GA4 Data-Driven Default
Since mid-2023, GA4 defaults to "Data-Driven Attribution" for conversions. This replaces the old Last-Click default from Universal Analytics. The algorithm is Google s proprietary machine-learning model that assigns fractional credit across touchpoints based on your site s conversion patterns.
On paper this is an upgrade: data-driven beats rule-based when you have enough data. In practice, most sites do not meet Google s own threshold to run the algorithm effectively, and Google silently falls back to a rule-based model.
Google s published threshold for data-driven attribution: 600 conversions in the selected channel grouping within 30 days, 3,000+ conversions total in 30 days across your property. Below that: data-driven results are based on a smaller data sample and are statistically shaky.
For most B2B SaaS sites, you will never hit 3,000 conversions/month. GA4 shows you data-driven results anyway, but they are noise.
The 6 Models Available in GA4
You can switch models in GA4 under Advertising → Attribution → Model comparison.
| Model | How GA4 implements it | Use when |
|---|---|---|
| Data-Driven | Proprietary ML on your data | 3000+ monthly conversions |
| Last Click (ex-Direct) | 100% to last non-direct touch | Checking DR baseline |
| Last Click | 100% to last touch including direct | Rarely useful |
| First Click | 100% to first touch | Demand creation analysis |
| Linear | Equal credit all touches | Short simple funnels (see linear guide) |
| Position-based | 40% first, 40% last, 20% middle | B2B SaaS default (see model ranking) |
| Time Decay | Half-life of 7 days | Sales-led, 30–90 day cycles |
Note: GA4 removed "W-Shaped" and other advanced models that existed in Universal Analytics. If you need W-shaped for B2B, you build it outside GA4 in your CRM or with HockeyStack.
When GA4 Attribution Is Wrong
1. Small data volume
Under 3,000 monthly conversions, data-driven is noisy. Numbers bounce between reporting periods without real business changes. You may conclude a channel is declining when it is just statistical wobble.
2. Consent-loss and cookie-less environments
GDPR consent banners, Safari s Intelligent Tracking Prevention, Firefox s enhanced tracking, and cookie wipes all reduce the data GA4 sees. In EU traffic, consent rates of 40–60% are normal — which means 40–60% of your data is missing. GA4 models on the visible data but calls it your total.
3. Cross-device journeys
Without signed-in users or Google Signals enabled (and opted-in to by users), GA4 cannot stitch a session on mobile to a session on desktop for the same user. Each device looks like a new user. For B2B SaaS buyers who research on their phone and buy on their laptop, this is catastrophic.
4. Long consideration cycles
GA4 s default lookback window for acquisition is 30 days. Many B2B SaaS buyers take 60–120 days. If the first touch was 90 days ago, GA4 does not see it — that first touch gets reclassified as "direct" or "organic" on subsequent sessions. Your demand-creation channels get systematically under-credited.
5. Offline conversions
Deals that close via phone call, email, or Zoom — without touching your checkout — are invisible to GA4 unless you manually push them back via the Measurement Protocol or a CRM integration. Most teams never do this.
How to Read GA4 Attribution Reports
In GA4, go to Advertising → Attribution → Model comparison. Open two models side-by-side: Data-Driven and Position-Based.
What you are looking for:
- Agreement on channel rankings: if both models rank channels in roughly the same order, trust the ranking. Use Data-Driven for the numbers, Position-Based as confidence.
- Disagreement on rankings: if models rank channels differently, you have a funnel where model choice matters. The direction to trust is usually Position-Based for B2B SaaS (see model ranking).
- Massive "Direct" attribution: if Direct looks unrealistically large, your branded search is likely paid-driven (direct = people typing your URL after seeing an ad). Compare Direct to paid ad spend trends — they usually correlate.
Supplementing GA4 (Server-Side + Custom SQL)
Two tactics to make GA4 attribution more honest:
Server-side tracking for paid channels
Install Meta Conversions API, Google Ads Enhanced Conversions, LinkedIn Conversions API. These send conversion events from your server to each ad platform, which restores some of the signal lost to iOS 14, Safari ITP, and consent banners. Every major ad platform has this — no excuse not to have it.
Custom SQL layer on top of GA4 BigQuery export
Enable GA4 s free BigQuery export. Write SQL queries to build attribution models GA4 does not offer natively — W-shaped, custom weighting based on your funnel stages, segment-specific attribution. This is the "graduate" move when rule-based models stop meeting your needs. See my custom attribution model guide.
Monthly Workflow
15-minute monthly GA4 attribution review:
- Open GA4 Model Comparison: Data-Driven vs Position-Based
- Check channel rankings and flag any divergences
- Check "Direct" share — if it has crept above 30%, investigate branded search + paid demand gen
- Spot-check a few closed deals against GA4 s attributed path to confirm sanity
- Document one budget decision for the month based on this review
That is the whole process. You do not need a $2k/month tool to do attribution well with GA4 — you need discipline, awareness of GA4 s blind spots, and a willingness to look at two models in parallel rather than trust any single default.