Attribution vs. Incrementality: They Answer Different Questions, and You Need Both

Incrementality and attribution get discussed as though they were competing lenses viewing the same data. They are not. They are designed to answer different questions, using different forms of evidence.

• Attribution asks which observed marketing touchpoints should receive credit for a conversion.
• Incrementality asks whether the marketing activity caused additional conversions that would not have occurred without it.

A refresher on attribution

Attribution was the favoured child of marketing analytics teams circa 2015. Marketers discovered that conversion paths contained multiple touchpoints:

Display → Paid Social → Organic Search → Email → Purchase

Attribution modelling provided frameworks for distributing that credit.

Splitting a $100 conversion

Model	Display	Paid Social	Organic Search	Email	How credit is assigned
First-touch	$100	$0	$0	$0	All credit to the first observed interaction
Last-touch	$0	$0	$0	$100	All credit to the final interaction before purchase
Linear	$25	$25	$25	$25	Divided equally among every observed touchpoint
Position-based	$40	$10	$10	$40	First and last weighted most, middle splits the rest
Time-decay	$10	$20	$30	$40	Progressively more credit closer to conversion
Data-driven	$30	$20	$20	$30	Distributed by estimated contribution

These are simplified examples. Position-based models can use different weighting rules, time-decay allocations depend on timing, and actual data-driven models vary.

Incrementality 101

Incrementality regained interest around 2020. Rather than doling out credit by equation, it relies on carefully guardrailed tests against real, live sales data.

How many of these sales were actually caused by this campaign — not counting how many would have happened regardless?

The answer is lift.

The example that makes it concrete

Say you want to discover the lift of a campaign. Divide your audience into two groups:

• A control group not exposed to the campaign
• An exposed group that sees it

Run the campaign for 30 days, then look at the results.

Group	Purchases
Exposed	1,000
Control	800
Incremental lift	200

Here is the crux:

An attribution model could associate many or all 1,000 purchases with the campaign, allocating value across the platforms and touchpoints involved according to the model you choose.

Incrementality would conclude that only those 200 additional purchases were actually caused by the campaign.

Same campaign, same data, a fivefold difference in conclusion. That is why conflating the two frameworks is dangerous.

Using them together

Where marketers go wrong is when they go all in on either framework. The two play nicely together — provided you use the right one to answer the right question.

• If you are optimising campaigns or deep-diving customer journeys, attribution is your best friend, giving a shared success metric for comparing platforms and touchpoints.
• If you are defending your budget from a proposed cut, incrementality is your strongest evidence regarding which channels actually create additional business rather than capturing business that would have happened anyway.

Side by side

	Attribution	Incrementality
Primary question	Which observed touchpoints should receive credit for a conversion?	How many additional conversions occurred because of the marketing activity?
Best use case	Ongoing campaign optimisation, understanding customer journeys, allocating credit across measurable channels	Validating whether an investment creates additional business value, informing higher-level budget decisions
Main blind spot	Correlation is not causation — a touchpoint may receive credit for a conversion it did not create	Tests can be expensive, slow or hard to design, and results may not explain which individual touchpoints influenced the customer
Most likely stakeholder	Channel managers, performance marketers, platform teams, marketing analytics	Marketing leadership, finance, data science, growth strategy, budget owners

"Why don't these numbers match?"

The question that has haunted the original author throughout her career.

It gets asked when a platform's reported revenue or conversions differ from the client's CRM, web analytics, or other source of truth. They almost never line up perfectly.

What is hard to explain succinctly:

The fact that they differ doesn't necessarily mean either is incorrect.

Each system applies its own logic based on the interactions it can observe, which conversions should qualify for credit, and how long after an interaction credit can still be attributed. Neither is "wrong."

An advertising platform may correctly observe and report that a customer viewed or clicked an ad before purchasing. But:

Evidence that an ad was seen before a purchase isn't necessarily proof that the ad caused it, nor is it proof that the ad didn't cause it.

Why this matters more with automation

The distinction becomes particularly important as platforms continue to push automated solutions.

Automated systems are designed to maximise performance based on the conversion signals defined inside the platform. They are simply not designed to maximise performance based on your carefully calculated incremental lift test results.

As a result, automated campaigns target audiences, placements and queries already associated with users likely to convert — existing customers, branded searchers, remarketing audiences.

Those conversions may be entirely valid according to the platform's attribution model, while creating less additional revenue than the campaign report implies.

In other words, automated campaigns can increase the number of conversions credited to a given campaign without actually causing an equal increase in total sales.

A note on in-platform lift studies

Platforms are increasingly offering lift studies and other incrementality-focused tools.

But it is a mistake to assume incremental value is automatically incorporated into automated campaign optimisation.

"Data without insights is meaningless, and insights without action are pointless."

A lift study will only affect performance if and when someone applies its findings to the campaign's objectives, inputs, or budget decisions.

Google Ads may provide a controlled lift experiment, but unless the advertiser applies the findings — or selects a campaign setting explicitly designed to optimise for incrementality — the measurement system and the delivery system are still working toward two different definitions of success.

Some platforms are addressing this. Meta now offers an incremental attribution model intended to optimise delivery toward conversions it predicts were directly caused by advertising. For now, though, that is a specific optimisation choice requiring advertiser action, not an inherent feature of every automated campaign.

Know which question you are answering

Attribution and incrementality aren't competing methods for finding one definitive metric. They're different tools designed to answer different questions.

You wouldn't try to use your Allen wrench as a hammer, would you?

• Attribution helps marketers understand which touchpoints contributed to a conversion and provides a shared basis for comparing channels
• Incrementality helps businesses understand whether their investment generated additional conversions that would not have occurred otherwise

The best marketers need both. Attribution provides the ongoing signals needed to optimise campaigns; incrementality validates whether those optimisations are creating new business value or simply capturing demand that already existed.

As automated campaigns take greater control over targeting, placements, bidding and budget allocation, understanding both sides will only become more important.

A system can become exceptionally efficient at maximising attributed conversions without becoming equally effective at producing incremental growth.

So the next time a platform report contradicts your CRM, don't assume either number is wrong — ask which question each number was designed to answer.

Practical guidance

1. Do not blend the two questions. When someone asks "how is this channel performing?", establish first whether that is a credit question or a causal one.

2. Schedule incrementality tests around big decisions. They are not a weekly instrument. Run them before major budget shifts or new channel investments.

3. Treat an attribution model change as a baseline change. Switching models makes past performance look different. Record the change date and separate comparison periods.

4. Do not read automated campaign conversions as incremental. Automation drifts toward easy conversions by design, and blended brand search magnifies the distortion — see Why Separating Brand and Non-Brand Campaigns Makes ROAS Honest.

5. Act on lift studies you run. Measuring without changing campaign objectives leaves the two systems chasing different definitions of success.

6. Secure data quality first. Missing parameters make both the attribution model and the test control group untrustworthy — see Google Analytics Adds Campaign Diagnostics for Missing Aggregate Identifiers.

7. Build a separate rule for turning numbers into decisions — the subject of A Dashboard Does Not Finish the Work.

8. Judge non-linear channels like SEO on blended metrics. Why channel-level CAC hides the contribution is covered in How SEO Lowers Blended CAC.