9 steps to make SKAN predictive (and yes, it’s possible)
Apple’s SKAN framework has been a key part of mobile growth professional’s daily working life for eight months now. But 90% of mobile marketers still struggle to make it useful, to make it helpful, to make it as good as the good old days of IDFA.
In short, to make SKAN predictive.
Apparently, it is possible.
In fact, sophisticated marketers are finding ways to make SKAdNetwork as good as IDFA in terms of predictive power. That might sound insane to those who are still struggling to make it work at all, but it’s something that Singular CTO Eran Friedman has personally witnessed and in fact helped brands achieve.
If IDFA was a 10 for predictive capacity, I asked him, what’s SKAN?
“When I work with the more sophisticated companies who really made the effort to try to figure it out, I’ve seen this score increasing further and further,” he says. “We see cases in which companies are saying: ‘We’ve got to the same level of scale that we had with IDFA, happy about the performance, and we’re happy about the data that we get.’ So it did require some iteration and work to get there. But I would say that you can definitely get to the comparable level of what you used to have.”
Which, I guess, is basically a 10.
That doesn’t mean SKAN gives you everything you used to have. IDFA remains the gold standard, because it was (and remains, if you get permission via ATT):
- Rich in data
So the question then is: how can you make SKAN predictive? What precisely do you have to do to generate the kind of marketing campaign optimization value from SKAdNetwork that you once got from IDFA?
According to Friedman, there’s at least nine things you need to make happen.
9 steps to making SKAN predictive
1. Get the basics right
The first hurdle is to get the basics right. Ensure the wires are connected, the tubes are hooked up, and you are actually getting data back from SKAdNetwork. This may seem simple, even simplistic, but many marketers have had serious issues right at this very initial stage.
2. Utilize modeled data: conversions
Apple’s privacy thresholds mean you only see part of the data. With the right data science and machine learning, however, you can get high quality modeled SKAN data that tells you with a high degree of confidence what’s actually happening.
3. Create modeled data: cohorts
Once you have your missing data, you can build predictive models for cohorts as well, which you have to estimate since you don’t have precise data on specific install times. That helps build predicted LTV and helps you understand what you’re actually getting from campaigns measured with SKAN.
4. Optimize your conversion models
You might start with a revenue model, and then try an engagement model or a retention model. Ultimately, you want to find which model works best for you and your app. The good news: you can update SKAdNetwork conversion models on the fly, in the cloud, using the Singular dashboard.
5. Build in multiple signals to your SKAN postback
You only get one postback and it’s only 6 bits. But you can encode multiple things into that one postback, such as data on cohort, revenue, behavior, registration, sign-up, or more. (Check a primer on how to optimize your postbacks in SKAN.)
6. Adapt to your vertical and app
Most gaming apps, especially hyper-casual, need almost instant feedback for ad campaign optimization. Fintech might work with longer feedback cycles. Adapt your strategies to your specifica app and and your vertical, because your LTV and customer lifecycle calculations are likely very different than someone else’s.
7. Play some trial & error
Don’t expect to get it right on the first try. Expect to have some weeks of figuring it all out. Invest that time, because SKAN is not a fad. It’s essentially the one thing you can be sure of having to measure iOS campaigns. (Think you’ll always have fingerprinting? Don’t be so sure …)
8. Get operational datasets perfect: installs, ROI, cohorts, LTV
Your first priority is to get operational datasets running and functioning well. That means basic data like cost, installs, cohorts, LTV, and ROI. You need this for daily/weekly optimization and reporting. The good news is that you can get this with SKAN.
9. Consider high-level strategic insights: incrementality, media mix modeling
Once you have the basics in place and if you have additional resources in data science and engineering, look into sophisticated techniques like incrementality and media mix modeling for strategic-level insights over longer-term periods. This will give you insight on splitting your budget between channels, the overall blending impact of all your marketing efforts and spend, and more.
Subscribe to Growth Masterminds
Subscribe to Growth Masterminds on your favorite podcasting platform: