Marketing Attribution Models: A Decision Framework for Teams Who Don't Have Unlimited Data

A marketing attribution model is a framework that assigns credit for conversions to the marketing touchpoints that influenced them. The six models that matter in 2026 each suit a different data maturity level — from GA4-only setups that can only support first-touch to warehouse-native teams running data-driven models. Which one is right for you depends less on the model itself and more on the data infrastructure you already have.

The question every attribution guide avoids

Every "which attribution model should I use" article we've seen follows the same structure: describe the models, list pros and cons, then shrug and say "it depends on your business." That's not a decision framework. It's deferred responsibility dressed as a guide.

After running attribution implementations across SMB and mid-market clients — from e-commerce founders tracking their first paid campaigns to SaaS marketing directors trying to rescue misallocated budget — the question we've learned to ask first is not "which model do you want?" It's "what does your data stack actually support?"

An attribution model is only as good as the signal feeding it. A team running GA4 with standard events cannot run data-driven attribution — the model requires 400+ conversions per month, cross-device stitching, and a BigQuery export GA4 doesn't provide out of the box. A B2B team with a 90-day average sales cycle will be misled by last-touch attribution regardless of how much they prefer its simplicity. The model must fit the data reality. That is the only framework that matters here.

TL;DR — the decision table

• First-touch — Data: Any pageview tracking | Best for: Long cycles, brand measurement | Fails when: Short cycles, multi-channel mid-funnel
• Last-touch — Data: Any conversion tracking | Best for: Simple directional benchmarking | Fails when: Multi-touch journeys
• Linear — Data: Multi-session user tracking | Best for: Even-weight multi-channel view | Fails when: Sales cycles under 14 days
• Time-decay — Data: Timestamped touchpoint data | Best for: Short to medium sales cycles | Fails when: Long research cycles where early touches matter
• Position-based (U-shaped) — Data: Multi-touch session tracking | Best for: Awareness + conversion balance | Fails when: When mid-funnel is complex
• Data-driven — Data: 400+ conversions/month, multi-touch history | Best for: Scaling paid + content | Fails when: Low-volume, early-stage, or GA4-only

What a marketing attribution model actually is

A marketing attribution model is a rule set that distributes conversion credit across the marketing touchpoints in a customer journey. It helps teams decide which channels, campaigns, or interactions deserve budget by quantifying their contribution to outcomes. If your tracking doesn't capture every touchpoint in the journey, every attribution model will return a partial picture at best and a misleading one at worst.

The six models that matter in 2026

We've reduced our working list to six. For each, we include the specific GA4 event and property fields required — the information most attribution guides skip because they're written by people who've never had to actually implement them.

1. First-touch attribution

How it works: 100% of conversion credit goes to the first touchpoint in the customer journey — the session that brought the visitor to your site for the first time.

GA4 data required: session_source, session_medium, session_campaign at the first-session level. Cross-device stitching via user_id if you want accurate first-touch across devices (requires authenticated users).

When it wins: long sales cycles where brand awareness matters — B2B teams with 60–180 day journeys, or businesses where the initial discovery channel (organic search, podcast mention, PR campaign) drives qualified prospects that other channels subsequently convert.

When it fails: any business with short sales cycles or significant mid-funnel investment. If your retargeting is doing real work, first-touch makes it invisible.

What this model will not tell you: anything about what happened between the first touchpoint and the conversion. If 80% of your conversions require three retargeting exposures after the initial visit, first-touch has no idea and doesn't care.

2. Last-touch attribution

How it works: 100% of credit goes to the last touchpoint before conversion. The "final click gets everything" model. GA4's default.

GA4 data required: session_source and session_medium at the conversion session. The simplest model to implement — zero configuration beyond a working GA4 conversion event.

When it wins: simple direct-response campaigns with a single channel and short sales cycles. As a quick directional sanity check, not as a basis for budget decisions.

When it fails: whenever multiple channels are in play. As Avinash Kaushik has noted repeatedly in his work on web analytics, last-touch attribution reliably overcredits retargeting and branded search (which capture demand other channels created) while systematically undercrediting the channels that did the upstream work. A September 2024 Snap/EMARKETER Media Measurement Survey of 282 US marketers found that 41% most commonly use last-touch attribution and 22% rely on it exclusively — despite 74.5% reporting they want to move away from it. The gap between what teams want and what they can implement is the clearest signal of how infrastructure-constrained most attribution decisions actually are.

What this model will not tell you: what drove the lead into your funnel in the first place. The continued prevalence of last-touch across marketing teams explains a significant portion of the chronic paid media misallocation we see in client accounts.

3. Linear attribution

How it works: equal credit distributed across every touchpoint in the conversion path. Five interactions means each gets 20%.

GA4 data required: multi-session user journey data via GA4's path exploration report, or a direct connection to BigQuery with user_pseudo_id continuity tracked across sessions.

When it wins: teams that want a fairer cross-channel view without the complexity of weighted models. Works well for businesses with 3–7 touchpoint journeys and relatively balanced channel mix.

