Departure · The problem
Every winner on the account was the same six-second ad
FlixBus's TikTok account was carried by a small set of incumbent price-led cuts. They were cheap — but they were a ceiling. New creative launched constantly and none of it competed: a creative audit of the top and bottom 20 files by CPA showed the losing patterns clearly.
Price openers + watermarks
"€2,99" on glossy templated production accounted for 13 of the 20 worst performers. It read as an ad, not a recommendation.
Faceless formats
Chat-bubble and montage concepts produced zero winners across every iteration. No human, no product in use, no chance.
Launch-and-duplicate sprawl
Up to 86 ads launched in a single week starved each file of learning data, with single files duplicated across 5+ ads competing against themselves.
No transferable learnings
Each market produced creative independently — a win in one country taught the others nothing.
Stop 1 · The strategy
One testimonial formula, engineered from winner DNA
The same audit showed what the top files shared: a single real creator on camera appeared in 10 of 10 top testimonials, while every faceless format sat in the bottom cohort. I codified those shared traits into a repeatable six-beat shotlist — 20 seconds, raw UGC, natural light, one creator carrying all six beats.
Hook
Creator's face, immediate — confessing they overpaid, or stressed at a flight-search screen. Never a price figure.
Turn
"I've officially switched." Quick cut to boarding, QR ticket to the driver.
Demo
Seated: legroom, charger, Wi-Fi — physically shown, narrated first-person.
App proof
Screen recording of the booking flow: destinations, easy plan changes.
Close
Spoken CTA over a suitcase walk-off — "if you've got a trip coming up…"
End frame
Hold on the creator. No price stamp, no logo card. Burned-in captions throughout.
Stop 2 · The test design
A hook × market matrix instead of a creative lottery
Rather than more volume, the test was structured: 36 creatives from five hooks, launched across four markets in dedicated ad groups so every hook earned a clean read in every country. All files launched simultaneously in mid-May from zero — an intentional learning phase with CPA inflation priced in.
It paid off fast. The moment budgets stabilised on the surviving files, blended CPA dropped from €24.76 to €8.48 in 30 days — and held at €8.14 through July. Winners kept winning: the top file sustained a 6.5% CTR, the account record, with a 15.6% hook rate.
Stop 3 · Localisation
One formula, four markets — no local re-invention
Localisation was the multiplier. The same script architecture was produced natively for France, the UK, the US and Italy — local creators, local idiom, identical beat structure. Instead of four creative strategies, the account ran one strategy four times, and every market landed in the same €7.6–8.7 CPA band by July.
FR€6.71
€25.27 → €6.16 → €7.62
Beats the market pool (€7.54) — home of the €4.25 top file
UK€7.91
€27.46 → €11.39 → €7.91
Beats like-for-like pool by 35% (€12.14)
IT€8.60
€21.87 → €9.05 → €7.97
Parity with two long-running incumbent files; beats the fresh test cohort
US€8.87
€25.12 → €9.03 → €8.71
The de-facto video program — holds the 6.5% CTR account record
The identical script converted in French, English and Italian with only native re-performance — proof the formula, not the language, was doing the work.In-market comparison, June–July, web purchase campaigns
Terminus · Outcome
From experiment to the account's creative program
By July the testimonial set was operating within 13% of an incumbent benchmark built over years — on roughly a ninth of the spend, before the matured files had a full run at scale. In two of four markets it beat the existing pool outright; in the US it effectively became the video program.
The agreed next phase: deploy the winning assets into additional markets, relaunch in existing ones, and iterate on the six-beat formula as the account's standing production system — turning a one-off test into a repeatable creative engine.