A marketing attribution model is a rule for deciding which touchpoint gets credit for a conversion. That's all it is. It doesn't discover truth, it applies a rule you chose, and every model gives a different answer from the same data.
For most businesses in 2026, data-driven attribution is the right default, because it's the only model that adapts to your actual customer behaviour instead of imposing a rule. Use last-click only when you need a simple, stable number that everyone already understands, and use first-click when you're specifically trying to find out what creates demand.
The uncomfortable part: every model is wrong. They all attempt to reduce a messy human decision, made across weeks, devices, and conversations you can't see, into a percentage assigned to a click. The goal isn't to find the correct model. It's to pick one that's wrong in a way you understand, and stop switching.
This article covers the six models, what each one systematically over-rewards, why your numbers will never reconcile across platforms, and how to choose.
- An attribution model is a credit-assignment rule, not a measurement. Different models give different answers from identical data.
- Last-click over-rewards the bottom of the funnel: branded search, retargeting, and email. It makes demand capture look like demand creation.
- First-click does the opposite and over-rewards whatever the customer touched first, even if it played no role in the decision.
- Data-driven attribution learns from your data instead of applying a fixed rule, but needs conversion volume and can't see anything your tracking missed.
- Every model is blind to untracked touchpoints. If a large share of your traffic is Unassigned or Direct, no model can fix that.
The six marketing attribution models, and what each one gets wrong
Take one customer journey and run it through every model. Same journey, six answers.
The journey: a user reads your blog post from Google (organic), sees a retargeting ad on Meta a week later, clicks a Google Ads branded search ad two days after that, then buys.
| Model | Who gets the credit | What it systematically over-rewards |
|---|---|---|
| Last click | Google Ads (branded) | The bottom of the funnel. Demand capture |
| First click | Organic | The top. Whatever touched them first |
| Linear | 33% each to all three | Nothing, and therefore everything equally, which is its own distortion |
| Time decay | Mostly Google Ads, some Meta, little organic | Recent touchpoints. A softer last-click |
| Position-based (40/20/40) | 40% organic, 20% Meta, 40% Google Ads | The two ends. Assumes the middle never matters |
| Data-driven | Whatever your data says | Nothing by rule, but inherits every gap in your tracking |
Look at the organic row. Under last click it earns nothing. Under first click it earns everything. The blog post didn't change. Only the rule did.
That's the entire lesson of attribution, and most arguments about it are really arguments about which rule someone's budget depends on.
Why last click is so popular and so dangerous
Last click is the default in most tools because it's simple, stable, and never disagrees with itself. It's also the model most likely to make you defund the thing that's actually working.
Branded search is the classic victim of this logic in reverse. Someone reads your article, decides to buy, googles your brand name, clicks the ad, converts. Last click hands the sale to branded search. The report says paid search is your best channel. It isn't. It's your best closer. The content created the demand and got zero credit for it.
Businesses then shift budget from content into branded search, which can't create demand, and wonder why growth flattens six months later.
Is your attribution built on complete data? The Free GTM Audit checks the tracking layer every model depends on. Enter a URL, get a report in about 10 minutes. No signup.
Data-driven attribution: what it actually does
Data-driven attribution (DDA) is now the default in both GA4 and Google Ads, and it works differently from the rest.
Instead of applying a fixed rule, it compares the paths of users who converted against the paths of users who didn't, and assigns credit based on which touchpoints actually shifted the probability of conversion. If people who saw your Meta ad converted at a materially higher rate than people who didn't, Meta gets more credit. If it made no measurable difference, it gets less.
That's a genuinely better idea than any fixed rule. It has three real limits.
It needs volume. DDA needs enough conversions and enough path variety to learn anything. On a property with a handful of conversions a month, it has nothing to learn from and quietly falls back to something closer to a rule.
It's a black box. You can't inspect why it gave 23% to a channel. When a CFO asks you to justify the number, "the model decided" is a difficult sentence to say out loud.
It only sees what you tracked. This is the one that matters most and gets discussed least. DDA can't assign credit to a touchpoint it never saw. Every untagged email, every stripped referrer, every consent-denied session is invisible to it, and the credit that belonged there gets redistributed to the channels that were visible.
That last point is why attribution work almost always turns into tracking work. If 30% of your sessions land in Unassigned or Direct, your sophisticated model is doing sophisticated maths on incomplete data, and the sophistication actively hides the gap.
Why your platforms will never agree
Every ad platform runs its own attribution, sees only its own touchpoints, and counts on its own calendar. They aren't supposed to reconcile.
- Google Ads credits Google Ads clicks, books conversions on the click date, and applies its own model.
- Meta credits Meta touchpoints, including view-through conversions, where someone merely saw an ad and didn't click. GA4 has no idea that impression happened.
- GA4 credits across all channels, books conversions on the conversion date, and uses GA4 attribution.
Add up conversions from Google Ads, Meta, and GA4 and you will exceed your real sales, sometimes by a lot. Each platform is claiming the same customer. Neither is lying. They're answering different questions, and only one of them, your backend, knows how many orders actually shipped.
