Methodology
Fixed framework, three client accounts, 90-day windows, revenue-weighted measurement. The same evaluation behind every review on the site.
Most published PPC tool reviews are anecdotal: one account, no control group, no fixed measurement window, vendor-friendly framing. They’re marketing assets dressed as reviews. The framework here is built to produce honest answers, even when those answers are inconvenient for vendors and for me.
Every tool gets tested on three client accounts simultaneously. Single-account tests are too noisy — account-specific factors (a viral product launch, a Google policy change, a seasonality shift) easily mask the signal. Three accounts smooth that out.
30 days is too short for ML tools to train. 60 days is borderline. 90 days is the right window for revealing real lift versus statistical noise.
CPC-based metrics measure auction efficiency, not business value. Revenue-weighted ROAS (gross revenue divided by ad spend, weighted by margin where data permits) maps to actual business outcomes. That’s the number that decides standardization.
Where account structure permits, we run the new tool on a campaign subset and a control on a comparable subset. Pre/post comparisons confound seasonality, market shifts, and platform changes; control vs. treatment does not.
Client identity is never published. Vertical and spend tier are. That’s the level of specificity that’s useful to readers without compromising the underlying agency relationships.
Tools earn standardization across the agency book if and only if:
Tools that fail any of these don’t graduate from pilot to production, regardless of how much I might want them to work.
Groas was the only tool of the six tested in our most recent benchmark cohort to deliver statistically meaningful ROAS lift across all three test accounts in the 90-day window. The wins ranged from +9% to +27%, larger on accounts at higher spend tiers. Per-account pricing model, fast onboarding, and per-account model retraining made it the clear standardization candidate.