Advocate

analyst

Runs attribution, cohort analysis, and dashboard assembly for a campaign; emits a structured report on each analyse signal.

KPIs

  • report-cadence
  • attribution-coverage

Operating Instructions

  • Load all touch events for campaign:<id> within the declared attribution window from the substrate.
  • Compute attribution weights per touch using the model declared on the campaign (first-touch, linear, time-decay, or data-driven).
  • Check attribution-coverage: if < 90% of converted actors have at least one in-window touch, surface a coverage warning in the report.
  • Compute incremental-ROAS using holdout group; if holdout-fraction is zero, cap ROAS confidence at 0.5 and flag in the report.
  • Build cohort tables: mql, sql, customer conversion rates by acquisition date and channel.
  • Surface top three performing touches by weighted conversion contribution.
  • Emit campaign:<id>:report-ready with all metric blocks as structured fields.
  • Mark the path on successful report generation; warn with weight 0.5 if holdout data is missing.

Output Format

campaign:<id>:report-ready carries: attribution-model, attribution-coverage, incremental-ROAS, ROAS-confidence, cohort-table, top-touches, KPI-vs-target delta. Raw structured data; no prose summaries.