AI is about to control a third of ad spend. What if it’s wrong?
Opinion
The conversation around AI has become unbalanced. We’re spending too much time discussing how intelligent the technology is becoming, but far less time asking whether we’re teaching it the right lessons, INCRMNTAL’s CEO writes.
The IAB’s recent forecast that AI-driven advertising will account for 32% of digital ad spend by 2030 should excite every marketer. AI is already transforming media buying, campaign optimisation and audience targeting, with platforms increasingly capable of making thousands of decisions in the time it takes a human to make one.
The assumption is that better technology naturally leads to better outcomes. The reality is more complicated.
AI doesn’t create strategy. It learns from the signals we give it. If those signals accurately reflect what drives business growth, AI can become one of the most powerful tools our industry has ever developed. But if they’re flawed, AI won’t recognise the mistake. It will simply automate it at scale.
The biggest challenge facing advertisers over the next five years won’t be adopting AI; it will be ensuring it learns from the right data.
AI is optimising against yesterday’s measurement model
Most optimisation systems today still rely heavily on attribution. This made sense 20 years ago, when digital advertising was largely built around clicks and relatively straightforward customer journeys. Last-touch attribution was easy to implement, happened in real time, and gave ad platforms a practical signal to optimise towards.
Today’s reality is very different.
A customer might discover a brand on social media, ask ChatGPT for product recommendations, watch a CTV advert later that evening, search for reviews a few days later and finally convert after clicking a paid search ad. Every one of those interactions may have influenced the purchase, yet many measurement systems still reward whichever platform happened to appear closest to the conversion.
The problem isn’t attribution itself. It’s assuming attribution measures causality. It doesn’t – it measures who received the credit.
Correlation isn’t the same as contribution
Recently, we analysed more than $1bn in advertiser spend across 65 enterprise advertisers to compare attribution with incrementality. The findings were difficult to ignore.
We found consistent structural differences between the channels receiving attribution credit and those generating genuine incremental impact. Search often appeared over-attributed because it naturally sits close to the point of conversion, while channels that create demand earlier in the customer journey often received far less recognition than they deserved.
This isn’t about declaring one channel “good” and another “bad”. Every channel has a role. The more important point is that attribution systematically favours observability over influence. It tells us which interaction was easiest to measure, not necessarily which activity persuaded someone to buy.
For years, marketers compensated for those limitations through experience and judgement. AI doesn’t have that luxury.
AI will faithfully optimise the wrong objective
Machine learning doesn’t question whether the optimisation target is correct. It assumes it is. So if an attribution model consistently over-credits one channel, AI will rationally direct more budget toward it. If another channel genuinely creates demand but rarely receives attribution credit, AI learns that the channel matters less than it actually does.
We’re already seeing examples of organisations handing greater responsibility to AI-powered campaign management, only to later discover that strong performance metrics concealed significant inefficiencies and wasted spend. The algorithms weren’t malfunctioning. They simply optimised against the wrong objective.
That’s why I believe the conversation around AI has become unbalanced. We’re spending enormous amounts of time discussing how intelligent the technology is becoming, but far less time asking whether we’re teaching it the right lessons.
Better AI starts with better data
This isn’t an argument against AI. Quite the opposite.
I believe AI fundamentally improves adtech. It can analyse complexity far beyond human capability, spot opportunities in real time, and continuously rebalance budgets as market conditions change.
But its success depends entirely on the quality of the data beneath it, especially measurement results.
The next competitive advantage won’t come from attaching another AI assistant to a dashboard. It will come from giving AI access to signals that reflect genuine business impact, not just observable behaviour.
If AI is expected to influence nearly a third of digital advertising spend by the end of the decade, measurement can no longer be treated as a reporting exercise. It becomes the training data for every future optimisation decision.
The industry has spent years asking whether AI is ready for advertising. A better question might be whether advertising is ready for AI.
Maor Sadra is the CEO and co-founder of INCRMNTAL