Every marketing team eventually runs into the same argument: Google Ads says it drove the sale, Meta says it did, and the CRM says the lead came from a webinar. Attribution models are the rules (or algorithms) that decide who gets credit when a customer touches several channels before converting. This guide is for marketers, founders and growth teams at small and mid-sized companies who need to choose a model without a data-science department. It covers how the six standard models work, shows one conversion credited to different channels under five of them, offers a scenario-to-model matrix, and lays out what GA4, Google Ads and Meta support in 2026.
What Is a Marketing Attribution Model?
A marketing attribution model splits credit for a conversion (a purchase, signup or qualified lead) across the touchpoints that preceded it: say a paid social ad, a blog visit, a newsletter click and a branded search. The model does not change what happened; it changes which channel your reports say deserves budget.
There are three families:
- Single-touch models give all credit to one interaction.
- Rule-based multi-touch models split it by a fixed formula.
- Data-driven models let an algorithm estimate each touch's contribution from converting and non-converting paths.
The Six Models at a Glance
| Model | How credit is assigned | Best for | Main weakness |
|---|---|---|---|
| First-click | 100% to the first touch | Judging awareness channels | Ignores everything after discovery |
| Last-click | 100% to the final touch | Short, impulse-driven purchases | Over-credits brand search and retargeting |
| Linear | Equal share to every touch | A neutral first look at a journey | Treats weak and strong touches alike |
| Time-decay | More credit the closer to the sale | Promotions and short campaigns | Undervalues early awareness work |
| Position-based | Typically 40% first, 40% last, 20% split across the middle | Lead generation with a clear first and last step | The split is a convention, not evidence |
| Data-driven | Algorithm estimates each touch's contribution | Accounts with enough clean conversion data | Opaque, and limited to the data it can see |
The Single-Touch Models
First-click attribution
First-click gives 100% of the credit to the first recorded touchpoint. It answers a useful question, what introduces people to a brand, so it is a reasonable lens for paid social, podcasts or content. But it ignores everything between discovery and purchase, so a channel that merely starts journeys looks like it closes them. It also depends on how far back tracking can see: a first touch in another browser or beyond cookie lifetimes is invisible, so "first" often means "first that could be seen".
Last-click attribution
Last-click gives 100% of the credit to the final touchpoint before conversion. It is simple, easy to audit and was the default in many tools for years. It also flatters branded search, retargeting and direct traffic, which usually appear at the end of a journey someone else started. It is not useless: for low-priced impulse purchases where the journey lasts hours, the last touch often is the main driver. Treat it as a baseline, not a verdict.
The Rule-Based Multi-Touch Models
Linear attribution
Linear splits credit equally across every touchpoint: four touches, 25% each. It is a fair start when it is unclear which stage matters most, and no channel in the path scores zero. The cost is that a casual ad view counts the same as the click that closed the deal.
Time-decay attribution
Time-decay gives more credit to touches closer to the conversion, using a half-life: a touch loses half its weight for every N days before the sale. It suits short promotions, seasonal campaigns and nurture flows. It undervalues slow awareness work, so if content and social sit early in a long cycle, it will keep suggesting they be cut.
Position-based (U-shaped) attribution
Position-based gives the first and last touches the biggest share, commonly 40% each, and splits the remaining 20% across the middle. It fits lead generation and B2B journeys with a clear first contact and demo request. But 40/20/40 is a convention, not a law of nature, and it undervalues middle touches such as a comparison page or case-study email.
Data-Driven Attribution
Data-driven attribution (DDA) estimates how much each touchpoint changed the probability of conversion by comparing converting paths with non-converting ones. Instead of assuming a shape, it learns one from the data. Google made DDA the default for new Google Ads conversion actions in late 2021, dropping the minimum-data requirement, and it is the default reporting model in GA4.
It adapts as behavior changes and avoids hand-picked weights. But it is a black box, since Google does not disclose how credit is assigned. It only models journeys it can see, so cross-device or consent gaps mean it works on broken data. And its view is one platform's: Google Ads DDA credits Google Ads interactions, and GA4 DDA sees site visits but not an impression of a Meta or TikTok ad.
