What Is Conversion Attribution? Models, Examples, and What Breaks Them

What is conversion attribution?

Conversion attribution is the practice of assigning credit for a conversion to the marketing touchpoints that led to it. A customer finds you through a Google search, leaves, comes back a week later from an email, reads a comparison, and finally buys after clicking a retargeting ad. That's five touchpoints. Attribution is how you decide how much of that sale each one earned.

It exists because the alternative is guessing. Without attribution you know your conversions happened, but not what produced them — so you can't tell whether to spend the next dollar on content, email, or ads. Attribution turns "we got 50 signups this month" into "organic search discovered them, email nurtured them, and retargeting closed them," which is the difference between a budget decision and a coin flip.

The catch — and the reason attribution is a whole discipline rather than a checkbox — is that the credit split depends entirely on the model you choose. Give the same five-touch journey to two different models and they'll name two different winners. So understanding attribution means understanding the models, and then understanding why even the right model gives wrong answers when the underlying data is broken. We'll take both in turn.

The attribution models, on one $1,000 sale

The clearest way to see what a model does is to run the same conversion through several. Imagine one $1,000 sale with five touchpoints in order: Organic Search → Blog → Social → Paid Ad → Email (the click that converted). Here's how the common models split that $1,000.

Source: Worked example — one conversion path across five channels, four models

• First-touch gives all $1,000 to Organic Search — the channel that created awareness. It answers "what starts customer journeys?" and systematically over-credits discovery.
• Last-touch gives all $1,000 to Email — the final click. It's the default in most tools because it's simple, and it answers "what closes?" while ignoring everything that set up the close.
• Linear splits $1,000 evenly — $200 to each of the five touches. Fair, but it treats a throwaway social impression the same as the decisive email.
• Time-decay weights recent touches more, so Email might get ~$480 and Organic Search only a sliver — on the logic that what happened just before the purchase mattered most.

Two other models round out the set: position-based (U-shaped) gives extra weight to the first and last touches (often 40% each) and splits the remaining 20% among the middle, and data-driven uses your own conversion data to assign fractional credit statistically rather than by a fixed rule [1][2].

Here's the whole set side by side — what each credits, the question it answers, and when it's the right call:

The lesson from one chart: there is no neutral model. Every choice encodes an assumption about what matters — discovery, closing, or the whole path. Picking one isn't a technicality; it's a claim about how your customers actually decide.

Which model do teams actually run?

In theory everyone should use sophisticated multi-touch models. In practice, here's the cohort.

Source: Attrifast 200-site cohort — configured default attribution model, 2026

52% of sites run last-touch — usually because it's the default and nobody changed it. First-touch is a distant second at 21%, multi-touch models (linear and position-based) together are 15%, time-decay 8%, and genuinely data-driven attribution just 4%. The gap between "what's recommended" and "what's configured" is the real state of SMB attribution: most teams are looking at a last-click view that over-credits their closing channels and hides everything that fed them.

That default has a predictable blind spot. Discovery channels — content, AI visibility, PR — rarely get the last click, so a last-touch-only shop systematically under-values exactly the top-of-funnel work that fills the pipeline. The fix isn't necessarily a fancier model; it's looking at first- and last-touch side by side, which is where the disagreement becomes informative.

How much the models disagree (and where)

If first- and last-touch mostly agreed, the model choice wouldn't matter. They don't.

Source: Attrifast cohort — revenue that switches channels between first- and last-touch, 2026

The share of revenue that gets reassigned to a different channel when you switch from last-touch to first-touch ranges from 22% (Social) to 58% (AI engines). Email swings 41%, Paid Search 33%, Organic Search 29%. The AI-engine number is the standout and the most strategically important: AI engines are overwhelmingly discovery touches — someone asks ChatGPT for options, clicks through, then converts later via a different channel — so last-touch barely credits them while first-touch credits them heavily. A team running last-touch only will conclude AI traffic doesn't convert, when really it's discovering customers who close elsewhere. That exact misread is why we built a separate view for it in AI revenue attribution, and why first-touch vs last-touch attribution is worth understanding channel by channel.

The real problem: the model is fine, the data is broken

Here's the part most attribution content skips. You can pick the perfect model and still get the wrong answer, because the model runs on a touchpoint path that's already missing pieces before it starts.

Source: Attrifast cohort — share of conversions with a broken attribution join, 2026

In our cohort, 34% of conversions were credited to the wrong channel — not because of a bad model choice, but because a source signal was destroyed before it could be recorded. Three failure modes do the damage:

1. Third-party cookie blocking (Safari, Firefox, Brave) erases returning-visitor identity, so earlier touches can't be linked to the converting visit and drop out of the path. This is the same erosion that makes cookieless analytics necessary.
2. Referrer stripping by AI clients and in-app browsers files real sources under Direct, so a touch that was really "Perplexity" gets credited as "Direct" — the mechanism behind what is direct traffic.
3. Consent-banner suppression removes some visitors from measurement entirely, so their touches never exist in the data.

