GA4’s Defaults Are Assumptions, Not Facts

A marketing lead defends a channel in the quarterly review using GA4’s own attribution report, confident the number is real. Three questions in, nobody in the room can say whether that credit came from a genuine model or from a fallback nobody ever flagged. The gap between confidence and verification is where the real audit starts, and it starts earlier than most teams think.

The Attribution Model You Think You’re Using Might Not Be Running

GA4 sets data-driven attribution as the default model for every property, crediting channels based on real conversion paths instead of a flat rule^[2]. The model needs volume to work, though, and Google’s own documentation does not publish exactly how much. Independent analytics consultancies report the practical threshold sits near 400 monthly conversions for a given key event, plus meaningful ad-interaction volume^[3]. A property under that bar reverts to last-click attribution automatically, with nothing in the interface saying so. Treat the exact figure as directionally reliable, not an official spec. Treat the silent fallback as real regardless, since it matches how the model has to work: no data, no machine learning, no choice but a simpler rule underneath.

A Series A SaaS team’s board deck showed data-driven attribution credit drifting toward organic search every month, and the shift read as an SEO win worth celebrating. A closer audit found their trial-signup key event sat well below the volume a data-driven model needs to run. The reports had been running last-click the entire quarter, quietly crediting whichever channel a lead touched last, dressed up in a dashboard still labeled data-driven.

The failure compounds because attribution feeds decisions that move real budget. A channel that looks strong under a silently reverted last-click model gets more spend next quarter. A channel that would have earned genuine multi-touch credit gets cut instead. Nobody on the team chose last-click. The conversion volume simply was not there, and GA4 never said so.

Does GA4 Count Customers, or Does It Count Clicks?

Google’s own documentation is explicit here, at least: “once per event” is the recommended default for any key event you create natively in GA4, and it counts every single trigger^[1]. Only key events migrated in from old Universal Analytics goals default to “once per session,” capping the count at one no matter how many times the event actually fires. Most teams never open that setting, and most never realize the two behave differently on the same click.

The default exists for a defensible reason. Two triggers of the same key event in one session are often genuinely different intents. Two demo requests from the same visitor a day apart are two real signals, and once-per-event preserves that distinction. The failure is not the default itself. It is nobody checking whether a specific key event, especially anything imported from Universal Analytics, is actually running the setting the team assumes it is.

A Series B team’s dashboard showed a 30% month-over-month jump in demo requests, and the news made it into a leadership meeting as a genuine win. Someone traced the raw events afterward and found a broken form validator. A meaningful share of submissions were double-firing, once on the initial click and again on a JavaScript retry after a validation error. Once corrected, the real lift was a fraction of the number that got celebrated. Nothing about that bug threw an error. It just quietly counted the same visitor twice.

The Deeper Assumption: One User, One Identity

Underneath both of the settings above sits a bigger assumption: that GA4 can reliably tell one visitor is one visitor, across devices and sessions. It cannot always, and the industry’s own research says the confidence problem is getting worse, not better, even as the tools improve.

GA4’s own identity resolution stacks three signals of decreasing reliability. First, a signed-in Google account through Google Signals, when available. Otherwise, device-based heuristics, or modeled statistical estimates for visitors who declined consent. Each layer down is a weaker assumption standing in for the layer above. Most B2B SaaS properties run mostly on the bottom two, since Google Signals coverage on a professional, work-device audience is thin to begin with.

In 2025, TransUnion and EMARKETER surveyed 196 US marketing professionals^[4]. 48% cite cross-channel deduplication issues as a top barrier to accurate measurement, and 55% name improving cross-channel attribution accuracy among their biggest challenges. 60% say internal stakeholders question the validity of their metrics at least sometimes. That is not a fringe complaint. That is a majority of marketers who do not fully trust the numbers they report on.

The scale of what identity errors can do to those numbers is larger than most teams assume. In 2026, LiveRamp and the Marketing + Media Alliance built a study on real impression logs: 1.9 million exposures across 147,941 users and four publishers^[5]. Dropping identity-resolution precision to just 50% collapsed measured ROI by roughly 70%, from a true return of $1.50 down to a measured $0.43. GA4’s identity model is a genuine assumption about who the visitor is, not a verified fact, and the cost of that assumption failing is not a rounding error. GA4 does not publish its own identity-resolution accuracy rate. No B2B SaaS marketing team can point to a number and say precisely how much of its reported ROI rests on a matched identity versus a heuristic guess. The LiveRamp findings are the closest available proxy for how large that risk can get.

A Series C team saw this play out directly. Paid and organic reports each independently claimed credit for the same signups, because cross-device identity resolution silently failed for visitors who started on mobile and converted later on desktop. Both channels looked healthy. Both were partly counting the same customer.

What to Actually Check This Quarter

This is not a replacement for the 30-minute GA4 audit. That audit verifies your events exist and fire. This is a layer under that: verifying the assumptions the platform makes about how to interpret events that are already firing correctly. Three checks matter most. Confirm the attribution model actually running for your highest-value key event by comparing its 30-day conversion volume against the reporting model shown in Admin, not just trusting the label. Confirm the counting method on every key event, especially anything imported from Universal Analytics, since a mismatched default can inflate or deflate a number nobody is watching. And treat cross-device identity as unverified until you have specifically tested it, not assumed it, the same discipline attribution fails at the edges already applies to URL-less clicks.

None of these three failures throw an error. That is exactly why they belong on a checklist instead of waiting to be noticed, the same lesson silent defaults already taught this blog once this month. A dashboard that looks fine is not the same claim as a dashboard that is fine, and the Analytics Trust Gap lives precisely in the space between those two.

Analytics tooling has genuinely gotten better at collecting more of everything: more events, more signals, more machine-learned modeling standing in for the gaps. It has not gotten better at telling a marketing team which default it quietly chose on their behalf. That gap between collection and disclosure is exactly where trust erodes fastest, and it is the gap this quarter’s audit should close.

Sources

1. Google, Change the Counting Method of Key Events – Official documentation; “once per event” is the recommended default for natively created key events, counting every trigger; “once per session” defaults only for key events migrated from Universal Analytics goals ↩
2. Google, About Data-Driven Attribution – Official documentation confirming data-driven attribution as GA4’s default cross-channel model; does not disclose specific minimum data thresholds or fallback behavior ↩
3. Seresa, GA4 Data-Driven Attribution Minimum Requirements – Jan 28 2026; practitioner-reported thresholds (~400 monthly conversions per key event, ~10,000 monthly ad interactions); reports silent fallback to last-click below threshold with no interface notification. Independently reported, not an official Google spec ↩
4. TransUnion / EMARKETER, The True Cost of Trust in Marketing Measurement – Fielded July 2025, n=196 US marketing professionals; 48% cite cross-channel deduplication as a top barrier; 55% cite cross-channel attribution accuracy as a top challenge; 60% say stakeholders question metric validity at least sometimes ↩
5. LiveRamp + Marketing + Media Alliance, The Missing Piece: Improving Confidence in Marketing Measurement – Presented July 20 2026; 1.9M real impressions, 147,941 users, 4 publishers; 50% identity-resolution precision collapsed measured ROI ~70%, from a true $1.50 to a measured $0.43 ↩

Written by Yasser Soliman

Fractional Head of WebOps

I've spent 5+ years embedded in marketing teams at B2B SaaS companies. I own the marketing website — performance, analytics, SEO, integrations — so your team ships without bottlenecks.