Attribution

You’ve tagged every link, governed every value, and your analytics finally shows clean, consistent campaign data.

Now the harder question. A customer touches four campaigns before buying. Which one gets the credit? Which one gets the budget?

That’s attribution, and it’s where the tracking work pays off. It’s also where clean data gets misread: pick the wrong model, or never realize you picked one, and perfectly tagged campaigns still produce misinformed budget decisions. By the end of this page you can name the model behind your reports, say what it systematically exaggerates, and run the comparison that exposes it.

What Attribution Models Are

An attribution model is a set of rules for distributing conversion credit across marketing touchpoints. A customer interacts with a paid search ad, then an email, then a retargeting ad, then buys. The model decides which of those touchpoints “get credit.”

That choice drives budget directly. Give all credit to the last touchpoint and bottom-funnel channels look disproportionately valuable while top-funnel channels appear to contribute nothing. The model you choose shapes the story your data tells, and the budget follows the story.

Common Attribution Models

• Last-click: all credit to the final touchpoint before conversion. The simplest model and the default in many platforms. Answers “what closed the deal?” and ignores everything before it.
• First-click: all credit to the first touchpoint. Answers “what brought them in?” and ignores nurturing and closing.
• Linear: equal credit across every touchpoint. A four-touch journey assigns 25% to each. Fair, but treats all interactions as equally important.
• Time-decay: more credit to touchpoints closer to conversion. Useful for long sales cycles where recent engagement signals stronger intent.
• Position-based (U-shaped): 40% to first touch, 40% to last, 20% distributed across the middle interactions. Balances acquisition and closing credit.
• Data-driven: GA4’s default. Machine learning analyzes your actual conversion paths and assigns credit based on observed patterns. Needs sufficient conversion volume to be reliable.

Model Comparison: One Journey, Five Stories

A prospect discovers your product through a LinkedIn ad, receives a nurture email a week later, reads a blog post via organic search, then converts after clicking a retargeting ad. The deal is worth $10,000. Watch each model allocate that revenue:

Touchpoint	utm_source / medium	Last-Click	First-Click	Linear	Position-Based	Time-Decay
LinkedIn ad	linkedin / paid_social	$0	$10,000	$2,500	$4,000	$1,000
Nurture email	newsletter / email	$0	$0	$2,500	$1,000	$1,500
Blog (organic)	google / organic	$0	$0	$2,500	$1,000	$2,500
Retargeting ad	google / display	$10,000	$0	$2,500	$4,000	$5,000

What that means at budget time:

• Under last-click, you’d double down on retargeting and cut the LinkedIn budget, even though LinkedIn created the initial awareness.
• Under first-click, LinkedIn looks like the hero and retargeting looks worthless, even though it closed the deal.
• Position-based pays both bookends, showing that the full journey matters.
• Time-decay favors recent touches, which makes sense for long sales cycles where recency signals intent.

Here’s the opinion the table is too polite to state: last-click is wrong for almost every business, and it’s probably what you’re using. It’s the default, and defaults win. No single model is “right,” but some are reliably misleading, and the misleading one is the one nobody remembers choosing. So run the same data through several models and watch where the credit lurches; a big swing marks a channel your current model is mispricing.

1. Ad Click
2. →
   Email
3. →
   Blog
4. →
   Retargeting
5. →
   CONVERT

Model	Ad Click	Email	Blog	Retargeting
Last-click	0%	0%	0%	100%
First-click	100%	0%	0%	0%
Linear	25%	25%	25%	25%
Position-based	40%	10%	10%	40%
Time-decay	10%	15%	25%	50%

Follow Along: Stride Footwear

The spring sale campaign runs for six weeks. A customer named Alex discovers Stride through a Facebook ad, receives the weekly newsletter a week later, clicks the Google search ad for “stride running shoes” two weeks after that, and finally buys a pair of trail runners after clicking the retargeting ad on Instagram. The order is worth $140.

