The Attribution Illusion

A customer sees a Facebook ad, visits the website, leaves. A week later they discover the company on Google, leave again. Then they receive an email, click through, and purchase.

Which channel deserves the credit?

Last-click gives email full credit. First-click gives Facebook full credit. Linear divides it equally. Time-decay weights the email heavily. Data-driven attribution applies an algorithm that most users cannot audit. All of them look at the same journey and produce different answers.

This is why attribution is one of the most misunderstood areas of email marketing. The numbers look precise because they appear in a dashboard with decimal places. The reality is that every attribution model translates assumptions into percentages. The precision is an illusion.

---

Why Attribution Is Fundamentally Hard

Customer purchase behaviour is almost never linear. A single transaction can involve multiple devices, multiple channels, multiple sessions spread across days or weeks, and multiple decision-makers in B2B contexts.

Factor	What It Means for Attribution
Cross-device journeys	A customer may research on mobile and purchase on desktop. Most attribution tools cannot connect these sessions to the same person.
Non-linear paths	The seventh touchpoint may matter more than the first, or the first may set the entire journey in motion. No model handles both cases well.
Offline influence	A podcast, billboard, or conversation may drive a search that leads to purchase. No online attribution model captures the offline starting point.
Long decision windows	A B2B purchase may involve 20+ touchpoints over three months. Recency-heavy models ignore the early work entirely.
Self-attribution bias	Direct traffic is often misattributed. The customer who types the URL directly was likely influenced by another channel first.

Attribution compresses all of this complexity into a single number. Useful for reporting — but not an accurate representation of what happened.

---

A Worked Example: One Purchase, Six Answers

Let's look at actual numbers.

A customer takes this journey:

1. Organic search — Searches "email marketing software", reads a blog post
2. Paid social — Clicks a retargeting ad, visits pricing page
3. Organic search — Searches "[company] vs [competitor]", reads comparison page
4. Email — Receives abandoned browse email, clicks through to features page
5. Direct — Types the URL, purchases the $1,000 plan

Same customer. Same journey. Six attribution models produce six different answers:

Model	Email Attributed Revenue	Who Gets the Credit
Last-click	$0	Direct ($1,000)
First-click	$0	Organic #1 ($1,000)
Linear	$200	All equal ($200 each)
Time-decay (7-day half-life)	~$380	Email ($380), Direct ($310), Organic #2 ($160)
U-shaped (40/20/40)	~$67	First & last ($400 each), middle three ($67 each)
Data-driven	Unknown	Depends entirely on the black box

Email's attributed revenue ranges from $0 to $380 for the exact same purchase. The only variable that changed was the model. If you report last-click, email looks weak. If you report time-decay, email looks dominant. Both reports are derived from the same data.

Most teams pick one model and present its output as truth. They are not lying — they are looking at one slice of a picture they cannot see.

---

The Attribution Model Zoo

Model	How It Works	What It Rewards	Best For	Worst For
Last-click	Full credit to the final touchpoint	Channels that close	Short sales cycles, direct response	Long cycles, brand building
First-click	Full credit to the first touchpoint	Awareness channels	Top-of-funnel analysis, content	Channels that convert later
Linear	Equal credit to every touchpoint	Channels that appear anywhere	Understanding distribution	Campaigns with many touchpoints
Time-decay	More credit closer to conversion	Late-stage channels	Short consideration windows	Early trust-building (email nurture)
U-shaped	40% to first, 40% to last, 20% to middle	Channels that open and close	Lead gen with clear funnel stages	Journeys where middle touchpoints matter most
Data-driven	Algorithm assigns credit statistically	Channels that lift conversion probability	Large datasets, full tracking	Small datasets, black-box methodology

The key insight: each model produces a different answer from the same data. A campaign that looks heroic under last-click may look average under linear and weak under first-click. None is wrong — they are all correct within their assumptions.

---

Incrementality: The Question Attribution Cannot Answer

Attribution tells you what channels were present in the journey. It cannot tell you whether the customer would have purchased without any of them.

This is the most important gap in marketing measurement.

Incrementality measurement — typically through holdout groups — answers the question attribution cannot: "What would have happened if this campaign had not existed?"

A real example:

A SaaS company sends a promotional email. Last-click attribution shows the email drove $50,000 in revenue. The campaign looks like a clear winner.

They run a holdout test: 10% of the list is randomly excluded from the send. After two weeks, the holdout group generated $42,000 in organic revenue — purchases they would have made anyway without the email.

True incrementality: $8,000.

The email did not drive $50,000 in new revenue. It accelerated $42,000 that was already going to happen and generated $8,000 in truly incremental purchases. The attribution model — any attribution model — cannot distinguish between acceleration and creation. Only a holdout test can.

How to run a holdout test:

1. Randomly split your audience into control and treatment before scheduling a campaign
2. Suppress the control group from the campaign
3. Measure the difference in conversion rate and revenue between the two groups
4. The difference is your true incrementality

Do not overcomplicate this. Even one well-designed holdout test per quarter gives you more useful information than the most sophisticated attribution model running on the same data every day.

