analyze-patterns

# Analyze Patterns Across Clay Data Synthesize insights from stored Clay analysis records. You ARE the synthesis engine — query the data and reason over it directly. ## Step 1: Understand what's available Call `list_analysis_types` to see what data is stored. Share the summary with the user so they know what's available. ## Step 2: Fetch relevant records Based on the user's question, decide the best approach: - **Broad pattern analysis**: Use `query_records` with the relevant `analysis_type`, fetch up to 50-100 records - **Topic-specific search**: Use `semantic_search` to find records matching a specific theme (e.g., "budget objections", "competitor mentions") - **Entity comparison**: Use `query_records` filtered by `entity_id` to compare analyses of the same deal/company across stages - **Time-based trends**: Use `since`/`until` filters to compare different periods ## Step 3: Synthesize When analyzing the returned records, look for: 1. **Recurring patterns** — what keeps showing up across multiple records? Don't just average scores — identify specific behaviors or themes that repeat 2. **Outliers** — which records deviate significantly? What makes them different? 3. **Trends** — are things getting better or worse over time? Compare earlier vs recent records 4. **Segments** — do patterns differ by rep, entity, source, or tag? Break down by dimensions 5. **Root causes** — separate individual issues from systemic ones. "One rep's problem" vs "team-wide gap" vs "process issue" 6. **Actionable recommendations** — what should change? Be specific: cite record IDs and data as evidence ## Step 4: Format findings Structure your analysis as: ### Key Findings - Top 3-5 patterns with specific evidence (cite record IDs) ### Breakdowns - Per-segment analysis if applicable (by rep, entity, tag) ### Recommendations - Specific, actionable next steps - Attribute each recommendation to the right owner (coaching, process, tooling) ## Cross-Stage Analysis When the user asks to compare analyses across stages (e.g., BDR vs AE assessment of the same deal): 1. Fetch records with the same `entity_id` but different `analysis_type` values 2. Compare the assessments side by side 3. Identify where evaluations diverge — this reveals handoff gaps 4. Note: "The BDR scored this B but the AE found it was a D" is a high-value insight ## Important - Always cite specific records as evidence — don't make generic claims - If there are too many records to process at once, work in batches and synthesize across batches - If the user hasn't imported data yet, suggest using the import-data skill first