Table of contents
Most agency case studies are unfalsifiable. A percentage with no period, no currency and no conversion action named is not evidence — it is decoration. So when we rebuilt our own case studies, we re-pulled every number from the platform that generated it and threw out the ones that did not survive.
These are the nine Google Ads results that did. Each row links to the full page, which states its period, its currency, the exact conversion action counted, and anything that moved the wrong way.
| Client | Headline result | Period |
|---|---|---|
| Texas JDM Motors | Revenue $507,873 → $1,476,093; ROAS 6.97× → 13.61× | 12 months |
| JDM Engine Direct | $97,245 of spend → $1,421,603 tracked revenue (14.62×); +90% on matched months | 12 months |
| WorldRemit (UK app) | Spend scaled 19× to £1.82M; tracked ROAS 0.63× → 2.14×; 18,448 in-app purchases | 12 months |
| Arsenal Électrique | Leads 36 → 444; cost per lead CA$220 → CA$28 | 12 months |
| L'Exterminateur en Ligne | Conversions 1,089 → 2,357; ROAS 3.27× → 5.51× | 12 months |
| Living Beauty | Purchases +225% to 937; revenue +264% to CA$133,192 | 7 matched months |
| L'Entrepôt de la Réno | Tracked purchases 477 → 2,513; revenue CA$356k → CA$1.90M | 8 matched months |
| Nettoyage Danfour | Leads 323 → 449 on a flat budget; cost per lead CA$35 → CA$26 | 12 months |
| K&R Strategic Partners | 0 → 233 conversions and 734 tracked calls from a standing start | 5 months |
The pattern across all nine
Six of these nine accounts were already spending money when we took them over, and in five of those the first meaningful gain came from measurement rather than from bidding. An account that counts the wrong thing is not underperforming; it is following instructions. Arsenal Électrique had 16 tracked phone calls in a year in a business where almost every job starts with a call. The building-materials account had a page-view action recording hundreds of thousands of "conversions". The app account had two systems both counting the same purchase.
The second pattern is that efficiency and scale were not traded against each other. Three accounts here grew spend substantially — 49%, 42% and nineteen-fold — and improved return while doing it. That happens when budget increases are sized to what the learning systems can absorb and held until the new level proves itself, and it does not happen when budget is raised because the quarter is ending.
Where we quote less than you would expect
Two of these pages deliberately withhold a metric. Texas JDM Motors quotes revenue and return but no conversion counts, because a tracking change mid-year corrupted them. Nettoyage Danfour quotes leads and cost per lead but no revenue, because its lead values are modelled rather than invoiced. Both are stated on the page.
How these numbers were verified
Every figure on this page is repeated from a case-study page, and every one of those was re-pulled directly from the platform that produced it — the Google Ads API, the Meta Marketing API, or Google Search Console and GA4 — rather than from a dashboard screenshot or an internal report. Three rules were applied without exception.
One conversion action, named. Where an account contained several conversion actions, we picked the single action that represents a real sale or a real lead and ignored the rest, even where that produced a smaller number. One account in this set carries more than 900,000 recorded page-view "conversions"; none of that appears anywhere here.
One event alias. Meta returns several aliases for the same event. Summing them inflates purchases and leads by two to three times. Each figure here uses exactly one alias.
Matched windows only. A year-over-year percentage is only quoted where both years have data for every month compared. Where they do not, the result is presented as a launch with absolute totals instead. That rule caused us to retract three claims we had originally drafted, and the affected case studies say so on the page.
