funnel-attribution-report
Reconcile cross-system marketing and sales attribution from traffic/ad/search/social sources through website events, calls, Vapi, bookings, CRM, completed jobs, and revenue. Use when the user asks whe
SKILL.md.claude/skills/funnel-attribution-report/SKILL.md
GitHubFunnel Attribution Report
Audit both the customer funnel and the measurement system used to describe it.
Workflow
- Define the business funnel and exact reporting period before comparing counts.
- Inventory systems and identifiers.
- traffic/source identifiers
- click IDs/UTMs when present
- session or lead IDs
- event_id or server/browser identifiers
- phone/call IDs
- appointment/CRM/customer IDs
- invoice/job/order IDs
- Map each stage and its source of truth:
- exposure/visit -> intent event -> lead/contact -> qualified lead -> booking/order -> arrival/completion -> revenue -> repeat/review.
- Reconcile counts between adjacent stages. Explain legitimate differences such as attribution windows, timezones, repeat interactions, modeled conversions, spam, cancellations, offline completion, and deduplication.
- Find measurement defects:
- double-counted events
- missing server/browser events
- overwritten UTMs
- lost click IDs
- calls without source mapping
- booking/CRM records with no originating lead
- platform conversions without backend receipts
- revenue records that cannot be joined back to acquisition
- Find operational leakage separately from tracking leakage. A missing booking can be a sales/process problem even when attribution is perfect.
- Quantify uncertainty. Use ranges or unresolved counts rather than false precision.
- Read
references/report-contract.mdand follow its output structure.
Rules
- Never force a single-source attribution story when the evidence supports multi-touch or unknown origin.
- Never reconcile by changing definitions silently.
- Preserve source-system counts and show derived reconciliations separately.
- Prefer deterministic joins; label probabilistic or heuristic matching.
- Protect customer-level data in outputs.
Useful triggers
Examples: "audit our attribution", "why does Meta say 80 leads but CRM says 42?", "trace calls to revenue", "where are customers dropping out?", "can we trust our conversion numbers?"