Notes.
Short pieces on what we learn while running accounts — tracking, attribution and the numbers behind the cases.
Brand search terms: the four-bucket method
"Bid on your brand" and "never bid on your brand" are both lazy answers. We sort every brand search term into four buckets, and then we know exactly what to do with it.
Read more →What your attribution is still worth under Consent Mode V2
Under denied consent Google Ads no longer sees a cookie, but it does see a cookieless signal it models into an estimated conversion. What can you still trust? Directionally yes, precisely no.
Read more →Your CPL is lying — and your agency probably doesn't know it
The cost-per-lead in your monthly report is almost never the number you think it is. Usually because Google Ads conversions and GA4 events have been mixed together.
Read more →'Call tracking doesn't work in the Netherlands' — and what it costs your bidding
Google's call-forwarding numbers don't work in the Netherlands, so most agencies stop measuring phone calls altogether. On an account where the phone is the business, that means you're bidding blind. There's an honest proxy.
Read more →One senior plus AI, or a team of juniors
At most agencies your account runs on a junior with a spreadsheet and a checklist. We run it the other way around: one senior making the decisions, with AI doing the work that would otherwise land on that junior. Why that works better, and where the limit is.
Read more →Smart Bidding is only as good as the signal you feed it
Most Smart Bidding debates are about the knobs: tCPA, target ROAS, budget. But the bidder optimises toward the conversions it can see. Feed it the wrong signal and you've got a strong algorithm running efficiently at the wrong target.
Read more →What a free audit finds every single time
We run a lot of free audits. Four things come up so often we'd bet money on them, and all four cost you money without you noticing it.
Read more →When "CPL fell 87%" falls apart in 30 minutes
We rewrote every case-page and reconciled the numbers against the source data. Most didn't survive 30 minutes of scrutiny — here's the honest version.
Read more →