GA4 attribution models for agencies: which to use and what they hide

TL;DR

An attribution model decides how GA4 splits credit for a conversion across the touchpoints that led to it. GA4 today offers two that matter: data-driven attribution (DDA) — the default, which uses machine learning to distribute credit across touchpoints — and last-click (cross-channel last non-direct), which hands all the credit to the final channel. The older rule-based models (first-click, linear, time-decay, position-based) were removed from GA4. The model you choose silently changes which channels look like winners, so agencies need to (1) pick one deliberately and stay consistent, (2) know it differs from Google Ads' own attribution (so the numbers will never match), and (3) remember the uncomfortable truth: the model matters far less than whether your tracking actually captured the touchpoints — no model can attribute a click your broken tag never recorded. Below: what each does, which to pick, and the trap.

---

Two agencies can look at the same client's GA4 and disagree about whether "email" or "paid search" drove the revenue — not because the data differs, but because they're on different attribution models, a setting most people never touch and fewer understand. It's worth understanding, because it quietly shapes every channel conversation you have with a client. Here's the practical version.

What an attribution model actually does

A conversion usually has more than one touchpoint — a user finds the client via organic, comes back from an email, clicks a retargeting ad, then converts. The attribution model decides how to divide the credit for that one conversion across those touchpoints. Change the model and the same conversions redistribute across channels — paid looks better or worse without a single thing changing in reality.

The models GA4 actually offers now

• Data-driven attribution (DDA) — the default. Google's ML looks at converting and non-converting paths and distributes credit based on each touchpoint's modeled contribution. No fixed rule — it's specific to the property's data. This is the recommended default for most accounts.
• Last-click (cross-channel, last non-direct). All credit goes to the last channel before the conversion — and crucially, Direct is skipped if there was any prior known channel (so a typed-URL return visit doesn't steal credit from the ad that started it). Simple, predictable, and what most clients intuitively picture.

Gone: first-click, linear, time-decay, and position-based were removed from GA4 (2023). If a client asks for "position-based attribution," that's a Universal Analytics memory — GA4 doesn't do it anymore.

Two related settings sit next to the model and matter just as much:

• Lookback window — how far back GA4 looks for touchpoints (e.g. 30 vs 90 days). Longer windows credit earlier touches; for long B2B cycles, the default may be too short.
• Reporting scope — cross-channel vs "Google Ads preferred" (which leans credit toward Ads). Most agencies should stay cross-channel for an honest picture.

Which to pick

For most clients: leave it on data-driven. It's the most defensible distribution and it's Google's recommendation. Switch to last-click only when you have a specific reason — a client who needs a dead-simple, explainable model, or when you're sanity-checking how much DDA is reallocating. The mistake isn't picking the "wrong" model; it's switching models constantly so reports never reconcile, or not knowing which one you're on when a client asks why a channel's numbers moved.

The discipline:

1. Pick one model per client and document it (in the measurement plan).
2. Match the lookback window to the sales cycle — longer for considered B2B purchases.
3. Stay cross-channel unless you have a reason not to.

Why GA4 and Google Ads will never agree

A predictable client question: "GA4 says paid drove 40 conversions, Google Ads says 55 — which is right?" Both, and neither. They use different attribution models, different conversion definitions, and different windows, and they count in different places. This mismatch is structural, not a bug — explain it once, pick one as the source of truth for each decision, and stop trying to make them tie out.

The trap: attribution can't fix missing data

Here's the part that matters more than the whole model debate: attribution only divides the touchpoints it actually has. If a tag is broken, a UTM is non-standard, or cross-domain isn't configured, the touchpoint is missing — and no model, however clever, can credit a click GA4 never recorded. A beautiful DDA report built on data that's quietly dumping a third of sessions into Unassigned is precisely-calculated nonsense.

So the priority order is the opposite of where most attention goes: fix the data capture first (so the touchpoints exist), then worry about the model. That's why verifying conversions and channels beats re-litigating attribution settings — the model is a lens, and a lens can't show what was never recorded.

Where this fits

Attribution is a reporting lens; the thing under it — every touchpoint actually being captured cleanly across a client's tags, channels, and pixels — is the part that determines whether the lens shows anything true. Phloz keeps that substrate legible: each client's tracking modeled and health-checked as part of the tracking-infrastructure map, so the attribution conversation rests on data you've verified rather than hoped for. The CRM for PPC agencies and pricing pages cover the workflow — but the one-line takeaway is: leave it on data-driven, stay consistent, and spend your energy on the data, not the model.