Amazon, Google, and Meta captured 59% of US ad spend in 2025, up from 47% in 2020. Each platform will happily tell you it drove the conversion. Add up their self-reported numbers and you'll count the same sale three times. The math doesn't work because platform dashboards are advocacy tools, not measurement tools.
That distinction matters more right now than it did even two years ago. Google retired rule-based attribution models in GA4 (goodbye, First Click, Linear, Time Decay, Position-Based) and defaulted to data-driven attribution. Meta rebuilt its ad ranking on LLM-scale systems, growing core ad revenue 33% year-over-year in Q1 2025. Google's AI Overviews compressed paid CTRs by roughly 68% on AI-triggering searches. The platforms are changing faster than most measurement setups can keep pace with.
So the question from Navah Hopkins's recent Ask A PPC column is the right one: how do you build a framework that compares performance across Google, Microsoft, Meta, and Amazon fairly? Here's the ops-oriented version of that answer.
Step 1: Audit Your Tracking Foundation (Before You Touch Attribution)
Skip this and everything downstream is contaminated. You need to confirm that conversion tracking fires accurately on every platform, that events are consistent (same naming, same trigger conditions), and that you're running a centralized tag management approach. Third-party cookies are unreliable enough now that server-side tagging and enhanced conversions using hashed first-party data aren't optional anymore. They're the baseline.
Run it this week: Pull one week of conversion data from each platform. Compare totals against your CRM or analytics platform. If the gap exceeds 15%, stop optimizing campaigns and fix tracking first. Use platform diagnostics plus a tool like Microsoft Clarity to verify that reported conversions match real user behavior.
Guardrail: If confidence in tracking is low, use data exclusion tools to remove low-confidence periods before making any budget decisions.
Step 2: Build an Independent Measurement Layer Above the Platforms
This is the single most important structural change. Platform-reported metrics are fine for in-platform optimization (bid adjustments, creative rotation, audience refinement). They should never be the source of truth for cross-platform budget allocation.
What you need: a unified measurement layer with a single dataset and consistent methodology sitting above all platform reporting. Think of reporting as a three-layer pyramid. Executive metrics at the top (net revenue impact, CAC, ROI). Operational metrics in the middle (campaign and channel performance, qualified pipeline). Tactical platform-specific metrics at the base for execution teams.
The key word is net revenue. Account for returns, refunds, and churn. Gross revenue flatters; net revenue tells the truth.
Governance matters here. Centralized UTM conventions and standardized tracking codes across every platform. Without this, your ops team spends half its time reconciling conflicting reports instead of actually analyzing performance. Document the conventions, enforce them, and audit quarterly.
Step 3: Use Attribution Directionally, Not as Gospel
Multi-touch attribution attempts to assign value across the full customer journey rather than crediting the last click. That's better than single-touch models. But MTA still runs on platform-reported data, which means it inherits platform bias.
The more rigorous alternative: incrementality testing. Holdout groups, geo-based experiments, or on/off tests that measure true lift by channel. This is where you get causal signal instead of correlation.
The trade-off is real, though. Incrementality tests require volume, time, and statistical rigor. Most B2B SaaS teams can't run clean holdout tests on every channel every quarter. So the practical move is to use MTA for directional reads on day-to-day optimization and run incrementality tests on your two or three largest spend channels once or twice a year.
The hypothesis (make it falsifiable): If we pause Channel X for 4 weeks in a geo-holdout region, then pipeline from that region will decline by at least 15%, because Channel X is generating incremental demand rather than capturing existing intent.
Success = statistically significant lift difference between holdout and control. Guardrail = monitor total pipeline in holdout region weekly. Stop-loss = if pipeline drops more than 30% in week 2, end the test early and restore spend.
Step 4: Layer in Human Signal (the Most Underused Data Source)
Platform data tells you what happened. It rarely tells you why. Ask customers how they found you. Ask sales what they're hearing about lead quality by source. Keep CRM source tracking accurate and align sales and marketing on lead attribution standards.
These conversations regularly surface gaps that dashboards miss. A prospect converts through Google but associates their discovery with a LinkedIn post from three weeks earlier. That perception matters for budget allocation even if no pixel captured it.
Don't treat human feedback as anecdotal filler. Operationalize it: add a "how did you hear about us" field, review it monthly, and cross-reference against platform-reported attribution. The delta between the two is where your measurement framework gets honest.
What to Measure (and What Not to Over-Interpret)
Primary metrics: CAC, ROAS (benchmark: 2.5–4× at scale; >4× for retargeting), net revenue per channel, qualified pipeline.
Leading indicators worth watching: on-site engagement quality, conversion rate by platform (benchmarks: Google Search 4–8%, Google Shopping 0.8–1.6%, Meta 0.9–1.8%), creative fatigue signals.
What to deprioritize: impressions, reach, and raw CTR. Google's AI Overviews alone compressed paid CTRs by ~68% on affected searches. A CTR drop might mean your ads are failing, or it might mean Google changed the SERP. Without downstream outcome data, you can't tell the difference.
Hopkins's original column made a point worth repeating: the goal isn't to crown a single winning platform. The goal is to reflect how users actually move through the funnel, then allocate budget based on where incremental value lives. Every platform will claim credit. Your measurement layer is the one that doesn't need to.
