Translate ads into other languages with AI dubbing and prompted copy

Luke Harries… and that's where we've really lent into AI. As a side note, we initially went down the route of you pay an expensive SaaS app for localization, you pay for an agency to then who have people to localize it. But I kept on having our team just sending screenshots of ChatGPT being like, "Oh, we should use this phrasing instead is way better." And I was like, "Well, if the LLMs are way better at translation than these agencies and expensive tooling."

Luke HarriesSo, we actually ripped out all the localization infrastructure. We built our own GitHub action and there's a thin library which just extracts the strings or wraps all the strings. And then each time we make a PR to our repo, it just sends it with a prompt per language about the localization guide, and that means we're able to localize way cheaper and much faster. And then once we've also localized the apps, you're then able to localize all the ads. There's stuff like our dubbing model, or again, just passing the copy with good prompts gets fantastic results. We think bottom's up, what's that next bottleneck for us to scale another 20% week on week, month on month? And then you start going through, okay, first of all, it's like funnel optimization, then it's probably starting to think about stuff like localization and so forth.

David BarnardAnd how do you think about measurement? And are you mixing in brand and other things that are harder to measure? I mean, with $100 million to spend, there's a lot more room to experiment and potentially put money into things that will pay off more in the long run. But are you looking at dollar for dollar ROI? Are you looking at for some channels, but not others? How do you think about measurement and effectively spending that?