A/B pre-lander tests and step-by-step audience optimisation reduced CPL by 38% in 6 weeks.
The task
The campaign launched from scratch: the advertiser had no buying history in this geo. Needed to reduce cost per lead in a French lead-gen campaign without sacrificing traffic quality.
We worked the "Lead Gen" vertical in the France market. Payment model — CPL per lead, main sources — .
How we did it
The plan was built so each step could be rolled back on its own.
- Launched three pre-lander variants in parallel from day one rather than sequentially: a direct-benefit version, a social-proof-led version, and a no-pre-lander version going straight to the form — so traffic quality differences couldn't be blamed on launch timing.
- Split audiences by "temperature": the social-proof variant ran on cold traffic, the no-pre-lander version on retargeting that had already seen the offer.
- Tightened exclusions for users who left the pre-lander in under 3 seconds — this signal correlated with poor lead quality more strongly than any demographic filter.
What went wrong
Not everything went according to plan. The no-pre-lander variant unexpectedly converted better on retargeting audiences than on cold traffic, where social proof won outright — no single universal winner emerged.
Instead of picking one variant for the whole campaign, both pre-landers were kept running in parallel — each on its own audience-temperature segment — effectively turning the campaign into two separate funnels.
The result
By the end of the period, the setup hit its planned numbers. Over the six weeks: roi — 248%, cost per lead — −38%, pre-landers tested — 3.
CPL dropped 38% over six weeks, but most of the gain came in weeks 4-6, after the fast-bounce exclusions from the pre-lander were tightened.
What we took away
- Launching pre-lander variants in parallel rather than sequentially from day one guards against wrongly blaming traffic-quality differences on launch timing.
- The same offer can need different pre-landers for cold and retargeting audiences — a single winner doesn't always have to be picked.
- Pre-lander bounce speed is a strong lead-quality signal, sharper than demographic filters.
These numbers belong to one specific setup, geo, and buying period. Economics will differ on another offer or in another season — we always recalculate them before launch.