AI Can Optimize the Ad Campaign. But Can It Tell You What the Ad Actually Caused?

Photo By: Luke Chesser

Advertising is becoming easier to execute and harder to interrogate.

At its unBoxed conference, Amazon Ads introduced new tools that give artificial intelligence a larger role in campaign management. Among them is Full-Funnel Campaigns, which brings sponsored ads, display, video, and streaming TV into a single campaign while using AI to optimize execution toward an advertiser’s objectives.

For marketers, the appeal is obvious. Instead of manually coordinating campaigns across formats, teams can increasingly define products, creative, budgets, and desired outcomes while automated systems handle more of the execution.

But as AI makes more advertising decisions, another question becomes increasingly important: How do brands know which of those decisions actually changed consumer behavior?

Optimization and Incrementality Aren’t the Same Thing

Imagine an advertising system identifies consumers who are highly likely to buy a moisturizer. It serves those consumers ads, they purchase the product, and the resulting conversions are attributed to the campaign.

The system may have performed exactly as intended. But some of those consumers may have bought the moisturizer without ever seeing the ad.

That’s the distinction between identifying a customer who is likely to convert and determining whether advertising caused that customer to convert.

For marketers, the difference isn’t academic. It can determine where the next advertising dollar should go.

As AI becomes better at predicting consumer behavior, an automated system could become exceptionally good at finding people who were already close to purchasing. That can improve campaign metrics without necessarily establishing how much additional demand the advertising created.

Automation Could Make Attribution More Complicated

AI isn’t simply changing targeting. It is increasingly capable of adjusting multiple elements of a campaign simultaneously.

Audiences can change. Bids can move. Creative can be generated and tested. Budgets can shift between placements and formats.

Meanwhile, advertising isn’t operating in isolation. The same consumer might see a creator discussing a product on TikTok, receive a retailer promotion, encounter a display ad, search for reviews, and ultimately make a purchase on Amazon. The brand itself may have changed its price or launched another promotion during the same period.

Sales increase. Which decision deserves the credit?

The irony is that AI can simplify the operational side of advertising while making the environment marketers are trying to measure increasingly complex.

ROAS Doesn’t Answer Every Question

Return on ad spend remains a useful metric. It helps marketers understand how much attributed revenue is associated with advertising expenditure.

But attributed revenue and incremental revenue aren’t necessarily the same thing.

Suppose a campaign is credited with $5 million in sales.

Executives still have reasons to ask how many of those purchases would have occurred without the campaign. Some consumers may have purchased organically. Others may have bought earlier than they otherwise would have. Advertising could have shifted purchases between channels rather than created entirely new demand.

There is another layer, too. Even incremental revenue doesn’t automatically mean incremental profit.

A campaign that generates additional sales at a high acquisition cost can produce a very different business outcome from one that attracts profitable new customers. For companies deciding how much capital to allocate to advertising, revenue, gross profit, and contribution margin can tell different stories.

The question therefore becomes more complicated than whether an AI system improved campaign performance. Did the decisions it made create enough additional economic value to justify the spend?

AI May Need a Quantitative Counterpart

This distinction is increasingly relevant as companies build AI into commercial decision-making.

Kapnova, an agentic revenue and profit optimization system for consumer brands, separates opportunity discovery from quantitative evaluation. AI agents can identify potential opportunities and patterns, while using different analytical methods. Including causal inference, econometrics, forecasting, optimization, and scenario simulation,can be applied depending on the business question.

In advertising, that distinction could help move the conversation beyond identifying which campaigns, audiences, or channels correlate with purchases.

Brands could instead ask what sales might have looked like without a particular campaign, which consumers actually changed their behavior, or what could happen if spending were redistributed across channels.

The objective isn’t to replace campaign optimization. It is to evaluate the economic consequences of the decisions being optimized.

That distinction becomes more valuable when AI is making those decisions at a scale humans could never manage manually.

The CMO and CFO May Ask Different Questions

Marketing teams and finance teams can look at the same AI investment through different lenses.

A CMO might reasonably want to know whether AI improved campaign performance, reduced manual work, identified better audiences, or increased conversions.

A CFO may want to know how much incremental profit those improvements created.

Both questions matter.

As automated systems receive more influence over media budgets, however, companies will increasingly need to connect them.

An AI system shouldn’t necessarily be considered successful simply because it found more customers who converted. From the perspective of the overall business, the more consequential question is whether its decisions changed consumer behavior profitably.

That requires knowing not only what happened after the algorithm acted, but what likely would have happened if it hadn’t.

Sometimes the Best Ad Decision May Be Not to Spend

AI is rapidly lowering the effort required to make another marketing decision.

It can generate more creative, evaluate more audiences, adjust more bids, test more variations, and redistribute budgets faster than human teams could alone.

Marketing budgets remain finite.

That means the next stage of advertising intelligence may depend as much on deciding where not to invest as identifying another opportunity to spend.

The question has traditionally been: Where should the next advertising dollar go?

As AI takes on more of that decision, companies may need to pair it with another: What would happen if we didn’t spend that dollar at all?