Theme
Language
Blog · AI & Performance
The data-driven route to a better ROAS
Advertising costs in Berlin keep climbing, and the pressure on performance marketing teams climbs with them. Meta CPMs sit between EUR 18 and EUR 30 in many sectors, and Google Ads CPCs for commercial keywords regularly pass EUR 4. If you are still working with the same manual methods you used three years ago, you are losing margin systematically.
At the same time, the technology has moved on radically. AI-driven performance marketing makes it possible to optimise bids in real time, test creatives automatically, shift budgets dynamically and reach audiences more precisely than before. The intended effect is a lower cost per conversion at a higher return on ad spend.
This article sets out which AI levers are actually available, what the difference looks like in numbers, and why the technology on its own is not enough.
The problem is not a lack of know-how. It is the speed and the complexity of modern advertising ecosystems.
A performance manager can realistically analyse and adjust two or three campaigns in real depth per day. Google and Meta, meanwhile, run thousands of auctions per campaign per hour. Every auction carries different signals: device, time of day, location, user history, competition. Manual bids cannot represent that level of granularity.
A typical e-commerce campaign with 50 ad sets produces millions of data points a month. Which combination of creative, audience, placement and bidding strategy is optimal is no longer something anyone can grasp intuitively. Without machine analysis, the best insights stay invisible. The same applies the moment you start to scale an e-commerce brand across several markets at once.
People are becoming more resistant to advertising. The average lifespan of a creative on Meta has fallen from 14 days (2022) to around 7 days (2026). If you are not testing new variants constantly, performance slips, but manual creative testing does not scale. Our overview of Meta Ads work in Berlin goes into the production side of this in more detail.
Google's smart bidding strategies (Target ROAS, Target CPA, Maximize Conversion Value) use machine learning to calculate a bid for every single auction. The system takes more than 70 signals into account in real time, among them device, browser, location, time of day, remarketing lists and search history.
AI tools generate and refine creatives in a fraction of the time. From automatic copy variants through dynamic image composites to AI-generated video ads, the options grow month by month.
Instead of defining audiences by hand through interests or demographics, AI systems use conversion data to identify the most valuable users on their own. Lookalike audiences, broad targeting with a machine learning backend and first-party data signals are steadily replacing classic interest-based targeting.
Last-click attribution distorts budget decisions systematically. AI-driven attribution, data-driven attribution in Google Ads for example, spreads the conversion value across all touchpoints and shows which channels genuinely contribute to the result.
How do you spread a budget optimally across 5 channels, 20 campaigns and 100 ad sets? Done by hand, it is guesswork. AI tools analyse historical performance data and move budget dynamically to wherever the next euro produces the highest marginal return.
Worked example: EUR 10,000 of ad spend with and without AI optimisation (illustrative).
To make the difference concrete, here is a specific scenario for an online shop in Berlin. Read it as a model calculation rather than a forecast: the numbers show how the levers interact, not what a particular account will do.
The difference does not come from one single factor but from the interplay: a lower CPC, a higher conversion rate and better targeting compound into a far bigger effect than any of them on its own. The middle term of that equation is largely a question of the landing page, which is why conversion rate optimisation belongs in the same project as the bidding setup.
Performance Max (PMax) is Google's AI-driven campaign format, pulling every channel into one. It is not a cure-all, though.
Our recommendation: use PMax alongside manually managed search campaigns, protect brand keywords explicitly in separate campaigns, and check the placement reports regularly. PMax tends to push budget into Display and Discover, where the ROAS is often lower.
Meta's Advantage+ suite covers several AI-driven features.
ASC combines prospecting and retargeting in a single campaign. The algorithm decides which users see which creatives in which placement. For e-commerce accounts with enough conversion data, it is often the most efficient campaign structure on Meta.
Meta adapts creatives automatically to different placements and user types: brightness, contrast, aspect ratio and even text overlays are optimised dynamically. The feature works best with strong source material, because AI does not make a weak creative good.
Rather than detailed targeting instructions, the advertiser supplies only audience suggestions. Meta treats these as a starting point and widens the audience automatically when the model finds promising segments outside the suggestion.
A practical note for Berlin: for local campaigns, so events, hospitality and local services, combine Advantage+ with geo-targeting on Berlin plus a 30 km radius. The model optimises inside that radius, while wasted reach beyond the region is avoided.
For all the possibilities, plenty of AI-driven campaigns still fail. The reasons are almost always human rather than technical.
AI algorithms optimise towards whatever signal you hand them. If tracking is faulty, if micro-conversions are counted as macro-conversions, or if CRM data is out of date, the model learns the wrong thing. Data quality is not a nice-to-have. It is the foundation of every AI optimisation.
AI optimises inside the parameters people set. Set Target ROAS too aggressively and you get too little volume. Hand creative testing to the machine without supplying strategic hypotheses and you are testing at random instead of systematically. The best work comes out of human strategy and machine optimisation operating together.
AI tools often optimise towards isolated metrics: the ROAS of a single campaign, the CPA of a single channel. Real business success, though, comes from the way all channels work together across the whole customer lifecycle. Look only at platform ROAS and you miss the view of blended ROAS, customer lifetime value and marginal cost.
The trend is towards fewer, larger campaigns, and in principle that is right, because the model gets more data per campaign. But too much consolidation means losing control: when a single PMax campaign spreads budget across Search, Display, YouTube and Discover, it is often unclear where the performance is actually coming from.
AI-driven performance marketing is not an optional upgrade. It is becoming a precondition for competitive advertising in Berlin. The technology is available, the tools are mature and the effects are measurable.
But AI does not replace a strategy. It accelerates one. The companies that combine clean data, clear goals and human steering with machine optimisation are the ones building the strongest competitive position over the next few years.
Start now and you train your algorithms earlier, gather more data and benefit from the compounding effect of machine learning. Wait, and you end up competing against rivals with better models. What that costs in practice, and what a sensible setup looks like at different budget levels, is covered in our breakdown of what a performance marketing agency costs. If you would rather talk it through directly, our contact page is the quickest route.
Wondering which AI levers are realistic for your own ad account? A free, no-obligation 30-minute call goes through the specifics.
Book a callRead next