + High Alignment (signal below 7, alignment 7+) = **"Explore — strategically important, gather more evidence"** - Low Signal + Low Alignment (both below 7) = **"Monitor — park for now"** Sort by quadrant (Act first), then by combined score within each quadrant. ### Part B: Opportunity Detail Cards For each opportunity (aim for 5–8 total), produce: ``` ## [##] [Opportunity Title] — [EVIDENCE TIER] **Problem**: [2–3 sentence description of the customer problem, not a solution] **Signal Strength: [score]/10 Strategic Alignment: [score]/10** **Sources**: [Which sources, with counts. Example: "Intercom feedback (14 items), Zendesk tickets (22 items / $380K ARR), NPS verbatim (8 responses)"] **Composing signals**: - [Specific sub-item with a customer quote and account examples] - [Specific sub-item with a customer quote and account examples] - [Continue for all meaningful sub-items] **Why now**: [1–2 sentences connecting this to the current strategic moment. What happens if we don't act on this in the next cycle?] ``` ### Part C: What We're Not Prioritizing List themes that appeared in the data but did not make the cut. For each: theme name, where the signal came from, and why it is parked. This section makes the prioritization defensible when a stakeholder asks why a topic is not on the list. ### Part D: Data Sources Summary Document what was analyzed: source name, row count, date range, key columns used, any data quality caveats (join rate, missing data, outliers removed). --- ## Output Format Produce the full output as a single markdown document. If the PM asks for a shareable document, export to the format their team uses. --- ## Efficiency Notes - Account and CRM datasets can be very large. Pre-filter to active customers only and extract only: Name, ARR, Health Score, Tier, Renewal Date. Discard the rest to save context window. - When joining feedback account names to CRM data, use fuzzy matching if needed. Note the match rate — if under 60% of feedback accounts match to CRM records, flag this as a data quality caveat in the output. - Support ticket exports often have dozens of columns. Extract only: Subject, Description (first 300 characters), Account Name, ARR if available, Priority, Created date. --- ## What Comes Next After the PM reviews the output and selects 1–2 opportunities to pursue, use the **opportunity-brief** skill to generate a full opportunity brief for each selected opportunity. Pass it the opportunity detail card from this output — the brief skill will expand it into a complete brief with customer evidence and problem framing ready for your team.