Apple’s privacy changes are rolling out soon and to get a complete picture of the wider impact of this change, we have presented a three-part series on the MMP, Ad Exchange, and DSP perspective on how Apple’s updates will impact our industry.
We had also the opportunity to sit down with Nicol Cseko, VP of Product at Aarki to get her perspective regarding what isn’t being talked about SKAdNetwork that is important to address and have prepared you this BONUS article.
There has been a lot of talk about the impact of SKAdNetwork on marketing attribution and the optimal way to select events to track. But there has not been enough direct discussion about some of the challenges and considerations of adapting some of the more popular conversion models. In this article, we aim to fill that gap.
- There are a lot of challenges and tradeoffs with the time-based conversion tracking model in SKAdNetwork, and determining what to actually measure is not as easy as picking the same events tracked today. Many advertisers are still trying to find a model that will work and doesn’t require specialist data science expertise to build. While each app is unique, we have three general tips for advertisers:
Don’t assume that you should optimize towards the same KPIs as you do today. For example, if you optimize to ROAS today, it can be tempting to pick a revenue model for SKAdNetwork where you increment revenue or put users into revenue bins and only track those revenue signals via conversion values. Such a model needs to take into consideration the monetization curve of a typical / target user, and the selected conversion delay. For example, if your users do not typically make a purchase until 3-5 days after downloading an app, but you are only tracking revenue for the first 24 hours with SKAdNetwork, it will not be useful. Even if you do see a significant number of purchases in the first 24 hours, the cohorts defined by each conversion value must meet Apple’s privacy thresholds, which can be challenging if a given conversion value represents only a small portion of your audience.
Do rethink your strategy. Going with the earlier example, if an advertiser’s ultimate goal is revenue, but it doesn’t make sense to measure it directly with SKAdNetwork, they can find other user events to measure instead. An advertiser in this case should find early indicators of downstream revenue or lifetime value (LTV), and encode these indicators in the conversion values. It is important that those early indicators still make sense in terms of the SKAdNetwork measurement period and privacy thresholds.
Do consider maximizing your conversion value bits and consider tracking more than one type of conversion model. A mix of revenue, engagement, specific actions, or timeframes provides flexibility for your buying partners to test different types of campaign optimization.
- Confidence in SKAdNetwork as a source of attribution and planning for conversion declines until SKAdNetwork 2.2 is at scale.
Having extensively tested SKAdNetwork attribution and comparing to MMP attribution across both LAT and non-LAT users, the variance in install counts across most apps remains high when comparing the two sources of installs. The challenge is that with Apple, you will not get log level data to analyze, so a full comparison to MMP installs today is not possible. Some of our observations include:
- Large amount of variance is due to lack of view-through attribution (VTA) support with SKAdNetwork today. The good news is that VTA is coming with SKAdNetwork version 2.2, however, in the short term, this will cause a fair amount of turbulence during the transition, as a significant number of installs cannot be attributed until version 2.2 is supported at scale on the supply side. For advertisers benchmarking CPIs or other metrics, you will need to take this into account. Furthermore, as 2.2 support will require yet another SDK update from the supply side that publishers need to implement, we expect it will take up to a few months to fully scale VTA attribution.
- Even comparing click-based attribution with SKAdNetwork, we see significant variance for many advertisers, for both their LAT and non-LAT campaigns. App advertisers should be keeping a close eye on these numbers as a large variance can be related to errors in SKAdNetwork attribution implementation.
- Will sparse signals overly favor advertisers with the most budgets?
Given Apple’s privacy thresholds on conversion events, will smaller advertisers with limited scale and budgets face more challenges optimizing beyond conversions? These advertisers will need to make tough choices on defining conversion events tracked in the SKAdNetwork framework to be more mass, and potentially of less utility just to try to get enough conversion values to optimize on if budgets don’t support the non trivial privacy thresholds Apple seemingly has placed on conversion events per campaign ID. Will they be pushed to focus on installs while those with deeper pockets can translate on post install behavior? Apple’s attribution has the potential to hurt smaller app developers who have been able to effectively reach target users through today’s sophisticated mobile marketing ecosystem, even with smaller budgets.
The mobile advertising industry will get a full picture of the impact of privacy changes soon. Before that, make sure that you are prepared for all the possible scenarios to make the transition as smooth as possible.