One Conversion, Five Models: A Worked Example
The journey below is an illustrative example, not client data. A customer buys a $1,200 annual plan after four touches:
- 12 days before purchase: clicks a Meta ad (paid social).
- 9 days before: reads a blog post found through organic search.
- 4 days before: clicks a link in a newsletter email.
- 1 day before: searches the brand name, clicks a Google Ads search ad, and buys.
| Touchpoint | First-click | Last-click | Linear | Position-based (40/10/10/40) | Time-decay (7-day half-life) |
|---|---|---|---|---|---|
| Meta ad | $1,200 | $0 | $300 | $480 | $159 |
| Organic blog post | $0 | $0 | $300 | $120 | $215 |
| Newsletter email | $0 | $0 | $300 | $120 | $352 |
| Google Ads brand search | $0 | $1,200 | $300 | $480 | $474 |
Reading down the columns gives five stories. First-click says Meta is the best channel and brand search is worthless; last-click says the opposite. Linear calls every channel equal. Position-based splits credit between Meta and brand search. Time-decay crowns brand search but gives the newsletter more than the blog post or the Meta ad.
Now the budget decision. A team on last-click sees $1,200 credited to branded search and may cut paid social. Yet the branded search happened because the Meta ad and blog post created awareness; remove them and branded searches shrink. Branded search and retargeting routinely collect credit for demand created elsewhere.
The fix is not to crown a new winning model but to compare two or three and focus on where they disagree. Meta's credit ranges from $0 to $1,200 here, which is the honest answer: its contribution is uncertain, and a holdout test or controlled spend change resolves it. A data-driven model is absent because it learns weights from thousands of paths and cannot be computed by hand for one journey.
Why Single-Touch Models Mislead
Single-touch models are incomplete, and the missing part is what budget decisions depend on. They reward position rather than contribution, so a channel earns credit for being first or last whatever it did for the customer. They create predictable budget drift: last-click pushes spend toward brand search and retargeting, first-click toward awareness channels that never close alone. And they hide interaction effects, such as email working only because an ad warmed the audience.
Which Attribution Model to Use: A Scenario Matrix
| Scenario | Typical journey | Suggested model | Caveat |
|---|---|---|---|
| Impulse ecommerce, low price | Hours to a few days | Last-click or data-driven | The last touch is close to the cause |
| Considered purchase or SaaS trial | One to four weeks | Data-driven, checked against position-based | Several research touches before buying |
| B2B lead generation with a CRM | Weeks to months | Position-based, then data-driven on qualified leads | Form fills are not revenue; join CRM outcomes |
| Judging top-of-funnel channels | Any | First-click as a lens beside last-click | Shows who starts journeys, not who finishes them |
| Seasonal sale or time-boxed promo | Days | Time-decay | Short half-life, or early touches vanish |
| Very low conversion volume | Any | Linear or position-based plus holdout tests | Algorithms have little to learn from |
| Budget split across Google, Meta, TikTok, LinkedIn | Cross-channel | Several models side by side on joined data | No single platform's report is neutral |
Two habits matter more than the model. Decide the question first: "which channel gets the next $5,000?" differs from "where do customers first hear about us?". And never change models mid-quarter and compare across the change; a switch redistributes credit without one extra sale.
What GA4, Google Ads and Meta Support in 2026
GA4 and Google Ads. Google retired first-click, linear, time-decay and position-based from both products in 2023, citing low adoption (under 3% of conversions), leaving data-driven and last-click. GA4 reporting offers data-driven, paid and organic last click, and Google paid channels last click. For first-touch-style analysis, GA4's path exploration shows early touchpoints without changing the reporting model.
Meta. On January 12, 2026 Meta removed the 7-day view and 28-day view windows, so the longest view window is one day; 28-day click remains for comparison only. A March 2026 change limited click-through to actual link clicks and moved other interactions into a one-day engage-through category. The default is now 7-day click, 1-day engage-through, 1-day view. Reported conversions can drop without any real change in performance, so 2025 and 2026 Meta reports are not like-for-like.
