Why $32B in AI Ad Spend Requires Independent Verification

US AI ad spending is projected to reach $32 billion in 2026 and $68 billion by 2030, large enough to change budget meetings before anyone has settled the measurement question.[1] That forecast stops being a market story and becomes an account-level problem when a platform-reported ROAS looks stable: should the next dollar move into Advantage+, Performance Max, or another automated campaign type?

The strongest independent evidence available does not support a simple yes. In Haus’s 640-test Meta incrementality dataset, Advantage+ outperformed manual campaigns on incremental ROAS in 42% of tests, which also means it underperformed manual campaigns in 58% of tests.[2] That result is not an argument to turn automation off. It is an argument to stop treating platform ROAS as sufficient permission to scale.

There is a separate debate over whether the $32 billion forecast is partly a relabeling of paid search, platform automation, and campaign types advertisers were already using. That matters, and it is worth reading alongside The $32B AI Ad Spend Bubble Is Mostly a Relabel. But even if a meaningful share is renamed spend rather than net-new budget, the operating question remains the same: are buyers scaling these systems because they create incremental customers, or because the platform interface makes the case for them?

The Independent Record Is Mixed, Not Ambiguous

Haus’s Meta study is the load-bearing evidence because it asks the question most dashboards avoid: what happened incrementally, compared with a manual alternative? Across 640 incrementality experiments, Advantage+ delivered 12% lower iROAS than manual campaigns on average, while running at 18% lower daily spend.[2]

That pairing matters. Lower daily spend can be a real benefit for lean teams or constrained budgets. It can also make an automated campaign look operationally cleaner while producing less incremental return. A CMO sees the spend restraint; a media buyer still has to explain whether the campaign is finding new demand or harvesting demand that another campaign would have captured.

The same study found that Advantage+ over-reported incrementality by 12 percentage points versus manual on average.[2] That is the number that should make a budget owner slow down. A campaign can look properly credited inside the platform and still fail the harder test: what would have happened if the automated treatment had not run?

The post-treatment window is another place where the difference becomes practical rather than academic. Haus reported post-treatment-window lift of 17% for Advantage+ versus 32% for manual campaigns.[2] In plain terms, manual campaigns in that dataset showed stronger delayed impact after the treatment window ended. If a brand has longer consideration cycles, repeat exposure paths, or delayed purchase behavior, cutting evaluation at the platform reporting window can miss part of the business result.

There is an important boundary around the finding: the Haus dataset represents brands spending $14 million or more per year on Meta, so smaller advertisers should not assume the same distribution will hold in their own accounts.[2] The right takeaway is narrower and more useful: among large Meta advertisers, Advantage+ was not reliably superior to manual campaigns on incremental ROAS, and the only safe answer is to test the account in front of you.

The Dashboard Can Stay Calm While CAC Moves

The measurement gap becomes easier to feel in the CAC example surfaced by Pixis, citing Wicked Reports data from 55,000 Meta campaigns. In that dataset, new-customer CAC on Advantage+ doubled from $257 in May 2024 to $528 in May 2025, while Meta’s reported ROAS held steady at about $4.52.[3]

This is secondary-source evidence, so it should not be treated like an audited universal benchmark. But the pattern is exactly the one buyers worry about: the platform metric does not panic, while the business metric deteriorates. A dashboard that holds ROAS flat can still be absorbing more spend to acquire the same kind of customer, or claiming credit for customers who were already close to purchase.

That distinction changes the scaling decision. If reported ROAS is flat and CAC is rising, adding budget is not a neutral optimization choice. It can raise blended acquisition cost, crowd out manual prospecting, and make the next incrementality test harder because the account has already reorganized around the automated campaign.

This is also why a holdout test feels so inconvenient in the room. The platform rep can point to forecasted conversions. The dashboard can show efficient return. The paid social lead is the one asking for a pause long enough to prove whether those conversions are actually new. That request sounds conservative only if the organization has decided that attribution is the same thing as incrementality.

Google Has the Same Verification Problem

The issue is not limited to Meta. Performance Max can be effective, but it also blends inventory and decisioning in ways that make cannibalization harder to see. Grow My Ads reported three unnamed cases in which brands cut PMax spend by 77% to 80%, shifted budget to Standard Shopping, and saw revenue increases of 35%, 45%, and 200%.[4]

Those are single-agency case studies, not statistical proof. They do not establish that most brands should cut PMax by a similar amount. They do show why an account that appears healthy at the campaign level can still deserve a cannibalization audit. If PMax is claiming conversions that brand search, Shopping, or organic demand would have captured, then the platform’s reported return can be directionally right for attribution and directionally wrong for budget allocation.

The practical test is not whether PMax has conversions. Of course it does. The test is whether total account revenue, new-customer volume, and marginal profit improve when PMax receives more budget, after controlling for what brand search and Shopping were already doing. A campaign that wins inside its own reporting boundary can still weaken the portfolio.

Why the Pressure to Automate Keeps Rising

Advertisers are not moving toward AI systems only because of vendor persuasion. The IAB reported that 64% of ad executives cited cost efficiency as AI’s top benefit, up from fifth place in 2024, and that 83% use AI in creative.[5] That combination explains the pressure: teams want cheaper production, faster iteration, and fewer manual account chores.

Platforms have their own pressure. CNBC reported that Meta raised its 2026 capital expenditure outlook to $125 billion to $145 billion and that Google raised its outlook to $180 billion to $205 billion; CNBC also reported that Meta shares fell 7% while Alphabet shares rose 7% as investors judged the companies’ AI spending paths differently.[6] Reuters separately reported that Meta shares fell on concerns over AI spending and legal scrutiny.[7]

Those market reactions do not prove that any given ad product is overcharging or underperforming. They do explain why automated monetization matters so much to the platforms selling it. When infrastructure spending rises, ad systems that can absorb more budget with fewer manual controls become more strategically important. Buyers should assume the default path will continue to tilt toward automation, not because it is always better, but because it is easier to scale.

What Has to Be Verified Before Scaling

The standard for trusting AI ad automation should be higher than “platform ROAS did not fall.” The evidence points to a tighter operating threshold: keep a manual baseline alive, run incrementality tests against it, and let iROAS and CAC trend decide whether the automated campaign deserves more budget.

The manual baseline is the easiest piece to sacrifice and the hardest one to rebuild. Once an account has been reorganized around automated campaigns, historical comparison becomes noisy. Creative mix changes, audience exposure changes, budget pacing changes, and the team starts debating whether the old structure was really comparable. Keeping a manual control may look inefficient in the short term, but it preserves the ability to answer the only question that matters when budget is on the line.

iROAS should carry more weight than reported ROAS because it measures the incremental return caused by the campaign rather than the return credited to it. CAC trend should sit next to it because finance feels CAC before it feels attribution nuance. Post-treatment-window lift matters because a campaign that creates delayed demand can be undervalued, while one that mostly captures ready buyers can be overvalued. Cannibalization checks matter because automated systems can win credit inside one product while reducing clarity across the account.

A clean test does not need to become a permanent research project. It does need enough separation to survive the budget conversation: defined treatment and control, a stated incrementality metric before launch, agreement on how delayed conversions will be read, and a rule for what happens if platform ROAS and business CAC disagree. Without that rule, the platform number usually wins by default because it arrives first and looks more precise.

A Reasonable Threshold for Trust

AI ad automation may work. The Haus data says Advantage+ beat manual campaigns in a meaningful minority of tests, and lower daily spend can be valuable for some teams.[2] PMax can also create operational leverage when it is not merely absorbing branded or Shopping demand. The problem is not automation; the problem is scaling automation faster than the evidence can support.

The $32 billion forecast should be treated as pressure, not proof. Before adding budget, the campaign should clear a threshold the platform dashboard does not control: positive incremental ROAS versus a manual baseline, no unexplained rise in new-customer CAC, observable lift beyond the immediate reporting window where relevant, and no material cannibalization of channels that were already converting the same demand.

If an automated campaign passes that standard, scale it. If it cannot, the forecast is not a reason to comply.

References

1. AI Ad Spending Will Reach $32 Billion In 2026, And Paid Search Teams Are Already Running It, Forbes, July 14, 2026
2. The Meta Report: Lessons from 640 Haus Incrementality Experiments, Haus.io, 2026
3. Advantage+ vs. Performance Max Head-to-Head (2026), Pixis.ai
4. We Cut Performance Max Spend by 80% (And Still Hit Our Goals), Grow My Ads, 2026
5. The AI Ad Gap Widens, IAB
6. Investors trust Google more than Meta when comes to spending on AI, CNBC, Apr. 29, 2026
7. Meta shares fall on concerns over AI spending, legal scrutiny, Reuters, Apr. 29, 2026