Reclaiming Attribution From Google AI Max: A CMO's Guide to Measuring What Actually Drives Growth

AI can optimize your advertising. But your company still needs an independent way to determine whether that advertising actually created incremental growth.

The problem isn't AI. It's who gets to define success.

I've spent enough time around paid acquisition to know that a beautiful ROAS dashboard can hide an ugly business reality.

A platform can tell you that performance improved.

Your agency can tell you that CAC is down.

Your attribution tool can tell you that three different channels contributed to the same customer.

And then the CFO asks the question that matters:

"Did marketing actually create more revenue?"

That's a much harder question to answer.

The shift toward AI-powered advertising makes it even more important.

Google's AI Max is changing how Search campaigns find queries, generate assets, and expand landing-page coverage. Google says advertisers using AI Max typically see more conversions or conversion value at similar CPA or ROAS, based on its internal data.

Those capabilities can be useful.

But there's a bigger strategic question sitting underneath the automation:

Who is measuring whether those conversions were actually incremental?

That's where I think CMOs need to rethink the role of attribution.

The answer isn't to reject Google, Meta, or automated bidding.

The answer is to stop asking the platforms to be both the growth engine and the referee.

---

The attribution problem every CMO eventually encounters

Imagine you are running $20 million a year through paid acquisition.

Google reports a strong ROAS.

Meta reports another strong ROAS.

LinkedIn reports meaningful pipeline contribution.

Your retargeting platform reports impressive conversion efficiency.

Your attribution platform then distributes credit across all of them.

Everything looks good.

Until you look at the business.

Revenue hasn't grown nearly as much as the dashboards suggest.

This isn't necessarily because anyone is doing anything wrong.

It's a structural problem.

Every platform has access to a different slice of the customer journey. Each platform also has its own optimization system, conversion definitions, modeling assumptions, and attribution methodology.

Google sees one part of the journey.

Meta sees another.

Your CRM sees what happened after the lead became a customer.

Your finance team sees the money.

Those aren't the same dataset.

And they shouldn't be treated as if they are.

---

AI-powered advertising makes measurement more important, not less

This is where the conversation around Google AI Max becomes interesting.

AI Max introduces more automation into Search campaigns, including expanded search-term matching, AI-assisted asset customization and Final URL expansion. Google has also been adding additional controls, reporting and experimentation capabilities as the product evolves.

That's not inherently a problem.

In fact, automation can be extremely valuable.

I don't want my team manually adjusting thousands of bids.

I don't want marketers spending their days trying to predict which auction will happen five seconds from now.

Let the machines do the work they're good at.

But there is an important distinction:

Optimization and measurement are not the same thing.

Google's algorithm can be excellent at finding users likely to convert.

That doesn't automatically mean Google created the entire value of those conversions.

A customer may already know your brand.

They may have seen your product through another channel.

They may have searched for you organically.

They may have been influenced by sales, PR, content, referrals or word of mouth.

The ad might have been the final touch.

The question is whether it was the incremental cause.

That's a very different question.

---

The difference between attributed revenue and incremental revenue

This distinction is becoming one of the most important concepts in modern marketing.

Attributed revenue asks:

> "Which channel can claim credit for this conversion?"

Incremental revenue asks:

> "What additional revenue happened because we spent this money?"

The first question is useful for campaign optimization.

The second question is much closer to what the CFO actually cares about.

Consider a simple example.

A customer searches for your company by name and clicks a paid ad before purchasing.

Google may reasonably attribute that conversion to the campaign.

But what happens if you stop buying that branded click?

Does the customer disappear?

Or do they simply click the organic result underneath it?

If the customer would have purchased anyway, the advertising interaction may have been attributed revenue without creating much incremental revenue.

That's the distinction CMOs need to understand.

Attribution tells you who touched the customer. Incrementality tries to tell you what changed the customer's behavior.

---

Stop asking the ad platform to prove its own value

This doesn't mean platform reporting is useless.

Far from it.

Platform reporting is extremely valuable for managing campaigns.

You need it to understand:

• Which audiences are responding
• Which creative concepts are working
• Which queries are producing qualified traffic
• Which campaigns are scaling
• Where conversion rates are changing
• How efficiently the platform is spending your budget

But those reports shouldn't necessarily be the final authority on whether an entire channel deserves another $5 million.

That's a different level of measurement.

The bigger the budget, the more important this distinction becomes.

A $10,000 campaign can be optimized from a dashboard.

A $50 million acquisition program needs an independent measurement layer.

---

What the modern CMO measurement stack looks like

I think about this as three connected layers.

1. Platform measurement

This is the tactical layer.

Google, Meta, LinkedIn, and other platforms tell you what's happening inside their environments.

Use this data.

Optimize against it.

But understand what it is:

platform measurement.

It isn't the complete picture of business impact.

---

2. Experimentation and incrementality

This is where things get more interesting.

Instead of asking an algorithm whether its advertising worked, you create situations where you can measure what changed when advertising changed.

That might include:

• Geo experiments
• Holdout groups
• Regional spend tests
• Conversion lift studies
• Audience experiments
• Incrementality tests
• Controlled budget changes

The methodology depends on the business, market, and available data.

