The problem: creative is the biggest lever, measured the least
You apply rigorous measurement to media planning, targeting and budget allocation — every lever except the one that decides whether the rest pays off. Creative quality drives more variation in outcomes than media alone, yet it remains one of the least systematically measured dimensions of marketing performance.
The reason is scale, not indifference. Focus groups, manual reviews and one-asset pre-tests were never designed for an environment where brands produce thousands of ad variations a quarter. On platforms like TikTok — fast feed, quick creative fatigue, a deliberately blurred line between content and advertising — that gap widens fast. The result is a growing volume of content that gets produced, published and scrolled past without measurable impact.
What changed is the technology. AI can now classify creative features across thousands of assets in days, correlate them against real performance data, and surface patterns manual analysis would miss. That is what this session demonstrates, using a real case.
What you’ll take away
In 20 focused minutes, you’ll see:
How AI decodes creative at scale — classifying features like logo frequency, human faces, food visibility and hook structure across thousands of ads, then linking them statistically to CPA.
Which creative features actually move acquisition cost — including several findings that run against category instinct.
How a Creative Score separates winners from losers — and why the gap between top and bottom creatives is far wider than most teams assume.
How insight becomes a usable playbook — translating data into clear, traceable rules for everyone who creates or commissions content.
Where Creative AI fits — alongside Brand Tracking, Campaign Evaluation and Marketing Mix Modelling as part of one continuous decision system, not a standalone test.
Wolt is one of TikTok’s most advanced advertisers in Europe, running thousands of always-on campaigns each year across user growth, brand building and courier recruitment in more than 25 markets. Creative was produced by many teams and partners, with different budgets, formats and local priorities — and no unified, evidence-based view of what “good” TikTok content actually looks like for Wolt.
The question was deceptively simple: why do some ads perform better than others, and what can be done to consistently raise creative performance? TikTok brought platform-level performance data, Nepa brought Creative AI analytics, and Wolt brought scale and a willingness to put creative under the microscope.
Key findings from the session
Concrete, counterintuitive takeaways you can apply immediately:
Brand logo: less is more. Low logo frequency performed best — yet 29% of Wolt’s ads still led heavily with the logo, signalling “this is an ad” and triggering scroll behaviour. The strongest ads integrate branding through environment, app UI and colour rather than a logo card.
Faces beat products. High frequency of visible human faces was by far the strongest performance driver — yet 41% of ads had low or no face visibility. TikTok rewards content that feels made by a person, not a brand.
Context beats category convention. Low food-in-frame frequency performed best, even for a food delivery brand. Generic food shots have become visual wallpaper in the category; the ads that broke through focused on the moment around the food, not the food itself.
A clear link to performance. More than 60% CPA difference separated the top 10% and bottom 10% of creatives, scored against the dimensions proven to drive effect.
A large, quantified gap. 86% of Wolt’s videos were below “good” creative levels at the time of analysis — a measure of headroom, not a verdict on quality, even for a sophisticated advertiser.
Brand and acquisition need different recipes. Creative quality mattered equally for both, but the winning features differed — so a single generic style guide isn’t enough.
From insight to action
Analysis only creates value when it changes behaviour. The project produced a custom TikTok Creative Playbook: a practical toolbox translating Creative AI findings into data-backed rules on opening seconds and hooks, branding approach, the balance of people, context and food, and format, length and CTA. Every recommendation is traceable back to the data — not opinion or platform folklore.
From there, Wolt and Nepa run a continuous Assess — Adopt — Measure — Refine loop: produce best-practice creatives, score every new asset, track CPA against baseline, and update the recommendations each quarter.
Where Creative AI fits
Nepa positions Creative AI analytics as a complement to — not a replacement for — concept testing, pre-testing and Campaign Evaluation. Traditional methods are ideal for deep evaluation of individual concepts and capture subjective dimensions AI can’t yet measure directly. Creative AI excels where they struggle: scale. Together with Brand Tracking, Campaign Evaluation and Marketing Mix Modelling, it makes creative measurement part of the steering wheel, not the rearview mirror.
Who should watch
CMOs and marketing leaders accountable for large creative and media investment who can’t afford to keep guessing which creatives drive efficiency.
Performance and growth leads managing high-volume creative production across markets and platforms.
Insight and effectiveness leaders looking to bring the same rigour to creative that they already apply to media.
Teams managing agencies and creative partners who want evidence-backed guidance on what to lean into — and what to avoid.
