Marek Gierat, Analytics, BI & Automation
Snapshot
In marketing since 2012, starting in e-commerce and marketing at a manufacturer, where I still run both today. Agency analytics and reporting since July 2022. I do not buy media. I own the layer above it: measurement, reporting, pacing dashboards and audits, which the team and the client use to decide where the money goes. On the data side I work with Google Ads, Meta Ads, Microsoft Ads, DV360, TikTok, Pinterest and Reddit.
What I do
I build the measurement layer (GA4, GTM, UTMs, dataLayer), bring platform and backend data into one place (BigQuery, Google Sheets, pacing dashboards), then audit whether it all adds up to the penny. Plus automated reporting, so the same report stops eating three days of someone's month.
Measurement audits (GA4 and GTM)
What your data does not measure.
What it means for you: You find out before you make a decision worth hundreds of thousands on it.
Pacing dashboards (Sheets, BigQuery, platform APIs)
Spend, revenue and progress to target every morning.
What it means for you: No more learning on the 5th of next month that the month was already lost.
Cross-validation
Platform against backend against warehouse.
What it means for you: One number instead of three that disagree, and you know why that one is true.
Automated reporting (APIs, scripts, AI)
The report someone assembled by hand for two days builds itself.
What it means for you: The team goes back to producing results instead of copying numbers.
Data into decisions
Not a chart, a sentence.
What it means for you: "This stopped working on 30 June, the cause is here, the move is this."
When I use which part of the stack
For me it is not "Meta or TikTok". It is a measurement stack I put in for almost every client. What changes is how deep I go.
GA4 and GTM
Always. Basic hygiene. Without it, the rest is guesswork.
A dashboard in Sheets
Always. Decisions happen daily, and the whole team and the client need to see them without logging in to five tools.
BigQuery
Almost always, because the questions quickly outgrow the GA4 interface: order-level data, cohorts, year on year across hundreds of thousands of orders.
Triple Whale, Adnova, Supermetrics
It depends. A ready-made layer where building our own makes no sense, and only after checking its numbers against the backend.
Who calls me
Industries
Hot takes
Hot take 1
Attribution is not measurement
It is a story the platform tells about itself. For years I watched teams optimise to the ROAS in the ads manager, happy, while the margin went down. Until you hold that number against the backend and against incrementality, it is not a result. It is a compliment the tool pays itself.
Hot take 2
An unchecked dashboard is worse than none
No report makes people ask questions. A pretty report with a silent error puts everyone to sleep for half a year, until someone notices the purchase event has been at zero since May.
Hot take 3
Most "analytics" is cleaning
Not modelling, not machine learning. Checking that the time zone matches, that UTMs are not written in three conventions, and that nobody pasted a phone number into a URL parameter. Not a line that wins applause at conferences.
How I work
- 1
First checks, always in this order
Does the key conversion have values across 12 months, not 90 days? Does the platform match the backend? What is really in the published GTM container, not the export? UTM and dataLayer hygiene, including personal data leaking into parameters. Time zone, currency and attribution model. Brand against non-brand.
- 2
After the first week
A list of what is broken in the measurement, with proof for each point. Not opinions: "this event has been at zero since 14 May, here is the screenshot". Plus a split: what we fix now, and what needs a decision on your side.
- 3
After the first month
Working measurement of the key events, one place to see spend and results daily, and the first report whose numbers match the backend to the penny. From then on, the conversation is about what to do with the data, not whether it is true.
Tools
Claude Code
Most of my work is not clicking in interfaces, it is writing something that does the same job next time without me.
Google Ads API
Data the interface will not give you: brand against non-brand, marginal ROAS, the daily split.
BigQuery
When the question does not fit in GA4, and as an independent source to check the spreadsheet against.
Google Sheets, Apps Script, Sheets API
The one tool every client opens without training. Everything else ends up here.
GA4 and GTM
The starting point of every audit, including the published container, not only the export.
Supermetrics, DataSlayer
Quick live platform data in a tab.
Triple Whale, Adnova
A ready analytics layer where building our own makes no sense.
Google Trends
Search demand as context for campaign results.
ClickUp, GitHub, Fireflies
So decisions do not get lost between the meeting and the implementation.
Things I have built for the team: a catalogue of all the team's AI skills that refreshes itself every morning; a standard for Sheets dashboards, packaged so a dashboard from anyone in the agency looks and works the same; a recurring 14-day Meta Ads report with a naming-convention audit; a weekly Google Ads report straight from the API; a creative concept generator working from an audience × angle × product matrix, with a human approving in the sheet; and a pacing dashboard with 60 charts, year-to-date and year-on-year tabs and daily spend plans. All of it runs through the team's API gateway: no keys on laptops. One rule we do not break: a conclusion is never published as a finding unless the system was actually queried.
Lessons
The costliest mistake
Restyling charts in a client's live dashboard, I filtered tabs by name prefix instead of exact name. The filter caught colleagues' tabs too. No data lost, but rolling the sheet back also undid my own formulas, which had to be rebuilt, on a sheet the client used that same day. Since then: exact names only on anyone else's work; version history works on the whole file, not one tab; and before I touch anything live, I write down what I am changing and where. The list takes two minutes. Undoing it took half a day.
Favourite metric
Marginal ROAS: what the last dollar added, not the average of all of them. The average can look fine while the last 20% of the budget is already losing money. Close behind: CPA split by brand and non-brand, because only that split says if the business is really growing.
Most overrated
ROAS from the ad platform. That is not measurement, it is a self-assessment. Close behind: CTR, the metric everyone reports and nobody has ever made a decision worth money on.
Quick tip
Open GA4, set the range to 12 months (not 90 days) and check, day by day, that your most important conversion has values the whole time. You are not looking for a dip; you are looking for a zero or a flat line. Twelve months will show you the exact day something broke, and with a date you usually know within an hour what caused it.
Off the clock
I act, I take photographs and I build dashboards. Each time the point is the same: the audience should understand what just happened.
- Theatre. The stage, not a hobby on the side: several theatre companies, outdoor shows included. I also founded a small chamber stage in Kraków, and trained as a musical theatre actor.
- Photography. I run my own studio: photo shoots and film sets. Which is why a dashboard has to add up and look good.
- Mountains. I walk them as often as I can.
FAQ
Do you buy media?
No. I own the measurement and reporting layer above the spend: dashboards, audits and the numbers the team and the client decide on.
What do I get after the first week?
A list of what is broken in your measurement, with proof for every point, and a split between what we fix now and what needs your decision.
Which tools do you set up for every client?
GA4 and GTM, and a dashboard in Google Sheets. BigQuery almost always. Ready-made tools such as Triple Whale only after checking their numbers against your backend.
Why 12 months and not 90 days?
A 90-day window cannot tell a structural error from an outage six months ago, or from something already fixed. Twelve months shows the exact day it broke.
Next step.
Bring the two numbers that disagree. Book a call with Marek.