But the underlying question stays the same:

What happened because we changed the marketing input?

That is much closer to causal measurement than simply counting conversions.

---

Geo-testing can turn a debate into evidence

One of the most practical approaches for larger advertisers is matched-market testing.

Imagine two markets with historically similar performance.

You continue normal advertising in one.

You materially change spend in the other.

Then you compare the difference in business outcomes.

The goal isn't to prove that advertising generates zero value.

The goal is to estimate how much additional value advertising generated.

A simplified incremental ROAS calculation looks like this:

iROAS = incremental revenue ÷ incremental advertising spend

The important word is incremental.

If reducing $100,000 in advertising produces a measurable decline in revenue, that's evidence that the spend was creating incremental demand.

If revenue barely changes, you have a very different conversation.

And that's exactly the kind of conversation a CMO should want to have before scaling the budget.

---

Modern MMM is becoming a practical executive tool

Marketing Mix Modeling used to have a reputation for being expensive, slow and difficult to maintain.

That is changing.

Open-source frameworks such as Google's Meridian and Meta's Robyn have made modern MMM much more accessible to sophisticated marketing organizations. Google's Meridian, for example, is designed for in-house modeling and uses aggregated data rather than cookie-level or user-level tracking. It can also incorporate experiments to help calibrate models.

Meta's Robyn similarly provides an open-source framework for marketing mix modeling, including modeling of media efficiency, adstock, saturation, and budget allocation.

The interesting part isn't the technology itself.

It's what happens when a CMO starts using it as a capital allocation system.

Instead of asking:

> "Which campaign has the highest reported ROAS?"

you start asking:

> "Where does the next dollar of marketing investment generate the greatest incremental return?"

That's a much more strategic question.

---

MMM and incrementality work better together

I wouldn't treat MMM and experimentation as competing methodologies.

They solve different problems.

Experiments can provide stronger evidence about causal impact in specific situations.

MMM can provide a broader view across channels, markets, and time.

Experiments can help calibrate the model.

The model can help put individual experiments into a wider business context.

That's where the measurement system starts becoming genuinely useful.

You're no longer trying to find one magical attribution model.

You're building a measurement framework that gets progressively better as evidence accumulates.

---

The other major shift: stop optimizing for revenue alone

Here's another problem I see regularly.

Marketing tells Google:

"A customer is worth $500."

But the business knows that's not really true.

One customer may generate $500 in revenue with an 80% gross margin.

Another may generate $1,000 with a 20% margin.

A third may generate $300 initially and become a $10,000 customer over three years.

From the perspective of the business, those customers aren't interchangeable.

So why should the bidding algorithm treat them as if they are?

This is where Value-Based Bidding becomes strategically important.

The better input isn't necessarily more conversion data.

It's better economic data.

Where appropriate, that can mean sending platforms signals based on:

• Gross margin
• Customer quality
• Qualified pipeline
• Closed-won revenue
• Subscription retention
• Customer lifetime value
• Fraud-adjusted value
• Downstream conversion milestones

The goal is simple:

Teach the machine what the business actually values.

---

The CRM should have a seat at the bidding table

For many companies, the biggest measurement mistake happens because the advertising platform gets a signal before the business knows whether that signal was valuable.

A lead comes in.

Google sees a conversion.

The campaign looks successful.

Three months later, sales discovers that half the leads were never qualified.

The advertising algorithm doesn't necessarily know that.

Unless you feed that information back into the system.

This is why server-side conversion architecture and offline conversion data matter.

The closer you can connect advertising signals to actual business outcomes, the better your optimization becomes.

The goal isn't simply:

more conversions.

It's:

more valuable customers.

---

First-party data becomes a competitive advantage

As measurement becomes more fragmented, first-party data becomes increasingly important.

Your CRM.

Your transaction database.

Your subscription data.

Your product analytics.

Your customer lifecycle data.

Your margin data.

Your retention data.

These systems know things that an ad platform doesn't.

The challenge is connecting those signals in a privacy-conscious way and creating a consistent definition of business value.

For larger organizations, data warehouses, clean rooms, and server-side integrations can become part of that architecture.

But again, the technology isn't the strategy.

The strategy is deciding:

What does a valuable customer actually mean to this business?

---

So what should a CMO actually do?

I wouldn't recommend ripping out automated bidding.

I wouldn't recommend going back to manually managing every keyword.

And I definitely wouldn't build an organization around fighting the ad platforms.

The machines are incredibly good at things humans aren't.

Use them.

Just don't let them define the scorecard.

Instead, I'd build the measurement architecture in this order.

1. Establish the business KPI

Start with the number that actually matters.

Revenue.

Contribution margin.

Qualified pipeline.

Gross profit.

LTV.

Whatever the business is genuinely optimizing for.

---

2. Separate platform reporting from business measurement

Keep your Google and Meta dashboards.

Just stop treating them as the only source of truth.

They are optimization systems first.

Your independent measurement layer should answer the broader business question.

---

3. Run incrementality tests

Start somewhere.

You don't need a perfect global experiment on day one.

Pick an important channel, market or campaign where a controlled test is possible.

