PPC Campaign Optimization: Fix the Measurement Before You Touch Bids
A myths-vs-facts operating guide for US performance marketers who need optimisation decisions built on trustworthy conversion data - not platform-reported ROAS.

Image via 123RF
In This Article
Measurement Before Optimisation
Automation Needs Guardrails
Consolidate and Test Incrementally
Why Most PPC Campaign Optimization Programmes Optimise the Wrong Number
Most accounts we audit at Prebo Digital do not have a bidding problem. They have a measurement problem. The reported conversion count in Google Ads, Microsoft Ads, Meta and TikTok is a modelled estimate produced by the same system whose performance you are judging. Optimising bids against that number is optimising against a scoreboard the platform keeps for itself. Here are five common assumptions US performance marketers make, the evidence against them, and the fixes we run before touching a single bid.
Myth 1: "Platform-Reported ROAS Is My True Return"
A dashboard showing 4.2x ROAS feels definitive. It is not. Platform conversions include modelled figures where consent or cookie signals are missing, view-through windows running several days, and cross-campaign double counting when one user touches Performance Max and a branded search ad in the same session.
The Fact: Platform Conversions Overlap, Model and Ignore Cost
Three distortions compound: attribution overlap (Google Ads and Meta both claim the same Shopify order), modelling (platforms fill consent gaps with estimated conversions), and COGS blindness (no platform knows your margin, returns or shipping, so revenue attribution is not profit attribution).
Warning: If your platform ROAS and your bank account disagree, the platform number is the one that is wrong for decision-making. Treat it as a directional signal, not a P&L line.
The Action: Build a Monthly Three-Way Reconciliation
Monthly, reconcile platform-reported conversions, GA4 purchase events and backend orders from Shopify, WooCommerce or your OMS. Variance in US accounts we review runs 15% to 40%, driven mostly by consent gaps and returns. Report blended MER (total revenue ÷ total marketing spend) alongside channel ROAS so leadership sees a number attribution overlap cannot inflate.
| Month | Platform Reported | GA4 Purchases | Backend Orders | Blended MER |
|---|---|---|---|---|
| Jan (example) | 1,240 | 1,080 | 1,015 | 2.9x |
| Feb (example) | 1,310 | 1,120 | 1,060 | 3.1x |
Myth 2: "Turn On Smart Bidding and Walk Away"
Automated bidding is not a strategy. It is a learning system that reflects the quality of the conversion signal and values you feed it. Google's documentation on Smart Bidding is explicit that it optimises toward the conversion actions and values you define. If your purchase event fires on every checkout initiation, or you pass gross revenue instead of margin, the algorithm will faithfully scale low-quality conversions.
How Automated Bidding Actually Learns (and Where It Breaks)
It learns from conversion events and their values within your budget, tROAS or tCPA constraints. It breaks in three predictable ways: a mislabelled conversion action, values that ignore margin, and a Performance Max campaign quietly harvesting branded queries and claiming cheap conversions as its own.
Tip: Pass margin-adjusted values via your dataLayer, not gross revenue. A $120 order with 40% margin should pass roughly $48, not $120.
window.dataLayer = window.dataLayer || [];
window.dataLayer.push({
event: 'purchase',
value: 48.00, // margin-adjusted
gross_revenue: 120.00,
currency: 'USD',
items: [{ item_id: 'SKU-100', item_margin: 48.00 }]
});
Myth 3: "More Keywords and More Campaigns Mean More Growth"
Fragmentation feels like control. In practice, splitting spend across dozens of small ad groups starves each of the conversion volume an automated bidder needs to learn. Five campaigns at $1,000/month each give the algorithm a fraction of the signal one consolidated campaign at $5,000/month provides, producing longer learning periods, noisier CPA and slower reaction to seasonality.
Why Consolidation Usually Beats Fragmentation at Scale
When we consolidate US accounts, the pattern repeats: fewer ad groups with tighter themes produce cheaper conversions within two to three weeks, not because the creative changed but because each bidding unit now has enough signal to learn. Consolidate anything below roughly 30 conversions per month into a broader themed ad group, then let query data decide where to split again.
Query-to-Intent Mapping: The Pruning Cadence That Moves CPA
The search terms report is the cheapest research tool in your account. We review it weekly above roughly $10k/month in spend and fortnightly below that. Every query gets one of three labels: add as exact keyword, add as negative, or leave. Queries that spent more than the target CPA over 30 days with zero conversions get paused or negated; queries revealing a mismatched landing page get routed to a new URL with matching message.
Myth 4: "Attribution Is a Reporting Detail, Not an Optimisation Lever"
Attribution decides which segment gets more budget. If the model attributes branded clicks to prospecting, you will scale prospecting and wonder why incremental revenue does not follow. Attribution is an input to optimisation, not a footnote.
US Consent, Cookie Loss and What It Does to Your Optimisation Inputs
Under California's CPRA and similar state laws, consent signals gate a meaningful share of observations. Google's help centre on Consent Mode and GA4 behavioural modelling explain how platforms fill these gaps. The fix is Consent Mode v2 plus server-side tagging that captures order events regardless of cookie state. Run a two-week event QA before any spend increase.
Info: A mislabelled purchase event can waste more budget in a week than a badly written ad ever will. Audit the tracking before you touch bids.
Myth 5: "Creative Is the Last Thing to Optimise"
In Performance Max, Meta and TikTok, creative functions as targeting. Post-ATT, the algorithm uses creative signals to find audiences. Treat creative iteration as a structured programme, not a single refresh before Q4: test new angles on a two-week cycle and keep a rolling log of winners by format.
Building a Weekly and Monthly Optimisation Cadence That Holds Up
Optimisation is a schedule, not a mood. This is the cadence we run for accounts between $10k and $150k monthly spend.
| Frequency | Task | Output |
|---|---|---|
| Daily | Budget pacing and anomaly check | Slack alert to owner |
| Weekly | Search terms review and negatives | Updated negative lists |
| Monthly | Three-way reconciliation and MER review | Revised budget allocation |
Optimising for Profit: Margin-Adjusted Values and Blended MER
Gross revenue rewards low-margin SKUs; margin-adjusted conversion values reward profit. Combine value-based bidding with a blended MER ceiling so no single channel claims more budget than the business can support. For a worked example on a Shopify catalogue, see our team's approach.
Testing for Incrementality Instead of Platform A/B Theatre
Platform-served A/B splits share audience and learning, which contaminates results. Run geo holdout tests instead: hold spend constant in a matched control market while changing the variable in the test market, then compare backend revenue. On a recent US homeware account, a 14-day geo holdout showed brand-search spend produced almost no incremental revenue, freeing budget for non-brand acquisition.