The consequence is subtle and expensive: a flawless last-touch calculation on a path missing half its touches produces a confident, wrong answer. Switching models doesn't help — you're applying a different rule to the same broken data. The fix is upstream, in the identity and source layer: durable first-party identity that survives cookie blocking, and source capture that survives referrer loss. Only after that does the model choice matter.

Conversion attribution vs revenue attribution

One more distinction that quietly changes every conclusion. Conversion attribution credits the conversion event; revenue attribution credits the money.

The gap matters because conversions aren't equal. A channel can drive a flood of low-value signups and look like a hero on conversion attribution, while a channel that drives fewer but far larger paying customers looks mediocre — until you weight by revenue and the ranking flips. "This channel drove 40 signups" and "this channel drove $12,000 in subscriptions" can point at completely different budgets.

Revenue attribution is harder because it requires joining the conversion to an actual payment record — matching the session to the Stripe charge server-side, not just counting a goal completion. That's why most tools stop at conversion attribution: GA4 attributes events, not dollars, because it has no native line to your payment processor. But for deciding where money goes, revenue is the only honest denominator. The practical rule: do conversion attribution, then weight every credit by revenue. That's exactly what Attrifast's revenue attribution does — conversion paths joined to Stripe, credited in dollars.

Longer journeys make the model matter more

A final reason the model choice can't be ignored: the more touches in a path, the more the model swings the answer.

Source: Attrifast 200-site cohort — median touches on the winning path, 2026

A B2B SaaS customer touches a median 6.4 channels before their first purchase; ecommerce buyers just 3.1; creators and publishers 2.3. A 6-touch B2B journey can be credited completely differently by first-touch versus last-touch — six ways to be wrong if you only look at one. A 2-touch ecommerce impulse buy barely moves between models. So single-touch attribution is far riskier for considered B2B purchases than for impulse retail: the more touches you compress into one credited channel, the more information you throw away. If your sales cycle is long and multi-touch, looking at a single model — especially last-touch — is where budgets go to die.

The bottom line

Conversion attribution assigns credit for a sale to the touchpoints that earned it — and the model you choose is a claim about whether discovery, closing, or the whole path matters most. On the same $1,000 sale, first-touch, last-touch, linear, and time-decay name different winners, and in the real world first- and last-touch disagree on anywhere from a quarter to nearly 60% of revenue by channel. But the model is the easy part. The hard part, and the one that actually breaks attribution, is that 34% of conversions get credited to the wrong channel before any model runs, because cookies and referrers were lost upstream. Fix the data layer first — durable first-party identity, source capture that survives referrer loss, and a server-side join to your Stripe revenue — then look at first- and last-touch side by side, weighted by dollars rather than signup counts. That's the difference between attribution that guides your budget and attribution that quietly misleads it.

FAQ

What is conversion attribution?

Conversion attribution is the practice of assigning credit for a conversion — a signup, a purchase, a paying customer — to the marketing touchpoints that led to it. When a customer discovers you through a Google search, comes back via an email, and finally buys after clicking an ad, attribution decides how much of that sale each touch gets credited with. It's how you answer 'which channels are actually driving conversions?' rather than guessing. The answer depends on the attribution model you choose, because different models split the credit differently across the same set of touchpoints. Without attribution, you know you got conversions but not what earned them, which makes every budget decision a coin flip.

What are the main conversion attribution models?

There are five common models. First-touch gives 100% of the credit to the first interaction — good for measuring what creates awareness. Last-touch gives 100% to the final interaction before conversion — the default in most tools, good for measuring what closes. Linear splits credit evenly across every touchpoint. Time-decay weights later touches more heavily than earlier ones, on the logic that recent interactions mattered more to the decision. Position-based (U-shaped) gives extra weight to the first and last touches and splits the rest among the middle. Data-driven models use your actual conversion data to assign fractional credit statistically. Most SMB teams run last-touch by default, but the 'right' model depends on whether you're trying to measure discovery or closing.

What is the difference between first-touch and last-touch attribution?

First-touch attribution credits the entire conversion to the first channel a customer ever interacted with — the one that created awareness. Last-touch credits it entirely to the final channel before they converted — the one that closed. On the same customer journey they often name completely different winners: first-touch might credit a blog post found via organic search, while last-touch credits the retargeting email they clicked to buy. In our cohort, first- and last-touch disagreed on 29–58% of revenue depending on the channel, with AI engines showing the biggest gap because they tend to be discovery touches that rarely get last-click credit. Neither is 'correct' — they answer different questions, which is why serious teams look at both.

