Advertising Campaign Analysis That Reflects Real Revenue
Why platform-reported ROAS misleads US advertisers, and how to reconcile ad spend against actual orders, CAC, and MER with clean GA4 and server-side data

Image via 123RF
In This Article
Platform ROAS Is Not Truth
Attribution Is a Design Choice
Fix the Data Pipeline First
Why Most Advertising Campaign Analysis Reports the Wrong Number
If your campaign report is built from the ad platform's own dashboard, it tells a story about modeled conversions, not orders. Advertising campaign analysis only becomes useful when reconciled against revenue that actually cleared in Shopify or Stripe - not the events a pixel claimed to see. This piece walks through five myths that corrupt post-campaign analysis for US advertisers, then shows the reconciliation model Prebo Digital uses to report channel performance with numbers finance will accept.
Myth 1: Platform-Reported ROAS Equals Real Incremental Return
A platform ROAS of 4.0 tells you what the platform attributed to itself under its own model and lookback window, after its modeling filled in missing events. It does not tell you what would have happened without the spend. Meta, Google, and TikTok each claim credit for conversions that may have occurred anyway - branded search, returning customers, email-driven visits.
Reported ROAS is a directional signal, not a P&L line. Treat it as the top of a reconciliation stack, never the bottom.
What Actually Breaks Pixel-Based Measurement in the US Market
Three mechanisms do most of the damage, and none are the marketer's fault:
- App Tracking Transparency and ITP. Apple's ATT framework and Intelligent Tracking Prevention shorten cookie lifespans, so a same-session pixel fire can miss a later purchase.
- Consent and opt-out signals. For brands selling into the EEA/UK, Google Consent Mode shifts conversions into modeled territory; US state privacy laws such as CCPA/CPRA add opt-out-driven gaps in pixel firing.
- Modeled fill-in. Platforms backfill missing events with modeled conversions - useful for optimization, dangerous for reporting.
Myth 2: More Pixels and Dashboards Produce Better Campaign Analysis
Adding a third dashboard does not add truth. If five tools disagree, you have five guesses. The bottleneck is almost always the event taxonomy - how purchase, add_to_cart, and begin_checkout are named, parameterized, and fired - not the visualization layer.
Myth 3: Last-Click Attribution Is the Objective View
Last-click feels objective because it names exactly one touchpoint. But GA4's data-driven attribution is the default for a reason: it distributes credit across touchpoints based on observed paths. Last-click systematically starves top-of-funnel programs - often the ones buying the audience the retargeting layer harvests.
Myth 4: One ROAS Target Fits Every Channel and Funnel Stage
A prospecting campaign and a branded-search campaign cannot share a target. Branded search always looks like a hero under any attribution model; prospecting always looks expensive on click-level ROAS. Judging them against the same number guarantees you cut the demand-creation layer that makes the rest of the funnel work.
From Reported Conversions to Reconciled Truth: A Three-Layer Model
Prebo Digital reconciles campaign analysis in three layers, each with a different job:
- Layer 1 - Platform. Reported conversions and ROAS. Use for optimization signal.
- Layer 2 - Analytics. GA4 events, server-side deduplication, and attribution models. Use for path analysis and channel comparison.
- Layer 3 - Back-end revenue. Shopify orders, Stripe charges, refunds, and new-vs-returning splits. Use for CAC, MER, and contribution margin.
Only Layer 3 is finance-grade. A good analyst explains the gaps between layers - and makes each gap smaller over time.
Layer Comparison: Platform, Analytics, and Back-End Revenue Views
| Layer | What it answers | Blind spot |
|---|---|---|
| Platform (Meta, Google, TikTok) | Is the optimisation signal healthy? | Modeled conversions, self-attribution |
| GA4 | How do paths distribute credit? | Cookie loss, consent gaps |
| Shopify / Stripe | What revenue actually cleared? | No channel-level attribution |
Myth 5: Incrementality Is Too Expensive to Test
Holdout testing and geo-lift experiments are cheap relative to scaling a channel that was never incremental. A two- or three-week channel pause in a single state, or a 5-10% holdout on a retargeting audience, often produces a clearer read than a year of platform reporting. If you cannot pause a channel without losing the revenue it claims to produce, you have your answer.
Build the Event Taxonomy Before You Build the Dashboard
Lock naming and parameters first. A consistent purchase event with value, currency, order_id, and event_id beats three clever dashboards layered on inconsistent events.
Document every event: trigger, parameters, destination (GA4, Meta, TikTok, warehouse), and owner. Checkout extensibility on modern Shopify themes changes when and how purchase events fire - validate in a test order before trusting any report built on top.
How to Deduplicate Server-Side Events (GTM Server Containers and Meta CAPI)
When you run both a browser pixel and the Meta Conversions API from a GTM server container, Meta needs a shared event_id to drop duplicates. Miss it and your purchase count doubles.
// Illustrative Meta CAPI payload with dedupe
{
"data": [{
"event_name": "Purchase",
"event_time": 1730419200,
"event_id": "ord_10482_purchase", // must match browser pixel
"action_source": "website",
"custom_data": {
"value": 1450.00,
"currency": "ZAR", // label as estimate in US reporting
"order_id": "10482"
}
}]
}
Reconciling Ad Spend to Shopify and Stripe Revenue With ETL
Pull ad spend daily per channel into a warehouse table, join it to Shopify orders and Stripe charges on order ID and date, and compute MER as total revenue divided by total ad spend over a rolling window. This join is where most reporting breaks - currency mismatches, timezone drift, refunds, and discount codes all distort the ratio. Clean it once, in a warehouse, using an ETL and analytics pipeline, and every downstream report inherits the accuracy.
Which Reporting Model Fits Your Decision: MER, CAC, or Contribution Margin
| Model | Question it answers | Blind spot |
|---|---|---|
| Platform ROAS | Is the optimisation signal healthy? | Self-attribution, modeled events |
| GA4 attribution | How do paths distribute credit? | Consent gaps, cookie decay |
| MER | Is total spend efficient vs total revenue? | Mixes channels, hides channel bloat |
| Incrementality | What did the spend actually create? | Slow, noisy on small budgets |
A Quarterly Campaign Analysis Cadence: Strategy to Report
Run analysis on a fixed cadence: weekly platform health checks, monthly GA4-vs-Shopify gap review, quarterly MER and CAC read with holdout results folded in. Bring finance into the quarterly review so the reporting model is agreed in advance, not argued post-hoc. See how this applies to your storefront by reviewing the Prebo Digital approach.
What to Look For in an Analytics or Attribution Partner
Ask three questions: Can they show an event taxonomy document from a live client? Can they demonstrate server-side deduplication across Meta and GA4? Can they join ad spend to back-end orders in a warehouse and show the reconciliation logic? If any answer is "we'll set that up later," the partner will report the same misleading numbers you already have.
Sources
- Google Analytics 4: Attribution models and cross-channel reporting
- Google Ads Help: About Consent Mode
- Meta for Developers: Conversions API
- Google Tag Platform: Server-side tagging
- Apple Developer: User privacy and data use (App Tracking Transparency)
- California Attorney General: California Consumer Privacy Act (CCPA)

