diffusr
← All posts

How we got Shopee’s cost per winner from $470 to $97

1,500 creatives in 30 days, two formats nobody on the account had tried, and the way we find formats like them.

Justin To
Justin To Co-founder, diffusr · Writes about what converts, and how we find new formats.28 October 2026 · 1 min read
Follow on LinkedIn

When we started with Shopee, their average cost per winning ad was $470. Thirty days later it was $97, with 1,500 creatives launched. Nearly 5x cheaper per winner.

It came down to two formats and a lot of volume.

The two formats

  • UGC + green screen. A creator in front of a scene from the drama, talking straight to camera like a friend recommending a show: “When the boss becomes the boyfriend.”
  • Reaction video. A creator reacting to the moment that makes you need to watch the next episode.

The hooks were plot twists and feelings, not product features. “3 dramas hiding a secret.” “What would you do for family?” Many were in Taglish, the way people actually talk about shows with their friends.

Why volume mattered

With one format you test one idea. With a format and a long list of shows, hooks and creators, you test hundreds of small variations of an idea that already works. Most lose. The few that win pay for everything, and they win cheaper because the structure was proven before we made a single one.

How we find formats like these

We don’t brainstorm them. We look for them.

  • Start in the ad library. We read ads that are live on Meta right now, grouped by format.
  • Sort by days running. An ad that has run for months has spend behind it. Nobody keeps paying for a loser.
  • Look one category over. The reaction format wasn’t new. It just wasn’t being used to sell shows on a shopping app.
  • Test small, then scale the winner. A few variations first. Whatever beats the account’s current cost per winner gets the volume.

You can run the first two steps yourself in our Ad Library. Type your category and see what other brands have kept running.

New formats come from other categories, not from a whiteboard.