---

The Two Email Attribution Traps

Email's position in the customer journey creates two opposite errors that often exist simultaneously.

Trap 1 — Over-credit. Email is frequently the final touchpoint — not because it created the desire to purchase, but because it happened to arrive at the right moment with a reminder. Last-click gives email full credit for conversions that multiple channels influenced. This creates a feedback loop: the model shows email as top channel, the team increases email investment and cuts others, demand declines because acquisition channels were quietly defunded. The model does not tell you it is eating its own source of demand.

Trap 2 — Under-credit. A subscriber joins, receives nurture content for six months, builds a relationship, then types the URL directly and purchases. Most models credit this as direct traffic. Email gets nothing. The relationship, trust, and intent were built through email, but the model records only the final click. It cannot see why the customer decided to buy.

The assisted conversion blind spot. Assisted conversion reporting looks at all touchpoints rather than just the final one. This reveals email's influence more fully:

Email's Role	Last-Click View	Assisted View
Closing conversions	Visible — full credit	Visible — counted once
Influencing conversions	Invisible — zero credit	Visible — appears as assist
Nurturing over time	Invisible	Partially visible

Teams that rely on a single attribution view see a fraction of email's actual contribution.

---

The Attribution Audit Checklist

Before debating which model to use, validate your data. This is the most overlooked step.

UTM hygiene:

• Every campaign uses UTMs consistently
• Naming conventions are enforced (email ≠ Email ≠ e-mail ≠ newsletter)
• All three standard parameters are populated on every outbound link
• Automated campaigns (welcome, abandoned cart, re-engagement) have UTMs

Link validation:

• Links are tested before every send
• UTMs survive redirects and link shorteners
• ESP tracking templates pass UTMs through to the landing page

Analytics configuration:

• Cross-domain tracking is configured if journeys span multiple subdomains
• Session stitching across devices is understood (most tools cannot do it)
• Your "direct traffic" definition is accurate (typing URL vs. unknown referrer)

Reporting audit:

• Channel totals match between attribution tool and analytics tool
• Spikes in "direct" or "other" are investigated (they are usually misattributed email)
• Reports are compared across tools to surface inconsistencies

Most teams spend more time debating models than validating tracking. The best model cannot compensate for data that says social when the channel is email.

---

The Email Attribution Maturity Model

Level	Description	What You Do	Next Move
1. No tracking	UTMs missing or wrong	Campaigns run without tracking	Standardise UTMs, add to every campaign
2. Last-click only	Default reporting, no alternative views	All decisions from last-click	Add assisted conversion reports, run one holdout test
3. Multi-touch aware	Using linear, time-decay, or U-shaped	Comparing multiple models	Invest in tracking hygiene, standardise methodology
4. Data-driven + experiments	Algorithmic attribution + holdout tests	Regular incrementality testing	Integrate findings into budget decisions
5. Holistic measurement	Attribution + MMM + experimentation	Triangulating across methods	Share methodology, contribute to industry standards

Most teams sit at Level 2. The highest-impact move is not jumping to Level 4 — it is getting Level 1 right and running one Level 4 experiment.

---

The Attribution Playbook

Stop searching for the perfect model. Use multiple views and make decisions based on direction, not precision.

Step 1: Fix tracking. Run the audit checklist. Inconsistent UTMs make everything else meaningless.

Step 2: Pick one model for trend comparison. Linear is a good default. The absolute numbers will be wrong, but the trend direction over time is useful. Email attributed revenue trending up or down is valuable information even if the precise number is uncertain.

Step 3: Run assisted conversion reports. Understand where email appears in the journey, not just where it closes. If email assists 3x more conversions than it closes, you are underinvesting.

Step 4: Run one holdout test per quarter. Incrementality is the closest thing to truth in marketing measurement. One well-designed test tells you more than months of attribution data.

Step 5: Track revenue per subscriber over time. This captures the program's health without requiring precise attribution. If it is growing, the program is working regardless of what the attribution model says.

Step 6: Compare channels using the same model. Internal consistency matters more than accuracy. Comparing email under last-click with search under linear produces a meaningless comparison.

---

What to Focus On Instead

Attribution is a tool for direction, not a machine for truth. The most useful questions have less to do with precise splits and more to do with whether the program is improving:

• Are subscribers becoming more engaged or less?
• Is revenue per subscriber increasing?
• Is email's contribution to overall revenue growing?
• Is acquisition quality (retention, LTV) improving?

A team that answers these four questions will make better decisions than a team that obsesses over whether email deserves 42% or 47% of a conversion.

The goal is not perfect attribution. The goal is good enough data to make better decisions than you could make without it. Attribution is a lens, not a measurement. Use it to see more clearly, not to pretend you have perfect vision.

---

Related Articles

• The Email Metrics That Actually Matter
• The Hidden Cost of Bad Email Reporting
• Why Email Benchmarks Are Dangerous
• Revenue Per Subscriber: The Metric That Changes Everything
• The Missing Layer in Most Email Reporting: Time

---