Marketing Attribution: Which Model Is Right for You?

Every attribution model has a flaw. Last-click inflates branded search. First-click ignores everything that actually closes the deal. And all of them quietly break the moment a user declines your cookie banner.

We've audited tracking setups across dozens of companies. The pattern is almost always the same: Google Ads reports 40% more conversions than GA4, the CRM tells a third story, and someone is making budget decisions based on whichever number they like best. This guide cuts through that.

The Five Standard Models - Ranked by Usefulness

Last-Click

Gives 100% of the credit to the final touchpoint before purchase. It's the Google Ads default and the fastest way to systematically over-invest in branded search while starving every channel that builds awareness upstream.

Use it only if your entire funnel is a single, impulse-driven step. So, almost never.

First-Click

All credit goes to the first touchpoint. Useful for understanding what generates initial awareness, but a terrible basis for budget decisions. Using first-click to allocate spend is like crediting your accountant for closing a deal because they were in the building on day one.

Linear

Splits credit evenly across all touchpoints. A reasonable starting point when you have no idea what your customer journey looks like but "treating everything equally" is just another way of saying you don't yet know what matters.

Time-Decay

Gives more credit to touchpoints closer to conversion. A better fit for subscription businesses with short trial-to-paid cycles. Just don't confuse recency with causality.

Position-Based (U-Shaped)

40% to first touch, 40% to last touch, 20% distributed across the middle. A pragmatic compromise when both acquisition source and closing channel matter. For B2B in particular, this is a reasonable default.

Data-Driven

Uses machine learning to assign fractional credit based on actual conversion paths the most accurate rule-based model, provided you have enough volume (Google recommends 300+ conversions/month). Below that threshold, you're feeding noise into a black box and calling it science.

The Problem Nobody Wants to Admit

All of these models break the moment users stop being trackable. And in 2026, most of your users are not fully trackable.

The Privacy Layer

When a user declines consent, your pixels fire nothing. GA4 records nothing. Google Ads reports nothing. The conversion still happened you just have no idea which channel drove it.

Consent Mode v2 introduces modeled conversions to partially fill this gap. Google's algorithm estimates conversions for non-consenting users based on patterns from those who accepted. It's better than nothing. It is not the same as actual data.

Server-Side GTM Is the Baseline

If you're still running entirely client-side tracking, your attribution model is built on incomplete events. Ad blockers, Safari's ITP, and consent suppression all eat client-side tags. Server-Side GTM moves event collection server-to-server, bypassing most of these gaps and improving data quality for GA4, Facebook CAPI, and CRM syncs simultaneously.

This isn't advanced infrastructure. It's the minimum foundation for any attribution model worth trusting.

The Google Ads vs. GA4 Discrepancy

Google Ads uses view-through conversions by default, a different attribution window than GA4, and modeled data that doesn't produce a matching GA4 event. The result: a team optimizing toward Google Ads numbers is chasing a metric that doesn't match what actually happened. Reconcile both sources and your CRM - before building any model on top of them.

Marketing Mix Modeling vs. Attribution: When to Use Which

Most companies ignore this distinction until they're spending €500K+ per month. You shouldn't wait that long.

What MMM Is

Marketing Mix Modeling is a regression-based technique that measures the incremental revenue impact of each channel using aggregated historical data - no user-level tracking, no cookies. It works at the macro level: total spend in channel X, total revenue in period Y, with seasonality and external factors controlled for.

The Honest Comparison

Use attribution for tactical decisions - which ad to scale, which keyword to bid on. It answers: of the users who converted, which touchpoints did they interact with?

Use MMM for strategic budget allocation, especially with offline spend, high-GDPR markets, or brand campaigns that work over months. It answers: if we shift €10K from paid search to paid social, what happens to revenue?

Under €100K/month? A clean attribution setup with server-side tracking is sufficient. Above it - particularly in subscription businesses where LTV varies significantly by channel you need both.

B2B Attribution: A Different Beast

Standard models are designed for short purchase cycles. A B2B company with a 90-day sales cycle, six buying stakeholders, and a mix of LinkedIn, webinars, SDR outreach, and paid search is not well-served by a model that looks at the last click before a form fill.

The Multi-Stakeholder Problem

The CMO reads a LinkedIn article. The IT lead finds your pricing page via organic search. The economic buyer clicks a retargeting ad the day before the demo. Last-click credits the retargeting ad. LinkedIn gets nothing. Your team cuts LinkedIn. Six months later, pipeline dries up.

What Actually Works

Account-based attribution: track touchpoints at the account level, not the individual, and connect them to CRM deal stages. In practice this means joining GA4 and ad platform data with your CRM in BigQuery then building cohort analyses that show which channel combinations correlate with faster sales cycles and higher contract values.

How to Choose and Build Your Model

Step 1: Fix Your Tracking Foundation First

No model is more accurate than the data feeding it. Before choosing anything, confirm: Server-Side GTM is deployed, Consent Mode v2 is configured correctly, GA4 and Google Ads conversion actions are reconciled, and CRM events are firing cleanly. Running a sophisticated model on broken tracking is like calibrating a precision instrument on a cracked wall.

Step 2: Map Your Actual Customer Journey

How many touchpoints typically precede a conversion? What's the median time from first touch to purchase? For subscription businesses, extend this past the conversion event - do customers from specific channels retain longer, upgrade more, or churn in month two?

This is where SQL in BigQuery becomes necessary. A single query joining acquisition source with subscription event data often tells you more than six months of dashboard staring:

SELECT
  first_touch_source,
  COUNT(DISTINCT user_id)   AS customers,
  AVG(ltv_12_month)         AS avg_ltv_12m,
  AVG(months_to_churn)      AS avg_months_retained
FROM `project.analytics.customer_ltv`
WHERE acquisition_date BETWEEN '2025-01-01' AND '2025-12-31'
GROUP BY first_touch_source
ORDER BY avg_ltv_12m DESC

Step 3: Match the Model to the Decision

Step 4: Connect Attribution to LTV

Most attribution stops at the conversion event and that's where subscription companies make their biggest mistakes. A channel that looks expensive on a CAC basis might acquire customers with a 24-month average lifetime, versus 8 months for the "cheaper" channel. You're not buying conversions. You're buying revenue streams.

Build LTV cohort analyses by acquisition channel in BigQuery. Once you can model LTV at the channel-cohort level, you stop optimizing for conversions and start optimizing for business value.

Step 5: Pick One System of Record

Most teams default to Google Ads as source of truth because that's where the spend is. Wrong choice. Your CRM contains actual customers - not modeled ones. Build attribution reports against CRM data, with ad platform data as input, not the other way around.

Mistakes That Cost Real Money

• Optimizing toward view-through conversions. Google Ads counts these by default - users who saw an ad but never clicked, then later converted. For most businesses this inflates display ROI by 3–5×. Weight them at 0.1× or turn them off.
• Ignoring the consent gap. In high-GDPR markets, 30–60% of conversions may be invisible to your model. Check what share of your Google Ads conversion volume is modeled vs. observed before acting on it.
• Using the same model for brand and performance campaigns. Brand search converts well under any model - it's capturing demand you already created. Crediting that to the brand campaign is circular logic. Separate the analysis.
• Treating attribution and LTV as separate workstreams. Every attribution decision has a downstream LTV consequence. Build the connection in from day one.