When is a discrepancy a tracking bug instead of a reporting difference?

It becomes a tracking problem when the gap is unexplained after scope alignment, or when one source is clearly missing sessions, events, revenue, or campaign context that should be present.

First-Touch vs Last-Touch vs Position-Based: GA4 Attribution Decision Tree (2026)

Which attribution model should I use in GA4?

Use Data-Driven Attribution (DDA) if your property has 400+ conversions per month with sufficient consent-granted data — DDA is GA4's default for paid channels and produces the most accurate cross-channel credit assignment. Below that threshold, GA4 silently falls back to last-click without warning. Use last-click for short-cycle e-commerce where the closing channel is the meaningful credit.

Use first-click for top-of-funnel awareness analysis where you need to credit channels that initiated journeys. Avoid linear, time-decay, and position-based models — they're available but offer little advantage over DDA when DDA has enough data.

The four models GA4 supports

GA4 dropped the older first-click, linear, time-decay, and position-based options as default reporting models in 2023. The remaining options in 2026:

Model	Credit assignment	Use case
Data-Driven Attribution (DDA)	Algorithmic — uses ML to distribute credit based on observed conversion patterns	Default; needs 400+ conversions/month to work
Last-click	100% to the last non-direct touchpoint	E-commerce, short-cycle conversions, paid-channel ROAS reporting
First-click	100% to the first touchpoint in the path	Awareness analysis, demand-gen channel evaluation
Linear / Time-decay / Position-based	Fixed-rule distributions	Available but rarely chosen — DDA does this better when it works

DDA is GA4's default. It runs continuously, learning from observed conversion patterns to assign fractional credit across touchpoints. The more data, the better — and below threshold, it silently disables.

The 400-conversions threshold

GA4's DDA requires at least 400 conversions per conversion type per month with sufficient touchpoints in the conversion path. The exact mechanics:

Threshold is per conversion event type, not aggregate. A property with 200 purchases + 200 leads doesn't qualify either model — neither hits 400.
Threshold applies over a trailing 30 days. New events need 30 days to qualify.
Insufficient consent can drop the qualifying count even if raw conversions are above 400.
When threshold isn't met, GA4 silently falls back to last-click — no UI warning, no email notification.

The fallback is the dangerous part. A stakeholder reading the Attribution report sees data attributed across channels and assumes DDA. In reality, it's last-click — but reported under the DDA label. Always check the model dropdown in the report to confirm which model is actually active.

The decision tree

Use this flowchart to pick a model:

1. Do you have 400+ conversions per event type per month?

No → Use last-click explicitly. Don't rely on DDA's fallback. Make the model choice deliberate so stakeholders understand what they're seeing.
Yes → Continue to question 2.

2. Is your conversion cycle short (under 7 days from first touch to convert)?

Yes → DDA is the default; last-click is a reasonable alternative for simpler reporting and ROAS calculations against paid channels.
No → Continue to question 3.

3. Do you need to credit awareness/top-funnel channels?

Yes → Use first-click alongside DDA. Compare the two views to understand the gap between awareness drivers and closing channels.
No → DDA is your single best model.

4. Are you comparing channels for paid budget allocation?

DDA preferred. Last-click systematically under-credits awareness channels (display, paid social) and over-credits brand search.

Want to see whether attribution loss is already distorting your channel data?

Audit attribution

The honest answer for most properties: DDA when it qualifies, last-click when it doesn't, with first-click as a secondary view for awareness analysis.

What "data-driven" actually does

DDA uses machine learning trained on your property's conversion data. The model:

Observes conversion paths (touchpoint sequences leading to conversions)
Compares paths that converted vs paths that didn't convert
Identifies which touchpoints had the highest correlation with conversion
Distributes credit fractionally based on the touchpoint's observed lift

The output: each conversion has fractional credit distributed across multiple channels. A purchase preceded by Display → Email → Paid Search might be credited 35% Display, 25% Email, 40% Paid Search based on observed patterns.

Limitations of DDA:

Requires the threshold. Below 400 conversions, the model can't learn — falls back to last-click.
Requires sufficient consent. EU/UK properties with low Consent Mode V2 acceptance may not have enough qualifying data.
Cannot credit unmeasured channels. Offline conversions imported via Enhanced Conversions improve the model; channels with no observed touchpoints get zero credit.
Sensitive to time-zone changes. Property time-zone changes can disrupt the learning window.

The silent fallback problem

The single biggest attribution gotcha in GA4: DDA silently disables when threshold isn't met, and the report continues to label itself as DDA even though it's serving last-click data underneath.

How to detect:

Check Admin → Property Settings → Attribution Settings. Confirms the model selected.
Check Reports → Advertising → All channels. The Attribution Model dropdown shows the active model. If it's DDA but your conversion volume is below 400, what you're seeing is last-click data with a DDA label.
Run the same report with model = DDA and model = last-click. If the channel splits are identical, DDA isn't actually running — it's falling back.

Mitigation:

For low-volume properties: explicitly set last-click as your primary model. Don't rely on DDA's fallback.
For volume-borderline properties: monitor monthly conversion count. If it drops below 400, switch to explicit last-click before stakeholders see misleading reports.
For all properties: annotate any change in attribution model in your dashboards. Stakeholders should never be confused about which model is active.

Lookback windows

The lookback window controls how far back GA4 looks when assigning credit:

Acquisition events (first_visit, sign_up, account_creation): default 30 days, configurable up to 30 days.
All other conversions (including purchases): default 90 days, configurable to 30, 60, or 90 days.

Most properties leave both at default. Shorter windows reduce credit assigned to top-of-funnel touchpoints (display, paid social) and increase credit to closing channels (paid search, direct, email). Longer windows do the reverse.

The strategic choice:

30 days — short-cycle e-commerce. Captures the buying cycle without crediting touchpoints from months ago.
90 days — high-consideration B2B, expensive products, long sales cycles. Credits the awareness investment that initiates the journey.

Changing the lookback window applies retroactively to GA4's standard reports — recent reports get recomputed with the new window. BigQuery exports preserve the original assignments per session.