Making SKAN actionable by modeling missing data

89% of mobile marketers in a recent webinar told us that measuring and optimizing iOS app install campaigns using SKAN data is extremely challenging.

In other words, it sucks.

That leaves three options. Growth pros can continue to mourn the loss of the IDFA, they can take a big non-compliance risk and focus their user acquisition efforts on the few remaining ad sources that enable fingerprinting, or they can put their big boy boots on, get with the program, and learn how to make SKAdNetwork work.

Good news: there’s help on the horizon.

Even better news: there’s help right here and right now, too. Including for data that is simply missing due to Apple’s privacy thresholds.

Apple’s SKAN privacy thresholds activate based on the number of installs per SKAN campaign ID in the previous 24-hour period. Fewer than 10-15, and you won’t see data. More than 20-25, and you will. Over 30, and your data gets significantly better. There’s a threshold for installs per publisher (because if you got too few installs from Publisher 732, you’d know with a high degree of confidence that the five recent users you acquired came from that source) and a threshold for conversion values (because if you got 25 new users from Publisher 732 but they all had somewhat varying conversion values, you could likely tease out individuals from the group).

Privacy is important. But how can mobile marketers still get the data they need for campaign optimization?

Modeling missing SKAN data

“We’ve been spending a lot of time and energy trying to figure out how this mechanism works,” Singular product manager Evyatar Ram told me recently in a Growth Masterminds podcast episode. “For 80% of those installs, you’ve got a conversion value. And for the remaining 20% … we’re trying to model or estimate what would be the conversion values.”

The technology is SKAN Modeled Metrics, and it allows marketers to get accurate modeling of revenue, events, and conversion value counts despite the fact that 20% of their conversions are missing.

How accurate?

The more data you have, the more accurate it will be. And you’ll always know exactly what the accuracy level is, Ram says.

“One of the things we’re trying to do is just be really transparent about it,” he says. “So every time we present this metric, we’ll show you a confidence interval that … basically gives you the range of values which we are pretty certain that the real number resides in.”

In other words, if the confidence interval is 20, marketers can be 90% sure that a modeled and reported number of 100 is 90% likely to be between 80 and 120. And that gives you the ability to make informed decisions about ad spend. The alternative, of course, is that you have campaigns with 20% of not just all your data, but 20% of your most valuable data — the conversion data — missing. That 20% makes the difference between a campaign that’s a dog and one that is a lion.

The point?

Actionable data to improve results.

SKAN: low scale equals bad data

One other important point: mobile marketers need to concentrate iOS campaigns to varying levels with different partners to ensure that they exceed privacy thresholds.

A single ad network campaign could be built with multiple SKAN campaign IDs. Networks are doing this to get insight on targeting, creative, and other components of the campaigns so they can improve performance, but it has a negative impact on the scale marketers ultimately end up with in each SKAN campaign ID.

Remember, you only get 100 SKAdNetwork campaign IDs.

Some networks, like Aarki and Liftoff, pretty much have a 1:1 ratio between internal campaign IDs and SKAN campaign IDs. Others, like Applovin and Facebook, use far more SKAN campaign IDs per overall campaign.

How each ad network manages their business is their business, but you should be aware of this ratio when designing campaigns and scaling campaigns. Low-scale campaigns with a high-ratio partner is going to result in limited SKAN conversion data, meaning you’re going to have to trust that your ad partner is delivering the value they’re claiming … because you won’t be able to verify it. The solution is to align high-volume campaigns with those partners, and low-volume campaigns with low-ratio networks.

We have literally seen a network peak at a 49:1 ratio of SKAN campaign IDs to internal campaign IDs.

If that stayed consistent over time, you’d need 1,500 installs a day to escape privacy thresholds. For one partner …

(Caveat: assuming that they’re evenly distributed between SKAN campaign IDs.)

That’s going to impact all but the highest-volume campaigns. And it’s going to impact the network itself, because if marketers can’t verifiably drive impact with them, they’re going to look elsewhere.

Align scale with partners; model the rest

Once you’ve aligned the scale of your campaigns with your partners (and yes, you can just ask them, or check their documentation regarding how many SKAN IDs fit into each of their campaign IDs), you can model the rest to have a high degree of confidence that you have good data — not perfect, but good — valuable information, and actionable insight.

So many mobile marketers are struggling with iOS 14.5, iOS 15, SKAdNetwork, and the loss of the IDFA.

But I am seeing many break through into the light at the end of the tunnel.

Listen or subscribe to Growth Masterminds here

Talk to an expert

Need some help to work through it all? No problem.

Book some time, get a demo, and share your frustrations. There is a path through.