Who decides in your marketing — and who is responsible?
We build the system behind your marketing decisions — and then we stay with it.
Google Ads · Meta Ads · GA4 · Google Tag Manager · Databox · ActiveCampaign · Semrush
There are two decision-makers in your company.
You have never spoken to one of them.
Google and Meta decide on what you hand them: the conversion values, the feed fields, the signals that arrive or do not. Not on your brief, and not on what you meant.
Four systems counted last month. You got four numbers.
Google says 1,000 purchases, Meta says 10,000, GA4 says 2,850, your own webshop says 2,500. Only one of those is where the money actually changed hands. And each platform optimises on its own count, so a wrong number is not a wrong report — it is Google and Meta buying more of the wrong thing, accurately, every day.
ChatGPT answers with your sector average.
Ask it about your company and what mostly comes back is the median for your sector — and it is most confident exactly where you were different. NP Digital ran 600 prompts across six models in February 2026: the best of them got brand facts right 59.7% of the time.
You are the one who has to explain it.
The owner asks you, not Google. And you cannot ask the Google Ads model what it decided on; the same goes for Meta. So what helps is a record: what was decided, on what basis, and which of your numbers it stood on.
We look at it, we build it, and we go back to see what it did.
Audit
A conversion event is never wrong on its own. It is wrong against what somebody meant to count — and usually nobody wrote that down. So we establish that first, and check against it after.
What we need is Viewer rights in GA4, read access in Google Tag Manager, and the Google Ads and Meta accounts. No admin rights, no developer, nothing changed while we look, and you can revoke it the day it is done. What comes back is the findings, each one with the thing it rests on next to it.
Build
Three things get built, and none of them is a dashboard. The measurement, matched to a specification and written down, so a number means the same thing next quarter. Your brand, your guides and your decisions in one place an AI can read. And a record where every decision carries what was expected of it.
This is the part that stays after we leave. It is also what decides whether your own AI answers out of your company or out of the general internet.
The feedback loop
We once found a stopped in-store campaign a week after it stopped. Weekly, that is the next day.
One half is the watching: it says something when a campaign stops, when measurement drifts, when the feed drops out of a campaign, when the conversion setting wanders from what the bidding is chasing. The other half closes the loop. Every decision went in with what was expected of it; we read back what actually happened, and that goes into the record. So the next decision starts from what the last one did, not from what somebody remembers. It is the only part of this worth more every year — and it is pointed at our own work too, not only at yours.
And if you want to run it yourself, there is a path for that: further up, your own Claude Code or Codex connects to our system over MCP.
What we do not promise.
We tell you which numbers can carry a decision.
How much better it gets depends on campaigns we do not run — anybody who promises you a percentage is guessing. What you get is which of your numbers holds up under the decision you want to make, and what has to be repaired first.
A finding is about what is set, not about who set it.
A setting is not a fault until we have asked whether it was deliberate. The answer often closes the question — and when it does not, your agency gets something it can work with too.
Every finding shows you where it came from.
Which campaign, which setting, which number. Check it yourself, or hand it to anybody who will.
Twenty-five years of marketing and sales decisions.
Hundreds of clients, thousands of separate jobs, from practice and not from books. Few teams in Hungary have been doing search optimisation and search advertising this long.
Spend twenty-five years inside one company and you learn that company very well. We spent them across hundreds, which means we can usually tell you which of your problems is yours alone and which one almost everybody has. The second kind is the one with a known answer.
The method and the ten levels →We read what you already run.
We do not replace your stack. We ask for read access, and every finding carries where it came from and when.
Google Ads
Paid searchCampaigns, search terms, quality signals and what the bidding is actually optimising for.
Google Tag Manager
MeasurementWhat fires, when, and on whose consent. Server-side containers included.
GA4
AnalyticsWhat visitors do, against what somebody meant to count.
Meta Ads
Paid socialCampaigns, creative and delivery, read the same way and against the same checklist.
Databox
ReportingWhere the numbers are already collected, so we read them rather than rebuild them.
Semrush
MarketWhat the market does, not what it says about itself. Limits stated as properties of the method.
The parts you run yourself are not open yet.
Leave an email address and you hear from us when they are. The audit, meanwhile, is available now.