Where AI Actually Helps in Paid Media in 2026, and What Is Still Missing

Monday morning, on an account that runs well. Smart Bidding adjusted the bids over the weekend. Advantage+ found new audiences. Performance Max moved part of the budget from YouTube to Search. New creative variants were generated. Every platform reports that things are going well.

And yet the question that decides the month’s result has an answer in none of these interfaces: does the next euro go to Google or to Meta?

That is where AI stands in paid media today. Optimization inside each platform has become the norm. Optimization between platforms is still largely left to do.

Optimization inside each platform has become the norm

It should be said without reservation: platform automation is good, and it is everywhere.

At Google, Smart Bidding bids for conversions or conversion value in every auction, using signals such as device, location, time of day or the actual query. Performance Max reaches all of Google’s inventory, from Search to YouTube, Discover, Gmail and Maps, and steers the budget toward the placements that perform. AI Max for Search, out of beta this year, has already been adopted by half a million advertisers, according to Alphabet.

At Meta, end-to-end automation is turned on from the start for sales, app and lead campaigns. The annual revenue run-rate flowing through these Advantage+ solutions is over 75 billion dollars, according to Meta, up from 60 billion nine months earlier. Budget, audience, placements, creative variants: all of it can be handed to the system.

TikTok follows the same path with Smart+, which automates audience, budget, placements and catalog creative.

Each platform measures these results on its own product, and they should be read that way. But the direction is clear: writing, iterating, launching, bidding, targeting and allocating budget inside a single platform is done. It has become a commodity.

Every platform counts conversions its own way

The trouble starts when the results are put side by side. Each platform grades its own work, with its own rules.

Google Ads uses data-driven attribution by default, but it is computed only on interactions with Google ads, over a 30-day window after the click. Meta counts by default 7 days after a click or 1 day after a simple view, on a last-touch model, and reports the conversion on the date of the impression. TikTok also counts 7 days after the click and 1 day after the view, and adds attribution after 6 seconds of video watched.

None of these platforms deduplicates its conversions against the others. The consequence follows logically, even if none of them puts it that way: the same sale can be claimed two or three times, and the sum of reported conversions can exceed actual sales.

Meta says as much about its own account: "comparing across different attribution models will lead to inaccurate conclusions". Putting a Meta ROAS next to a Google ROAS to decide where the budget goes is exactly that case.

Attributed does not mean caused

There is a second, deeper gap: an attributed conversion is not a conversion caused by the ad.

The platforms make the distinction themselves. Meta presented to its investors an "incremental" attribution that, according to Meta, identifies 24% more incremental conversions than its standard model. In other words, the standard counter and the real effect are two different things.

Research goes further. A study published in 2023 in Marketing Science, based on 663 large-scale Facebook experiments with a Meta researcher among its authors, compares the real effect measured by experiment with the best statistical methods that do not use experiments. In the lower funnel, the median real effect is 5%. The most sophisticated non-experimental method estimates it at 24%. On paid search, experiments run at eBay had already concluded that real returns were "a fraction" of conventional estimates.

None of this disqualifies the platforms. But it does mean that a budget arbitrage based on reported numbers starts from a fragile base.

The platforms themselves send the next-euro question elsewhere

The most telling part is where the platforms file this question.

In its help center, Meta lists "Where should my next marketing dollar go?" among the questions answered by its marketing mix model, Robyn. Not by Advantage+. Google draws the same line: Analytics lets you manage spend "across all accounts and marketing channels", while Google Ads manages it inside its own accounts.

The tools exist, and they deserve to be named. Google publishes Meridian, an open source marketing mix model, and Meta publishes Robyn. Google Analytics offers, in beta, cross-channel budget planning that includes non-Google costs. Google and Meta both offer lift tests from a 5,000 dollar budget.

What none of these tools does is close the loop. They plan, they project, they measure after the fact. Meta itself notes that its model needs analysts to tune the parameters and interpret the results. In practice, the arbitrage between platforms happens once a quarter, in a spreadsheet, and nobody executes it on Monday morning. And no native lift test measures a budget transfer from one platform to another: each one measures the effect of its own ads.

The last euro is not worth the average euro

There is a structural reason why this optimization cannot wait for the end of the quarter: returns diminish.

Google’s researchers put it plainly: advertising has lag effects and diminishing returns. Meridian’s response curves show it channel by channel: each extra euro returns less than the previous one. A study of 288 consumer brands, on television, found negative marginal returns for more than 80% of them. The context is not digital, but the lesson holds everywhere: an average ROAS says nothing about the last euro spent.

And that last euro is precisely the one that moves between two platforms. A platform with an excellent average return can be saturated at the margin, while another still has room. Neither will tell you, since each one only sees its own scope.

What it takes to optimize between platforms

Closing this gap does not require switching off platform automation. It requires a layer on top, with four elements:

1. 1A source of truth outside the platforms: your sales, your analytics, your e-commerce data, against which every platform is judged the same way.
2. 2Shared definitions: the same window, the same attribution rule, so that two ROAS figures can finally be compared.
3. 3Reasoning at the last euro, not the average: where one more euro still pays, and where it no longer does.
4. 4A continuous optimization loop: propose a move between platforms, get it approved, apply it, then measure what it produced against the neutral source.

How we approach it at Laier

That is the layer we built. Laier leaves each platform what it does well inside its own walls. It still challenges some of its optimizations, and steers it toward what it does best. But above all, it takes care of the optimization between them.

Laier continuously reads your Google, Meta, TikTok and all your other ad accounts, brings their numbers back to shared definitions and checks them against your source of truth. Whatever your goals, brand building, performance measured in return on investment or pure performance, optimization is judged on the same yardstick. When a euro would be better placed elsewhere, Laier proposes the move, with the number that justifies it. You approve. Laier applies it on the connected platforms and measures what it produced.

Not once a quarter. Every week, on every account.