Should the revenue of organic installs be included in a ROAS calculation?

Marked as spam
Posted by unknown (Questions: 1, Answers: 1)
Asked on June 5, 2019 9:30 pm
Private answer

Yes, but some things to consider:

- organic users' monetization performance often is worse than that of paid users. Applying an uplift on installs is likely skewed: if you arrive at an estimate that organic installs uplift is 0.2 of paid installs, most likely the appropriate revenue uplift factor is more like 0.1 or less (depending on how high the quality of your paid users is)

- when calculating the uplift factor, analyze your baseline granularly: search traffic that you get based on competitor or category keywords moves significantly less than search traffic for your brand. And if there is not a significantly different trend in organic traffic in countries where you run paid acquisition vs. where you don't, then likely other factors are at play. 

- we've experimented with loads of different models, and what has been working best for us is rather going aggressive on attribution lookback windows instead of applying generous static uplift factors. For example: Facebook's default setting is 28d click and 1d view attribution. If you set this to 28d click and 28d view, Facebook will already track around 20-30% more revenue than with the default setting. Of course an argument can be made that Facebook is incentivized to overstate their effectiveness (versus other channels), but I believe the accuracy of their method is still better than a static factor model. 

- In most cases I've seen and worked on, after a launch period the correlation between organic and paid traffic becomes smaller unless you change your campaign strategy (chart pushes / volume campaigns have a different organics impact than strict direct ROAS campaigns). 

- because of that last point, in general I'd advise caution unless you're in some kind of competitive landgrab situation. The potential bottom line contribution you burn by misattributing organic revenue and overspending on UA can be way more painful for your business' health than scaling a little slower. Monitor your bottom line, and if it looks like this, something's going wrong:

(these are things I have observed and applied for mobile games - I don't have experience with other types of apps)

Marked as spam
Posted by justin_stolzenberg (Questions: 1, Answers: 2)
Answered on June 7, 2019 6:09 am
Private answer

I think philisophically you have to decide how much of a role paid marketing plays in your product's organic magnetism. If you think all traffic would drop to 0 on a long enough timeline if paid marketing campaigns were shut off, then I think it's worth including revenue from organics into your paid acquisition ROAS calculations -- there'd be no organic without paid. More on that in this answer: How can I measure the impact of brand marketing campaigns on my product?

I see developers get into trouble when they drastically change their marketing budgets and project forward assumptions that don't materialize about how organic growth will change. If you have a measured / observed virality coefficient, I don't think it's fair to assume that it grows larger when spend increases; it could just as easily go down as go up, especially if your baseline spend is quite high. I've seen developers build models that bake in assumptions about how organic installs will increase at a ratio greater than the increase in their spend just because they need that organic revenue in their ROAS calculations in order to break even on the increased spend. I tend to be conservative when projecting organic ratio growth against spend growth.

Marked as spam
Posted by (Questions: 42, Answers: 111)
Answered on June 6, 2019 11:14 pm
Private answer

If you wanted to be more aggressive and there is a correlation between paid and organic installs, you can attribute the slope increase that you get in organic for every paid install. For example if the organic install and paid install relationship is linear:

Y = Mx + B, where x is the paid installs and Y is organic

You can use the slope M as a % contribution of organic LTV.

Need to remember in your query to exclude a couple of things:

  • Don't count organics during a feature period
  • Short time frame, since over a long period of time since organics should dominate
  • Don't count installs from some kind of bursting effects
  • Need to somehow account for potential chart position 
  • Need to make sure that the segment youre correlating has the same underlying properties to eliminate heteroscedasticity
Marked as spam
Posted by cfarm (Questions: 5, Answers: 21)
Answered on June 6, 2019 6:39 am