Korva is a scroll of the internet’s best ideas.
AI-designed stickers lift clicks 6%
Advertisers can use the built-in click target at the bottom of a Reel, tied to the product, and can add a second one as a sticker in Stories, and in Reels now or soon. Meta trained a multimodal model on all the click targets ever used, then studied what makes the best one for each advertiser.
The result is a set of custom stickers, optimized by AI, that advertisers can choose from. Everyone hasn't adopted them yet, and a few have.
It is one example among a bunch of similar tools. Together they sit behind the 3 or 4 percent increase in conversion that Meta reported on its earnings call.
How does AI learn a good click target?
A multimodal model turns images and text into the same kind of numbers, so it can compare how a design looks with how it performed. OpenAI's CLIP, published in 2021, is a well-known example, trained on 400 million image and text pairs. Given enough past click targets labeled by how often people tapped them, a model can pick up which visual patterns tend to draw taps.
Why don't more clicks guarantee more sales?
Click-through rate counts taps, while conversion rate counts how many of those visitors go on to buy or sign up. An accidental or curious tap adds to the first number and nothing to the second. Advertisers therefore track both, along with the cost of each purchase.
More from The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