Rafal runs growth at a Polish e-commerce company doing about €90,000/month in ad spend. His platforms reported 1,840 conversions for March. His warehouse shipped 1,190 orders. He spent a month trying to reconcile the gap before we reframed the problem: the gap was not an error to eliminate, it was double-counting to expect. What he actually needed was a single source of truth to plan against, and that was the backend number, not any platform.
The practical rule: pick one system as your source of truth for how many sales happened, and use the platforms only to compare channels relative to each other inside that system. Do not add them together. Ever.
How to choose a marketing attribution model
Answer three questions honestly.
1. Do you have conversion volume?
Under roughly 300 conversions a month, data-driven attribution has too little to learn from. Use last click or position-based, and revisit when volume grows. Sophistication with thin data produces confident nonsense.
2. What decision are you making?
- Optimising ad bids → last click or data-driven. Both reward what closes.
- Deciding whether to invest in content or brand → first click or position-based. Last click will always tell you content is worthless, because content rarely closes.
- Allocating a whole budget across channels → data-driven, plus an honest acknowledgment of what your tracking can't see.
3. Is your tracking complete enough to deserve a sophisticated model?
If Unassigned plus Direct is more than about 25% of sessions, fix that before you touch attribution. A better model applied to worse data gives you a more confident wrong answer, which is worse than an obviously crude one.
Most attribution projects should start as tracking projects. That's a boring conclusion, and it's the one I keep arriving at.
What to do about the touchpoints you can never see
No model sees the podcast someone heard, the colleague who recommended you, or the Slack message that started it. That isn't a tracking failure you can fix, it's a limit of the method.
Two things help.
Ask people. A single optional field on your checkout or signup form ("How did you hear about us?") gives you self-reported attribution. It's fuzzy, biased, and it will disagree with GA4. It also routinely surfaces channels that no model knew existed, and for word-of-mouth-heavy businesses it's often more truthful than the analytics.
Watch the aggregate, not the attribution. If you switch off a channel and total sales drop more than that channel's attributed revenue, it was creating demand that the model was crediting elsewhere. That's a real experiment, and it beats any model. Turning things off is the only attribution method that can't lie to you.
Frequently asked questions
What is a marketing attribution model? It's a rule for assigning credit for a conversion across the touchpoints that preceded it. Different models apply different rules and produce different answers from the same data. A model doesn't measure truth, it applies a convention you chose.
Which attribution model is the most accurate? None of them is accurate in an absolute sense, because all of them are blind to untracked touchpoints. Data-driven attribution is usually the most useful, because it learns from your data instead of applying a fixed rule, but it needs conversion volume and complete tracking to be worth anything.
What is the difference between first-click and last-click attribution? First-click gives all credit to the first touchpoint, so it rewards what creates demand. Last-click gives all credit to the final touchpoint, so it rewards what closes. Last-click systematically undervalues content and brand, and overvalues branded search and retargeting.
Why do Google Ads, Meta, and GA4 report different conversion numbers? Each uses its own attribution model, sees only its own touchpoints, and books conversions on different dates. Meta also counts view-through conversions that GA4 never sees. Adding their numbers together will exceed your real sales, because they are all claiming the same customers.
Should I use data-driven attribution? Yes, if you have enough conversion volume (roughly 300 a month or more) and your tracking is reasonably complete. If a large share of your traffic is Unassigned or Direct, fix the tracking first. A sophisticated model on incomplete data just hides the gap more convincingly.
How do I attribute offline or word-of-mouth conversions? You can't, with a model. Add a "How did you hear about us?" field to your forms for self-reported attribution, and run switch-off tests: if turning a channel off costs you more revenue than that channel was credited with, it was creating demand the model assigned elsewhere.
Next steps
Attribution isn't a measurement problem. It's a decision problem wearing a measurement costume.
Start by writing down which decision you're trying to make, then pick the model that answers it. If you're setting bids, last click or data-driven is fine. If you're deciding whether content is worth funding, last click will lie to you and you need first-click or position-based to see anything at all.
Then check the foundation. Before you invest in a better model, find out what percentage of your sessions arrives with no usable source. If it's above a quarter, your model is confidently allocating credit from data that's missing a quarter of the story, and no amount of modelling sophistication recovers information that was never collected.
If you want that foundation checked properly, the Free GTM Audit inspects the tracking layer everything else depends on. And if you want an independent monthly check on whether your attribution is telling the truth, that's what the GTM and GA4 monitoring retainer is for. €150/month for GTM, €250/month for GTM plus GA4, written report, no calls. Full pricing here.
Sources and further reading
- Attribution models in GA4 (Google)
- About data-driven attribution (Google)
- Attribution comparison tool (Google)
Is your GA4 attribution telling the truth?
The Free GTM Audit inspects the tracking layer that feeds your channel data: broken tags, consent gaps, and firing order. Enter a URL, get a report in about 10 minutes. No signup.
Start Free GTM Audit