Here’s what each attribution model says about which channel deserves credit for that $140:

Touchpoint	UTM values	Last-Click	First-Click	Position-Based
Facebook ad	facebook / paid_social	$0	$140	$56
Newsletter	newsletter / email	$0	$0	$14
Google search	google / cpc	$0	$0	$14
Instagram retargeting	instagram / paid_social	$140	$0	$56

The position-based math, spelled out: 40% of $140 is $56 to each bookend (the Facebook ad and the Instagram retargeting), and the remaining 20% splits across the two middle touches, $14 apiece. Multiply this across 800 orders and the budget story changes dramatically. Under last-click, Dana would conclude Instagram retargeting is the top performer and consider cutting the Facebook awareness budget. Under first-click, Facebook looks like the hero and retargeting looks like waste. Under position-based, both bookends share credit, and the full journey stays visible.

At the board meeting, Dana presents the position-based view and notes: “Our Facebook ads create demand. Our retargeting closes it. Cutting either one breaks the chain.” The CEO asks for the incrementality test she’s planning for Q3; she’s already thinking about what UTMs can’t measure (see “What UTMs Can’t Track” later in this chapter).

How UTMs Feed Attribution

Every UTM-tagged click creates a touchpoint in the attribution path. A click on utm_source=linkedin&utm_medium=paid_social&utm_campaign=q1_awareness is recorded as one node in the conversion journey, and the model uses those nodes, potentially across multiple sessions and days, to assign credit.

Which means UTM consistency is not optional for attribution:

• Consistent tagging = clean touchpoints. If LinkedIn ads always use utm_source=linkedin, attribution models can accurately aggregate credit for that source across campaigns.
• Inconsistent tagging = fragmented attribution. If some links use linkedin, others LinkedIn, others li, the model treats these as three separate sources. Credit scatters across phantom channels that are one and the same thing.
• First-touch vs. last-touch capture in forms. CRM and marketing automation platforms typically capture UTM values at form submission. Some store only the current session’s values (last-touch); others store the original visit’s values (first-touch). Capture both. First-touch tells you what drove awareness; last-touch tells you what triggered the conversion.
• Multi-touch attribution requires persistent user identity. Models beyond first- and last-click need to connect multiple sessions from the same user via cookies, login state, or other identifiers. A user visiting on three devices without logging in appears as three separate users, and multi-touch attribution breaks down.

Treat Attribution as a Lens, Not a Fact

• No single model is correct. Last-click tells you what converts; first-click tells you what acquires; linear and position-based give a holistic picture. Choose based on the decision you are making.
• Last-click overvalues bottom-funnel channels. Branded search and retargeting will always look like heroes under last-click because they are typically the final touch. They may simply be capturing demand that other channels created.
• First-click overvalues top-funnel channels. Awareness campaigns can appear solely responsible for revenue even when significant nurturing was required to close.
• Start with last-click, then compare. It’s the default in most tools, so it’s your baseline. Then run your data through whatever other models your platform still offers and see how the story changes: Adobe’s Attribution IQ keeps the full menu (first touch, linear, U-shaped, time decay, and more), while GA4 cut its list to data-driven and last click in November 2023, so there the comparison is one switch. Large discrepancies mark channels whose true contribution is ambiguous.
• Data-driven attribution is a strong default if you have the volume. Google publishes no minimum for GA4’s model; the 400-conversions-a-month threshold still quoted around the industry was a Universal Analytics requirement. The mechanics haven’t changed, though: a property recording a few dozen key events a month (a key event is what GA4 has called a conversion since 2024) gives the model very little to learn from.
• Privacy restrictions degrade attribution accuracy. ITP, cookie consent requirements, and shortened cookie lifespans reduce the ability to connect sessions across time, and multi-touch models lose fidelity as cross-session tracking becomes less reliable. Chapter 12 goes deeper.
• Attribution is one input, never the verdict. Combine it with incrementality testing, qualitative feedback, and business judgment.

Attribution modeling is the part of marketing analytics where the math outpaces the usefulness. Don’t let a model tell you something your judgment already knows.