When it fails: when touchpoints are not equally valuable. A paid search click that converts a prospect who arrived from organic content does not deserve the same credit as the content that built the initial relationship. Linear treats all touches as equivalent — which they aren't.

What this model will not tell you: which part of the funnel each channel is actually serving. A blog post that drove initial awareness and a retargeting ad that nudged the final click both get 50% in a two-touch linear model — but they're doing completely different jobs.

4. Time-decay attribution

How it works: touchpoints closer to the conversion receive more credit, with credit decreasing as the gap between touchpoint and conversion increases. GA4's time-decay implementation uses a 7-day half-life by default.

GA4 data required: timestamped multi-touch session data. Ideally exported to BigQuery via GA4's BigQuery export for full control over the half-life parameter. Available natively in the GA4 attribution settings without BigQuery, but not configurable.

When it wins: short to medium sales cycles (7–30 days) where recent intent signals genuinely are more predictive of conversion. E-commerce and SaaS free trials work well here.

When it fails: long research cycles where the initial discovery touchpoint created the relationship that eventually converted. B2B deals closing over six months often originate from an organic content piece or a referral — the "most recent touchpoint" near conversion (a branded search or retargeting impression) may have added no incremental value.

What this model will not tell you: whether the early touchpoints actually influenced the buyer's decision, or whether the recency weighting is just a more mathematically sophisticated version of last-touch bias.

5. Position-based (U-shaped) attribution

How it works: first and last touchpoints each receive 40% of credit. The remaining 20% is distributed evenly across middle touchpoints. The model reflects the "opening and closing the deal" structure of many marketing journeys.

GA4 data required: same as linear — multi-session tracking with user_pseudo_id continuity, plus reliable conversion event tracking. The "position-based" model appears as a named option in GA4's built-in attribution comparison tool, but custom weighting requires a BigQuery export.

When it wins: businesses where there's a clear distinction between acquisition (the first touch) and conversion (the final push), with a middle journey that varies in length. Maps well to B2B demand generation combined with retargeting.

When it fails: when there are more than five touchpoints in the journey, or when a middle-of-funnel channel (a nurture email campaign, a webinar, a case study view) deserves more credit than the evenly distributed 20% implies.

What this model will not tell you: whether any specific middle-funnel activity was actually doing meaningful work, or just along for the ride.

6. Data-driven attribution

How it works: machine learning algorithms distribute credit across touchpoints based on what actually predicts conversion in your specific data — using Shapley values (from cooperative game theory) to estimate each touchpoint's counterfactual contribution. Google's DDA implementation in GA4 compares conversion paths against non-conversion paths to estimate incremental value.

GA4 data required: minimum 400 conversions and 300 unconverted paths per month, per conversion event. Requires Google Ads connection and at least 30 days of qualifying data. Google Signals must be enabled. GA4 silently falls back to last-touch when below the threshold — a fact that's buried in the documentation and responsible for significant reporting confusion.

When it wins: scaling operations with sufficient conversion volume and multi-channel investment. Once you have the data volume, data-driven models consistently outperform rule-based models for budget allocation decisions, because they're built on your actual conversion data rather than a generic assumption about how touchpoints work.

When it fails: any team under 400 conversions per month. Early-stage businesses, highly seasonal campaigns, niche B2B teams. Also fails for any channel outside Google's ecosystem — organic social, podcast mentions, offline events, and email are invisible to DDA's modelling.

What this model will not tell you: what happens outside Google's walled garden. DDA credits touchpoints Google can see. If your highest-impact channels are non-Google, you'll be optimising toward the wrong signal.

Two models we've retired from the working list

W-shaped attribution

W-shaped adds a third weighted moment — the lead creation event — alongside first and last touch, producing a 30-30-30-10 split across first touch, lead creation, last touch, and middle touches. The problem: "lead creation" is a Salesforce CRM concept that doesn't map cleanly to GA4 event data. The model requires a CRM integration to define the lead creation event, and most of the teams we work with don't have the infrastructure to reliably stitch that event into their attribution layer. If you run Salesforce with full-funnel tracking and a Marketo or HubSpot integration, W-shaped can be useful. For everyone else, it creates false precision.

Lead-conversion touch attribution

Same structural problem. This model requires marking specific funnel stages in a CRM that feeds your attribution layer — a marketing ops pattern for enterprise SaaS stacks with dedicated RevOps teams, not a general-purpose attribution choice. We've stopped recommending it unless the client already has the CRM infrastructure in place.

Both models assume more data infrastructure than most teams have. We removed them to avoid sending teams down implementation paths they'll abandon halfway through.

The decision framework: match model to data maturity

GA4-only, no BigQuery export: You have three real options — first-touch, last-touch, and linear. Use first-touch if brand measurement and acquisition-channel clarity matter. Use last-touch as a directional sanity check only, never as your primary model. Use linear as your default working model for budget discussions. Data-driven attribution is unavailable or unreliable at this tier.

