What complete data means
The technical difference between what GA4 measures (≈13% in the EU) and what you need to defend a number to your CFO.
"Complete data" is one of those phrases that sounds obvious until you try to define it. This chapter draws the line — what we mean when we say SealMetrics captures 100%, what GA4 means when it shows the dashboard you've been reading, and why the difference is a budget problem, not a vocabulary one.
Written for measurement leads, performance marketers, and anyone who has to defend a conversion number to finance.
Where the 13% comes from
The 13% figure is not a slogan. It is the output of a three-stage cascade that every analytics tool relying on client-side cookies pays in full.
Start with 100 real EU visitors arriving at your site. Each stage applies to the survivors of the previous one — the cascade is multiplicative, not additive.
real EU visitors. Every person who actually loaded the page.
after consent rejection. Average rejection rate across EU enterprise CMPs is ~55% — outright rejects, banner closures, and reject-all preferences combined.
after ad blockers. Of those who accepted, roughly 40% run uBlock, AdGuard, Brave Shields, or Pi-hole. Analytics requests are stripped silently — the user sees your site, GA4 does not see the user.
remain in GA4. Safari ITP, Firefox Total Cookie Protection, and Brave's defaults expire or block the cookies GA4 needs to keep sessions intact. Of the remaining 27, roughly half use browsers that interfere materially.
Which is why "we'll improve our consent UX" doesn't recover the loss — it only narrows the first stage. Even a perfect 80% accept rate leaves you at roughly 20% of real traffic after the other two stages apply.
The 13% is not a worst-case scenario or an edge case. It is the median outcome for European eCommerce sites with proper CMP implementations. You can verify the math on your own traffic using our data loss calculator.
Cascade percentages from SealMetrics measurement studies of EU enterprise CMPs. Methodology in our published analysis. Last revised May 2026.
Sampling vs complete measurement
The 13% is what reaches the dashboard. The dashboard, in turn, samples it.
GA4 applies sampling when a query exceeds certain thresholds — by default, 10 million events per query in the standard tier. Above that, it analyses a subset of events and extrapolates the result. The report is presented as a measurement. It is an estimate of a measurement.
The sampling rate is not always surfaced. Some reports show a "sampled data" warning; many do not. For high-volume eCommerce sites with deep filters, the analysed ratio can drop below 1% of events on complex queries — a sample of a sample.
"What appears as a 1.3% conversion rate is a calculation on a sample of a fraction of the real population. Defending that number to a CFO is a coin toss dressed as analytics."
SealMetrics does not sample. Every event lands in ClickHouse at full resolution. A query against 4 billion events queries 4 billion events. We pay for the storage and the compute so that the answer matches the question.
Modeling vs measuring
When GA4 cannot measure — because consent was rejected, the ad blocker stripped the request, the cookie expired — it does not leave the cell empty. It models.
"Modelled conversions", "behavioural modelling", "consent mode v2": all variants of the same mechanism. A regression trained on the events GA4 can see predicts the events it cannot. The output is reported as a number, indistinguishable in the UI from anything that was actually observed.
Modelling has legitimate uses — directional analysis, time-series smoothing, exploratory work. Optimising a media budget against modelled conversions is not one of them. The bidding algorithm interprets a modelled conversion as a real one. The budget shifts toward whichever channel scored highest in the model. The channel that drove the actual converter is, by definition, the one whose data was missing — and the budget never reaches it.
Complete means measured.
Sampled is a degradation. Modelled is an inference. Each step away from measured is a step further from a decision being defensible. Wherever the phrase "complete data" appears in Open, it means: measured server-side, before any of the three losses fires, on the full population.
Where SealMetrics sits
Three tiers operate in analytics today. The category matters because comparisons across tiers are dishonest — and most published comparisons quietly are.
Enterprise analytics
GA360, Adobe Analytics, Piwik PRO, SealMetrics
Six- and seven-figure decisions. Complete data, compliance posture, and audit trails are non-negotiable.
Lightweight privacy analytics
Plausible, Fathom, Simple Analytics, Umami
€9–50/month. Page-level metrics, beautiful UIs, zero PII. Excellent at what they do — and not built for revenue attribution or billion-event datasets.
Free with sampling
GA4 default
Free at the entry tier. Constraints (consent loss, sampling, modeling) compound at scale.
SealMetrics competes in Tier 1. The lightweight tools are not competitors — they are the right choice for a different problem. If you are evaluating SealMetrics, you are not choosing between us and Plausible. You are choosing between complete data and the budget consequences of the alternative.
The rest of Open builds on this definition. Chapter 05 describes the infrastructure that makes it possible. Chapter 09 lists what we will not do with the data we collect. The chapter on GDPR compliance — and the one on how the pixel works internally — are in progress; in the meantime, the legal grounding lives at /security.
Continue with chapter 05 — Architecture and performance.
Was this chapter useful? Tell us →