FoundationalIndustry primer · Attribution·12 min read

What is attribution? The mostly-broken science of crediting conversions to ads

Attribution is the discipline of assigning credit for a conversion to specific marketing touchpoints. It's also the most-distorted, most-argued-about, and most-misunderstood metric in modern marketing. This primer explains what attribution actually is, the six models in use today, and the gap between teams that trust the platform dashboard and teams that read attribution as a triangulation across multiple methods.

Start here

Attribution is the science of crediting conversions to touchpoints - and it's harder than it looks

A conversion happens. The buyer saw the brand on TikTok 14 days ago, clicked an Instagram ad 3 days ago, opened an email yesterday, then converted via a Google branded search today. Who gets credit? That's the attribution question. The answer depends entirely on which model you use.

Different models produce wildly different credit distributions. Last-click gives 100% credit to the Google search. First-click gives 100% to the TikTok view. MTA gives partial credit to each. MMM doesn't credit touchpoints at all - it estimates channel-level contribution against baseline. Each method has assumptions; each is wrong in different ways.

Modern attribution literacy means knowing which model produces which kind of bias. Last-click flatters bottom-of-funnel and direct response. MTA flatters touchpoints with high frequency. MMM hides creative-level differences. Incrementality (the closest to truth) is expensive and slow. Read them together; trust none of them in isolation.

Common misidentifications

It's not this. It's that.

The most-common confusions, lined up side-by-side.

Not this

Attribution = ROAS

This

Attribution = the model that produces a specific ROAS number; different models = different ROAS numbers

Not this

Attribution = tracking

This

Tracking = capturing the events; attribution = assigning credit across them

Not this

The platform's ROAS is the truth

This

The platform's ROAS is one attribution model's answer - usually flattering itself

Not this

Incrementality replaces attribution

This

Incrementality calibrates attribution - tells you how much to discount the dashboard number

Anatomy

The 6 attribution models in active use

Each model has assumptions about how credit should flow. Knowing the assumptions is the literacy that separates 'reads ROAS' from 'reads attribution'.

Why it matters

Easy to compute, easy to explain, systematically wrong. Over-credits bottom-of-funnel; under-credits upstream awareness.

Concrete example

Buyer sees TikTok ad, clicks email, converts on Google search. Last-click: 100% to Google. Reality: TikTok and email did most of the persuasion.

The gap

The 8 differences between amateur and elite attribution practice

Attribution is where most operators are confidently wrong. The gaps below separate dashboard-watchers from triangulation-practitioners.

Dimension
Amateur
Elite
Default trust
Platform ROAS = the truth
Platform ROAS = one model's answer, flattering itself
Models read
Last-click only
Last-click + MTA + MMM + incrementality - triangulated
Incrementality cadence
Never
Quarterly geo-holdouts per channel
Attribution lift multiplier
Doesn't apply one
Applies measured lift multiplier to platform numbers - usually 0.6-0.8x discount
Knows when which model is wrong
Doesn't think about it
Last-click over-credits bottom; MTA over-credits frequent touch; MMM hides creative variance
Reports up
'Meta says ROAS is 4.2x'
'Meta-attributed ROAS 4.2x; incremental ROAS ~2.6x; recommend funding at incremental rate'
Tool literacy
Uses native dashboards only
Has Northbeam or Triple Whale for MTA + Recast or Haus for MMM + GeoLift for incrementality
Cross-channel decisions
'Meta ROAS is higher than TikTok, shift to Meta'
Reads incremental contribution per channel - Meta's incremental ROAS may be lower than its attributed ROAS

Pitfalls

The most common mistakes

Each one alone is recoverable. Several stacked together break the practice.

Pitfall 1

Trusting one model

Every attribution model is wrong in a specific direction. Trusting one model = funding the bias built into that model. Triangulate across at least 2 (platform + MMM or platform + lift).

Pitfall 2

Comparing across models without normalization

Last-click ROAS 4x and MMM ROAS 2x aren't comparable. Last-click is touchpoint-level; MMM is channel-level. Apples and orchards.

Pitfall 3

Ignoring view-through

Last-click and many MTA models discount view-through (saw but didn't click). For video-heavy DTC, view-through is meaningful - hiding it inflates click-led channels.

Pitfall 4

Treating ATT as solved

iOS 14 / ATT broke user-level tracking. MTA models built on user-level data are systematically degraded post-2021. MMM and lift studies are partial replacements - not perfect ones.

Glossary

Related terms you should know

The vocabulary that surrounds this concept. Bookmark this section.

Attribution

The discipline of assigning credit for a conversion to specific touchpoints.

Last-click

Attribution model giving 100% credit to the final touchpoint before conversion.

First-click

Attribution model giving 100% credit to the first touchpoint.

Linear

Equal credit to every touchpoint.

Position-based

Weighted credit - typically 40% first, 40% last, 20% middle.

MTA (Multi-Touch Attribution)

Data-driven credit allocation across touchpoints, often ML on user-level data.

MMM (Marketing Mix Modeling)

Statistical decomposition of total sales into channel contributions + baseline.

Incrementality

The lift caused by an intervention vs no-intervention control. The truthful version of attribution.

Lift multiplier

Ratio of incremental ROAS to platform-attributed ROAS. Typically 0.5-0.8x.

Attribution window

Time window in which a touchpoint can claim credit - typically 7-day click + 1-day view.

Where Shuttergen fits

Foundational knowledge in. 25 variants out.

Once you understand the discipline at this level, the bottleneck moves to production. Shuttergen turns one validated concept - anchored to your starting image - into 25 brand-safe variants you can test. The strategist stays in the loop; the production grind goes away.

Try Shuttergen free

Related Shuttergen reading

Where to go next

The connected pages that compound on this one.

Sources

What we read to build this

Foundational knowledge. Now ship the variants.

Shuttergen turns understanding into output - one validated concept into 25 brand-safe variants in hours, not weeks.

Start free