What UTMs Can’t Track, and What to Use Instead

UTM parameters measure clicks. That’s their strength and their fundamental limitation, because a significant portion of marketing activity generates value without producing a trackable click. Pretend otherwise and you’ll systematically underfund the channels that build demand in the first place.

Touchpoints that UTMs cannot measure:

Activity	Why UTMs miss it	What actually happened
Display/video ad impressions	No click = no UTM data. A prospect sees your brand 12 times before ever clicking.	Brand awareness was built, but only the final click gets credit
Podcast listens	A listener hears your ad, remembers your brand, and searches for you directly later. No link was clicked.	The podcast drove a branded search visit that gets attributed to “organic search”
Conference/event attendance	Someone meets your sales team at a booth, has a great conversation, and visits your site from their hotel that night by typing your URL.	The event drove the visit, but analytics records it as “direct”
Word-of-mouth / referrals	A customer recommends you to a colleague who searches your name.	The recommendation drove the visit, but analytics credits “organic search”
Billboard / TV / radio	Mass media builds awareness that manifests as branded search, direct visits, and social mentions, none of which carry UTMs.	Your brand campaign is working, but click-based attribution says it’s contributing nothing
Social impressions (no click)	Someone sees your LinkedIn post, doesn’t click, but remembers your product when they have the need a month later.	The impression contributed to the eventual conversion, but it left no trackable trace
Dark social sharing	A prospect forwards your article to their boss via Slack. The boss reads it and later visits directly.	Your content drove a decision-maker to your site, but it’s invisible in analytics

The pattern: UTMs capture the last mile, the click, and miss the miles that led to it. If you invest in brand, content, events, or any top-of-funnel activity, click-based attribution alone will systematically undervalue those investments.

Complementary measurement approaches:

Brand lift studies. Controlled experiments run by ad platforms (Google, Meta, LinkedIn all offer them) that measure the incremental impact of ad exposure on brand awareness, consideration, and recall, independent of clicks. Available for campaigns above certain spend thresholds. Best for: validating that display and video campaigns are moving perception.

Marketing mix modeling (MMM). Statistical analysis of aggregate data (spend, impressions, revenue, seasonality, competitive activity) to estimate each channel’s contribution to business outcomes. Doesn’t rely on user-level tracking at all; it works from time-series data. Increasingly important as privacy restrictions degrade click-level attribution. Best for: budget allocation across channels, especially when a significant share of spend goes to non-clickable media.

Incrementality testing. Holdout experiments: suppress marketing activity for a randomly selected group and compare their behavior to the exposed group. The difference is the true incremental impact of that channel. The gold standard for measuring causal impact. Best for: answering “what would happen if we stopped spending on X?”, particularly for retargeting, where attribution models heavily overstate value.

Survey-based attribution (“How did you hear about us?”). A simple question on your signup form, demo request, or checkout page. Low-tech, and it captures signals that no tracking system can: word-of-mouth, podcast listens, conference conversations, social impressions. Best for: capturing offline and untrackable touchpoints that analytics misses entirely. Pair with UTM data for a more complete picture.

Post-purchase surveys. Deeper than a single attribution question: ask new customers which content, events, or interactions influenced their decision. Reveals the qualitative journey that quantitative data flattens. Best for: understanding the why behind the conversion, not just the what.

The Measurement Stack

Method	Precision	Coverage
UTM / click-based attribution (measures exactly what was clicked)	High	Low (clicks only)
Survey-based attribution (captures untrackable touchpoints)	Low	High (everything)
Incrementality testing (measures causal impact)	High	Medium (one channel at a time)
Marketing mix modeling (estimates channel contribution from aggregate data)	Medium	High (all channels together)

No single method is sufficient. The strongest measurement programs layer multiple approaches.

UTM-based attribution is the foundation: precise, actionable, and free. It is not the ceiling. Make serious budget decisions on click data alone and you’re working from a partial picture, one where the channels that build demand (brand, content, events, podcasts) always look undervalued and the channels that capture demand (branded search, retargeting) always look like heroes. Knowing the bias exists is the first step to compensating for it.

The measurement question isn’t settled.

The click is where your data ends, and it was never where the decision began.