GA4 with BigQuery export: Add time-decay and position-based to your options. You can now build genuine multi-touch path analysis. Start with position-based (U-shaped) as your default working model and check whether early-touch channels look under- or over-credited compared to your channel investment and intuition.

Server-side tracking with warehouse-native data: You have the full toolkit. Data-driven is viable if your conversion volume meets the threshold. At this tier, the more important question is whether you should graduate from multi-touch attribution to marketing mix modelling — see below.

A SaaS client we work with was allocating 70% of their paid budget to retargeting based on last-touch data. When we migrated them to position-based attribution using their GA4 BigQuery export, first-touch organic search absorbed a substantial portion of the credit that had been sitting entirely with retargeting. They reallocated 30% of their retargeting budget to content production and saw a 22% increase in new visitor acquisition over the following quarter. The attribution model didn't change the business — it changed what they were willing to see about the business. More decision-framework resources across the full attribution and analytics pillar are linked in the pillar index.

What cross-device gaps and cookieless tracking do to each model

Every multi-touch attribution model depends on your ability to stitch sessions to a single user. In 2026, two structural realities degrade this.

Cross-device tracking: Without user_id (which requires authenticated users), GA4 stitches cross-device journeys using probabilistic modelling built on Google's identity graph. The accuracy of that stitching varies by audience — B2C e-commerce with high login rates gets better cross-device data than B2B audiences who never log into anything on your properties. For cookieless B2B audiences, treat cross-device journey data as directional rather than precise.

Third-party cookie deprecation: Safari and Firefox have blocked third-party cookies since 2019 and 2022 respectively. Chrome's deprecation timeline has been delayed repeatedly, but the direction is clear. For any model relying on advertising platform pixel data — the Facebook/Meta pixel, Google Ads tag in cookie mode — your journey coverage is already degraded for a meaningful portion of your audience. Server-side tracking closes most of that gap. For teams spending more than 30% of their budget on paid media, the infrastructure cost of server-side implementation is typically recovered in the first attribution cycle through better channel allocation.

For the infrastructure layer beneath this approach, Campakt's guide to first-party tracking covers the engineering trade-offs in detail.

MTA vs MMM: knowing when to graduate

Multi-touch attribution (MTA) works well when you can track individual customer journeys and there's a direct relationship between specific touchpoints and conversions. The key difference between MTA and MMM is the unit of analysis: MTA works at the individual journey level; MMM works at the aggregate channel level. MTA breaks down when:

• Your highest-impact channels aren't digitally trackable (events, PR, word-of-mouth, radio, TV)
• You're running brand campaigns with 4–6 month revenue lag effects
• Conversion volume is too low for data-driven modelling to be statistically reliable
• You need to understand the relationship between total spend and total revenue, not touchpoint-level credit allocation

Marketing mix modelling (MMM) solves a different problem. It uses statistical regression on aggregated spend and outcome data to estimate channel-level contribution — no cookies, no sessions, no user identifiers required. The floor for MMM to deliver strong ROI sits around £4M–£5M in annual marketing spend across four or more channels; below that, the model's statistical confidence intervals tend to be wider than the actionable signal. Gartner's 2024 data shows 67% of marketing leaders planning to increase MMM investment and adoption has risen 212% since 2023 — but the data requirements remain the barrier: 18+ months of clean, consistently tracked spend and revenue history before the model has enough to work with.

Most SMB and early-stage mid-market teams should not be running MMM. The teams that benefit most are those where a large share of their investment is in channels multi-touch attribution structurally cannot track.

Implementation checklist

Before setting up or switching attribution models:

1. Confirm GA4 captures user_pseudo_id consistently across sessions — missing values break path analysis
2. Enable cross-device tracking via user_id if you have any authenticated user base
3. If targeting data-driven attribution, verify you meet the 400 conversions/month threshold for each conversion event — check the "Attribution settings" section in GA4 Admin
4. Export events to BigQuery before switching away from last-touch — you'll want historical comparison data
5. Audit your UTM parameter structure before analysing first-touch data — inconsistent UTMs are the single most common reason first-touch data is untrustworthy
6. Run two models in parallel for 30 days before switching budget decisions — the divergence between last-touch and your new model is itself a data point

The most common failure mode: teams switch to a more sophisticated model without cleaning their UTM tagging first. The model is sophisticated. The data feeding it isn't. The result is garbage in, garbage out — with a more impressive label.

Takeaway

Most marketing teams are using the wrong attribution model — not because they don't know better models exist, but because they're running on data infrastructure that can't support anything more sophisticated than what they already have. The question is not "which attribution model is best?" It's "what data do I actually have, and which model gives me accurate answers with that data?"

GA4 without BigQuery: linear is your working model. GA4 with BigQuery: position-based. 400+ conversions/month with full multi-touch tracking: data-driven. Everywhere else, you're building false precision on top of incomplete data — and making budget decisions that will quietly compound the mistake every month you leave them in place.