What is the difference between conversion attribution and revenue attribution?

Conversion attribution credits the conversion event — a signup, a purchase, a form fill — to the touchpoints that drove it. Revenue attribution goes one step further and credits the actual money: not just 'this channel drove 40 signups' but 'this channel drove $12,000 in paid subscriptions.' The distinction matters because conversions aren't equal — a channel that drives many low-value signups can look great on conversion attribution and poor on revenue attribution. Revenue attribution requires joining the conversion to a payment record (a Stripe charge), which is why it's harder and rarer. For deciding where to spend, revenue attribution is the more honest number because it weights each conversion by what it was actually worth.

How does conversion attribution work technically?

It works by recording every touchpoint in a visitor's path and then applying a model to split credit when they convert. On each visit, analytics captures the source (referrer or UTM parameters) and stores it against a visitor identifier. When the visitor converts, the tool has an ordered list of their touches — say organic search, then email, then direct — and the chosen model distributes credit across them. The fragile part is maintaining that path: if the visitor is identified by a cookie that gets blocked, or a referrer that gets stripped, touches drop out of the path and credit lands on the wrong channel. Reliable attribution needs a durable first-party identity and, for revenue, a server-side join to the payment.

Why is conversion attribution often inaccurate?

Because the data feeding it is incomplete before any model runs. Three things break the touchpoint path: third-party cookie blocking in Safari, Firefox, and Brave erases returning-visitor identity; referrer stripping by AI clients and in-app browsers dumps real sources into Direct; and consent banners suppress a chunk of visitors entirely. In our cohort, 34% of conversions were credited to the wrong channel because a source signal was lost — the model was applied correctly to broken data. Choosing a better model doesn't fix this; a perfect last-touch calculation on a path missing half its touches still gives the wrong answer. The fix is upstream: durable first-party identity and server-side revenue joins, then pick a model.

Which conversion attribution model should I use?

Match the model to the question. Use first-touch when you're evaluating top-of-funnel channels that create awareness — content, PR, AI visibility. Use last-touch when you're optimizing the closing step — retargeting, cart-recovery email. Use linear or position-based when the journey is genuinely multi-touch and you don't want to over-credit either end. For most SMBs the practical answer is to look at both first- and last-touch side by side rather than committing to one, because the gap between them tells you which channels discover customers versus which close them. And whatever model you pick, weight it by revenue, not raw conversions — a model applied to dollar value is far more decision-useful than one applied to signup counts.

How many touchpoints does a typical conversion involve?

It varies sharply by business type. In our 200-site cohort, the median B2B SaaS customer touched 6.4 marketing channels before their first purchase; services and agencies averaged 4.8; ecommerce 3.1; and creators or publishers 2.3. The longer the path, the more the attribution model matters — a 6-touch B2B journey can be credited completely differently by first-touch versus last-touch, while a 2-touch ecommerce impulse buy barely changes. This is why single-touch attribution is riskier for considered B2B purchases than for impulse retail: the more touches you compress into one credited channel, the more information you throw away.

Can Google Analytics do conversion attribution?

GA4 does conversion attribution, but with significant blind spots for SMBs. It offers data-driven and rule-based models and a conversion-paths report, so mechanically it can split credit across touches. The problems are the same data-loss issues that affect all cookie-based tools: GA4 loses returning-visitor identity to third-party cookie blocking, misfiles referrer-stripped sources into Direct, and can't natively join a conversion to the actual Stripe revenue it produced — so it attributes conversions and events, not dollars. For a Stripe-based business that wants revenue attribution rather than event attribution, GA4's model is applied to an incomplete path and stops at the conversion rather than the payment, which is why many teams add a first-party layer on top.

References

1. Google Analytics Help — Attribution and attribution models in GA4
2. HubSpot — Marketing attribution models explained
3. Google Analytics Help — Default channel group definitions for GA4
4. MDN Web Docs — Referer header and Referrer-Policy
5. WebKit — Intelligent Tracking Prevention overview
6. Stripe — Webhook delivery and idempotency
7. SparkToro — Attribution and dark-traffic research
8. Ahrefs — How marketing attribution works
9. StatCounter — Browser market share worldwide
10. Backlinko — ChatGPT and AI search statistics

For the channel-by-channel view, see first-touch vs last-touch attribution, and for the money layer, what is revenue attribution. To run conversion attribution joined to real Stripe revenue on your own site, Attrifast's revenue attribution credits paths in dollars, and the best conversion tracking software comparison covers the tools that can.