How Creative AI analytics helped Wolt unlock a potential 40%+ CPA reduction on TikTok

When creative is your biggest performance lever, gut feel is not a strategy. Wolt, TikTok, and Nepa used Creative AI analytics to decode what actually drives acquisition cost — across thousands of ads, 25+ markets, and two campaign types.

Talk to our experts

THE OVERVIEW

WOLT

CLIENT
WOLT

Year
2014

Industry
Food delivery / Q-commerce

SOLUTIONS
Creative AI Analytics, Campaign Evaluation

Wolt runs thousands of always-on TikTok campaigns each year across user growth, brand building, and courier recruitment. Creative was produced by many teams and partners, with different budgets, formats, and local priorities — resulting in no unified, evidence-based view of what makes a Wolt TikTok ad perform. Working with Nepa and TikTok, Wolt used Creative AI analytics to classify and score thousands of ad assets.

Sessions: IIEX 2026 Creative Performance with AI

See the case presented live

Watch the full IIEX Europe 2026 session, where Nepa walks through how AI decoded thousands of Wolt TikTok ads and turned the findings into a creative playbook.

Watch the session

THE CHALLENGE

Creative at scale — without a consistent standard

The scale problem

Wolt is one of TikTok’s most advanced advertisers in Europe, producing hundreds of creative assets every quarter across a wide network of teams and agency partners. This volume made it impossible to manually evaluate what was working — and why.

No unified playbook

Without a shared, evidence-based standard, creative decisions relied on individual judgement, platform folklore, and what looked right. Spend continued to flow toward ads that may have been significantly underperforming against their potential.

The core question

Why are some ads performing better than others — and what can Wolt do to consistently raise creative performance across 25+ markets?

Creative as the missing lever

Wolt applied rigorous measurement to media planning, audience targeting, and budget allocation. But creative quality — the single largest driver of performance variation — remained largely unmeasured at scale.

The Approach

Three steps from creative chaos to creative clarity

Step 1 — AI classification of creative features

Nepa’s Creative AI classified the features present in all Wolt TikTok ads over the preceding year — thousands of assets. This covered visual elements including brand logo presence and frequency, human face visibility, food and product in-frame, hooks, text overlays, narrative structures, and TikTok-native mechanics. Every ad received a comprehensive feature profile.

Step 2 — Statistical drivers analysis

Nepa then established correlations between creative features and campaign performance via statistical drivers analysis — identifying which features have the strongest positive or negative impact on CPA. This moved the analysis from ‘what is in the ads’ to ‘what actually moves the needle’. Several intuitively appealing features turned out to hurt performance rather than help it.

Step 3 — Creative scoring and playbook

A composite Creative Score was established for each ad, accounting for performance across the dimensions proven to drive effect. The scores translated directly into a custom TikTok Creative Playbook — a practical, data-backed toolbox covering opening hooks, branding approach, people and context balance, format, length, and CTA guidance. Every recommendation was traceable back to the data.

KEY FINDINGS

Three findings that challenged the obvious

Brand logo: less is more

Low logo frequency was best for performance — but 29% of Wolt’s ads had very high logo frequency. Heavy logo presence signals ‘this is an ad’ and triggers scroll behaviour. The best-performing ads integrated branding through environment, app UI, colour palette, and voiceover.

Faces outperform products

High frequency of visible human faces was by far the strongest performance driver — yet 41% of Wolt’s ads had low or no face visibility. TikTok rewards content that feels made by a person, not a brand. UGC-style formats consistently outperformed polished, brand-centric production.

Food: context beats convention

Low food frequency in frame was best for performance — even for a food delivery brand. Generic food shots had become visual wallpaper. The ads that broke through focused on the situation around the food: the late-night craving, the convenience moment, the group of friends.

THE RESULT

The business potential of creative excellence

Creative quality was equally important for both brand and acquisition campaigns — but the success patterns differed significantly. Brands need separate creative playbooks for brand and acquisition, not a single generic ‘TikTok style guide.’ Both campaign types showed a potential CPA reduction of roughly 37–38% when best-practice guidelines were followed.