The grading-their-own-homework problem. Each platform counts conversions inside its own windows and logic, so what Google, Meta and an email tool claim usually adds up to more than real sales. Platform numbers are best used to optimize inside each platform, with a neutral source (CRM, store backend or joined dataset) judging the whole.
What No Attribution Model Can Fix
Models only redistribute the data they are given. Four gaps sit upstream of any model:
- Cross-device and consent gaps, where one person looks like several visitors, or like none.
- Offline conversions such as phone calls, demos and in-store purchases that never reach analytics.
- View-through influence inside walled gardens.
- Double counting across platforms that each claim the same sale.
Two methods complement attribution. Incrementality testing (a geo holdout or conversion-lift study) measures what happens when a channel is reduced, the closest thing to a causal answer. Marketing mix modeling works from aggregate spend and sales, needs no user-level tracking and suits budget questions across many channels.
Once journeys span Google Ads, Meta, TikTok, LinkedIn and a CRM, one joined dataset becomes necessary: spend, clicks and CRM outcomes keyed to the same campaign and customer. That is usually where teams outgrow single-platform reports and look at cross-channel analytics tooling for ROAS, CAC and journey views.
Get the Data Right Before Choosing a Model
A sophisticated model on poor tracking produces confident nonsense. The foundations come first:
- Conversion firing: the Google Ads conversion tracking setup guide covers the common breakages.
- Stores: send revenue events via the GA4 ecommerce events setup guide, so credit is measured in money.
- Lead-gen sites: track form submissions in GA4 without code so a lead is a real event.
- Campaign context: set up GA4 custom dimensions for the parameters being tagged.
- Journey comparison: build useful GA4 audiences and segments to compare short and long journeys.
- Sales outcomes: connect Google Sheets to the CRM, the cheapest first step toward revenue-level attribution.
Acquisition is only half the story; for post-signup behavior, compare Mixpanel and Amplitude.
FAQ
What is the best marketing attribution model?
There is no universal best. Data-driven attribution is the strongest default when the platform supports it and tracking is clean. Compare it with last-click and a position-based view, and treat big disagreements as a prompt to test, not a result to trust.
What is the difference between single-touch and multi-touch attribution?
Single-touch attribution gives all credit to one interaction. Multi-touch spreads it across several, by a fixed rule (linear, time-decay, position-based) or an algorithm (data-driven). Multi-touch reflects real behavior better, but still depends on touchpoints being tracked.
Does GA4 still offer first-click, linear, time-decay and position-based models?
No. Google removed them from GA4 and Google Ads in 2023. Rule-based views now need an external tool, a spreadsheet model built from exported paths, or a dedicated attribution platform.
Is data-driven attribution better than last-click?
Usually, not automatically. It captures earlier touches and adapts over time, while last-click ignores them. The trade-offs are opacity and a narrow field of view. For short, impulse-driven purchases the two often give similar results.
Which attribution model should a small business use?
Start with the platform default (data-driven in GA4 and Google Ads), keep last-click as a baseline, and add a position-based view if the sales cycle exceeds a week or two. With very few conversions, simple comparisons and occasional on/off tests beat trusting any algorithm.
How do attribution, incrementality and marketing mix modeling differ?
Attribution assigns credit to touchpoints in tracked journeys. Incrementality testing measures the extra conversions a channel causes by comparing exposed and unexposed groups. Marketing mix modeling estimates channel effects from aggregate spend and sales. Attribution is fastest; the other two are better evidence for big budget decisions.
TL;DR
- Data-driven attribution is the best default in 2026 where a platform offers it; keep last-click as a baseline.
- Linear, time-decay and position-based work as sanity checks for low volume or long sales cycles.
- First-click and last-click alone mislead: each gives 100% of the credit to one touch.
- GA4 and Google Ads only offer data-driven and last-click; Meta's view windows shrank to one day in January 2026.
- Compare two or three models, focus on where they disagree, and settle the big disagreements with holdout tests.
Originally published on Trackstack Editorial.
Top comments (0)