Build a repeatable experimentation muscle.

---

4. Build modern MMM when the organization has enough scale

For organizations with sufficient spend, geographic coverage, and historical data, MMM can become an important part of the planning process.

Tools such as Meridian and Robyn make the technical barrier lower than it used to be.

---

5. Feed better signals back into the algorithms

Once you understand which customers actually create value, send better conversion signals back into your advertising systems.

Don't optimize blindly for cheap leads.

Optimize for business outcomes.

---

6. Make finance part of the measurement conversation

This might be the most important step.

Marketing measurement shouldn't be something marketing argues about with finance once a quarter.

CMO and CFO should agree on the economic definition of success before the money is spent.

That changes the conversation completely.

---

The new role of the Head of Growth

The growth leader of the next few years isn't simply going to be the person who knows Google Ads inside out.

That skill still matters.

But it isn't enough.

The strongest growth organizations will sit at the intersection of:

Media + Data + Experimentation + Finance.

You need someone who understands how the auction works.

Someone who understands measurement.

Someone who can interrogate an MMM.

Someone who can design an incrementality test.

Someone who understands CRM data.

And, increasingly, someone who can explain marketing investment to a CFO in the language of capital allocation.

That's a very different profile from the traditional performance marketer.

---

The black box isn't the enemy

I've become less interested in the debate over whether AI-powered advertising is "good" or "bad."

That's the wrong question.

Automation is here.

And in many cases, it's going to outperform humans at the mechanics of media buying.

The strategic question is different:

Can your organization tell the difference between platform-reported performance and actual incremental business impact?

If the answer is yes, automation becomes incredibly powerful.

If the answer is no, increasing automation can make the measurement problem harder to see.

That's the risk.

The goal isn't to take control away from the machines.

It's to make sure the machines don't become the only ones deciding whether the machines worked.

---

My CMO rule for 2026

Let the platforms optimize the auction.

Let your data measure the business.

Let experimentation challenge your assumptions.

Let finance define the economics.

And let the CMO connect all four.

That's what attribution sovereignty really means.

Not rejecting Google.

Not rejecting Meta.

Not rejecting AI.

It's simply refusing to confuse reported performance with proven incremental growth.

Because when you're managing millions in acquisition spend, the most important question isn't:

"What did the platform say happened?"

It's:

"What would have happened if we hadn't spent the money?"

That's the question worth building your measurement strategy around.

---

Key Takeaways

• Google AI Max is an automation layer, not an independent measurement system. Google's current AI Max capabilities include expanded search-term matching, text customization and Final URL expansion.
• Platform ROAS and incremental ROAS answer different questions.
• Attribution measures credit; incrementality measures causal business impact.
• Geo-testing and controlled experiments can provide evidence that platform attribution cannot.
• Modern MMM can provide a broader view of channel contribution and budget allocation.
• Google Meridian and Meta Robyn are examples of open-source MMM frameworks available to sophisticated marketing teams.
• Value-Based Bidding becomes more powerful when conversion signals reflect actual customer economics.
• CRM and first-party data should increasingly influence advertising optimization.
• CMOs should separate the bidding engine from the measurement engine.
• The ultimate KPI should connect marketing spend to business value, not simply platform-reported conversions.

FAQ

What is Google AI Max?

Google AI Max is a set of AI-powered features for Google Search campaigns designed to expand search-term matching, customize ad assets and improve landing-page relevance. Google describes AI Max as a way to expand reach and automate aspects of Search campaign management.

Is Google AI Max ROAS the same as incremental ROAS?

No. Platform-reported ROAS and incremental ROAS measure different things. ROAS generally measures revenue attributed to advertising according to the platform's measurement methodology. Incremental ROAS attempts to estimate the additional revenue caused by the advertising investment.

What is incrementality testing in marketing?

Incrementality testing measures what changes when advertising exposure or spending changes. Common approaches include geographic experiments, audience holdouts and controlled spend tests.

What is Marketing Mix Modeling?

Marketing Mix Modeling, or MMM, is a statistical approach that uses aggregated marketing and business data to estimate the contribution of different marketing activities to outcomes such as sales or revenue. Google's Meridian is an open-source MMM framework designed for in-house modeling.

What is Google Meridian?

Google Meridian is an open-source Marketing Mix Modeling framework from Google. It uses aggregated data and Bayesian causal inference to help advertisers estimate marketing impact and make budget allocation decisions.

What is Meta Robyn?

Robyn is an open-source Marketing Mix Modeling framework from Meta Marketing Science. It uses machine-learning and statistical techniques to model media effectiveness, adstock, saturation and budget allocation.

Should CMOs stop using platform-reported ROAS?

Not necessarily. Platform reporting remains useful for campaign optimization and operational decision-making. The strategic issue is whether platform reporting is being treated as the organization's only measurement of marketing impact.

What should a CMO optimize for instead of conversions?

The appropriate metric depends on the business, but sophisticated organizations may optimize toward qualified revenue, gross margin, contribution margin, customer lifetime value or downstream business outcomes rather than raw conversion volume.

Let the machines optimize the media. Let the business decide whether it was worth the money. Let's Talk.