Reporting — Dossier

Objective

Deliver accurate, unified reporting to clients and internal ops — anchored on GA4 as source of truth, enriched with platform diagnostics and CRM outcomes.

Audiences

Audience	Views
Client	KPI dashboard, spend, conversions, simple narrative
Operator	Reconciliation, anomalies, approval backlog context
Leadership	Portfolio roll-ups across tenants (internal)

Attribution joins

Campaign-level reports join GA4 to platform cost/spend using platform campaign ID first (utm_campaign macros), then click IDs, then normalized name — targets ≥ 95–98% match — UTM & attribution spec.

Meta CPAS (Collaborative Ads) — Phase 2: conversions occur on marketplace PDPs, not the client site. CPAS campaigns use shared-item / catalog-segment metrics from Meta Insights (catalog_segment_value, converted_product_value, segment ROAS) as primary KPIs — not GA4-first. See Meta CPAS spec.

Data sources

Source	Role	Kobi operates BQ export?
GA4 (Analytics Data API)	Primary metrics — source of truth	No — read via API; client may run their own GA4→BQ separately
Google Ads (Kobi BQ + API)	Spend, structure, platform diagnostics; join to GA4	Yes — agency-owned accounts exported to Kobi BigQuery
Meta / TikTok / DV360	Spend, diagnostics (secondary to GA4)	Planned — API → Kobi warehouse as connectors mature
Meta CPAS (Phase 2)	Shared-item / catalog-segment conversion metrics on marketplace SKUs	Planned — Insights API; not GA4-primary
MMP	App attribution (if client uses)	No — client-owned; out of scope unless future integration
CRM	Pipeline, qualified leads, revenue	Internal CRM API — not platform BQ
Internal audit log	Plan versions, human actions	Kobi ops telemetry BQ

Standard reports

Client weekly

Active plan tracks — always-on, branding, engagement, active events — each with vN and approval date
Spend by channel and by track (vs that track's approved plan and vs prior week)
Plan drift indicator — per track manifest vs plan (highlight if revise recommended)
Primary KPI vs target
Top campaigns (GA4-attributed)
Optimization changelog — per track; what changed, why, auto vs human-approved
Pending plan revise / replan — per track diff summary if draft vN+1 exists
Upcoming special-day / event plans — scheduled event tracks and budget preview
Upcoming seasonal or scheduled plan changes

Internal daily ops

Tracking health panel
GA4 vs platform variance alerts
CRM upload status
Pending human approvals

Monthly executive (per client)

Trend vs prior month
Plan version history — v1…vN with revise vs replan labels
Optimization impact summary — KPI delta attributed to top change sets
Replan / revise history and approval latency
Vertical benchmark (when available)

Reconciliation

Automated checks documented in GA4 source of truth:

Flag when platform conversions exceed GA4 by > tolerance
Flag CRM qualified leads not matching GA4 lead events (tracking gap)

Reporting Agent

Schedules report generation
Detects anomalies (spend spike, conversion drop)
Builds optimization changelog and plan drift sections from optimization_log + execution_manifest
Proposes plan revise, full replan, or optimization investigation tickets — does not auto-apply
After plan vN+1 approved: rebaseline all "vs plan" metrics to new version

Report ↔ plan ↔ optimization data flow

Delivery channels

Kobi client dashboard (primary)
Email PDF/export (optional)
API for parent Kobi platform widgets

Human touchpoints

Narrative review before client send (configurable auto-send)
Investigation when reconciliation fails

Cost note (LLM)

Scheduled reports — daily digest, weekly narrative, monthly executive, per-track drift — are generated via Vertex Batch API (50% discount) since they tolerate async completion. Anomaly alerts are real-time (<1h SLA) and run at standard rates. See Agentic orchestration — Batch API for scheduled tasks.

SLA

Weekly client report: Monday 09:00 tenant local (configurable; batch generated overnight)
Anomaly alert: < 1 hour from detection (real-time, not batched)