Hacker News Reader

Comment by shostack | Hacker News Reader

Would you mind sharing some technical detail on how you used attribution as a backbone? What sort of modeling did you do around it? Just static attribution models? Or did you get into the dynamic algorithmic side of things like the big dedicated attribution platforms are doing?

View on HN

Attribution on mobile is a commodity now. 5 years ago before AndroidID and Apple IDFA, all you could use was IP, OS, Device Type, and random other bits of info the network would send you. So you'd need some machine learning inference to connect installed users to where they came from. Now, you just literally match the device-ids with what you get in the click from network partners (off-Facebook). Adjust, Appsflyer both do this.

On FB, just use FBSDK and it comes with its own attribution. They are starting to shift to a "multi-touch" model where you don't really know how they attribute -- its a black box that FB controls via their statistics. FB will report what the CPI is to you. It sucks, but thats market dominance for you.

What I'm also saying is, make sure you're also sending revenue, logins, and any other significant events to FB as well. Then you can figure out ROI and do more performance marketing. This works at any scale.

Thanks. I should have clarified--attributing an app install on Android to the source is definitely a solved problem. I was referring more to cross-channel attribution and determining dynamic weighting across multiple touch points.

Sounds like FB is working towards obfuscating that and not giving granular control which is disappointing to hear, but have you done any work around custom weighting like that?

Reply on news.ycombinator.com