How We Measure & Verify Results

No single model tells the whole truth. We triangulate across three lenses and reconcile the differences rather than trusting any one in isolation.

Platform-reported attribution (Meta, Google, and similar) is fast and granular, but it tends to over-claim because each platform credits conversions to itself within its own window. We use it for in-flight optimization, not as the final word on contribution.

Multi-touch attribution distributes credit across the touchpoints in a customer journey. It gives a more balanced cross-channel view than last-click, but it is still a modeling choice and depends on the quality of the underlying tracking.

Incrementality answers the question the others cannot: what would have happened anyway. Where the stakes and budget justify it, incrementality testing is the standard we anchor to, because it measures the lift a channel actually caused rather than the conversions it merely touched.

Metrics only mean something when their definition is fixed. Here is how we define and measure the headline numbers you will see in our case studies.

Incrementality testing is how we separate marketing that caused a result from marketing that simply rode alongside one. The common methods we use, matched to budget and risk, include:

• Geo holdouts: suppressing or scaling spend in matched regions and comparing outcomes against control regions.
• Audience-level experiments: platform-native conversion lift studies that compare exposed and held-out audiences.
• Spend-step tests: deliberate increases or pauses, read against a forecast baseline to estimate the marginal contribution of additional budget.

Results are reported with the test design noted, so the reader knows whether a figure reflects measured lift or modeled attribution.

Attribution windows materially change a number, so we state them. Click and view windows are agreed at the start of an engagement and held constant for the duration, which is what makes period-over-period comparison meaningful.

For results that compound, we report a clearly labeled before period and after period rather than cherry-picking a single peak month. Seasonality and promotional spikes are called out where they materially affect a result.

Measurement is a discipline of approximation, not certainty. We think naming the limits is part of doing it well.

• Signal loss from privacy changes, cookie deprecation, and consent gating means tracked conversions undercount reality, and modeled conversions fill the gap imperfectly.
• Attribution assigns credit; it does not prove causation. Only controlled tests get close, and even those carry confidence intervals.
• External factors (brand momentum, pricing, PR, market shifts) influence results and cannot always be isolated from marketing effect.
• Case-study figures reflect a specific brand, period, and context. They are evidence of what is possible, not a guarantee of what any other brand will achieve.

When a case study cannot be cleanly measured, we say so. The goal is not the most impressive number, it is the most defensible one.