Leveraging side-by-side device level and SKAdNetwork data in Automate: Adjust’s solution
Marketers depend on measurement to determine whether or not a campaign has been successful. Undoubtedly, iOS 14.5+ and working with SKAdNetwork has posed significant challenges and a shakeup to strategy and processes. Marketers and developers have reframed their approach to measurement to best work with a mix App Tracking Transparency (ATT) opted-in data and aggregated SKAdNetwork data concurrently for campaigns on iOS — which is the new reality of attribution on the platform.
Both data sets offer unique benefits and drawbacks. Device level data allows for detailed, granular reporting, and building of user cohorts mapped to revenue indications, but due to the consent required the final picture is often incomplete and partly inaccurate. High-level SKAdNetwork data is relatively accurate when focusing on totals per partner, but fundamentally lacks granular performance reporting. This diversity presents marketers with the challenge of deciding which data points they should base their daily decisions on, and has led to the rise in demand for unified and simplified reporting.
There are, however, multiple issues that make merging device-level MMP data with aggregated SKAdNetwork data complex and unreliable, which is the key reason why Adjust recommends working with side-by-side data via Automate.
What are the problems when merging MMP data and SKAdNetwork data?
- Deduplication: Before even considering combining SKAdNetwork install data with ATT level install data, a deduplication that takes out the installs from one of the sources that are attributed in both data sets must take place. As the SKAdNetwork data set is aggregated and not on a device level, the deduplication can be attempted by introducing a dimension that splits the device level attributed installs from the non attributed. This is an early concept of deduplication that would consume at least one bit of your conversion values and will not solve all the other problems when thinking about how to merge ATT level and SKAdNetwork data.
- Randomized install dates: The install date received from SKAdNetwork installs is always randomized and is not clearly identifiable. This means that SKAdNetwork installs can be valid for anywhere between 0 and 48 hours before it is received, and complicates the ability to remove them from a data set.
- Google related installs: The install dates of SKAdNetwork installs from Google are even more complicated to work with. By applying a level of modeling, Google attempts to determine the date of the related ad interaction (clicks or impressions) and link it to an install. As Google is one of the biggest self attributing networks and part of most channel mixes, this heavily impacts the ability to merge data.
- Attribution methods: SKAdNetwork and device-level attribution work differently, and it’s highly likely distribution across channels is also significantly different between the two. This is why it’s still expected to have duplicates for some channels when aggregating, even when following deduplication as mentioned above. For example, some device level installs in multi-touch marketing scenarios attributed to Facebook might get attributed to Twitter via SKAdNetwork.
- Randomization at conversion value level: There are a specific amount of conversion values set to be null for SKAdNetwork. As the information regarding whether the install is attributed, or not attributed, is packaged into the values itself, when this value is nulled, the install cannot be identified as attributed or not attributed. The distribution of null values is not necessarily linear across all conversion values and depends on the amount of installs per campaign ID. This means we cannot simply extrapolate on a percentage of null values vs. installs with a conversion value. There can easily be up to 40% null values across all SKAdNetwork installs. To reach a high enough level of installs to overcome this threshold and have a big enough data set to work with, a relatively large campaign spend is required. Because many marketers are already grappling with the challenges surrounding the privacy changes, we don’t recommend splitting budget across multiple campaigns unless you expect to reach this high threshold of installs. This can heavily impact deduplication and is another key reason to be cautious when merging device-level and SKAdNetwork data.
- Differences by country/region: For marketers working across multiple countries and languages, the country dimension is usually extremely important, as costs and performance can vary heavily from market to market. For SKAdNetwork installs, there is generally no country information available — another reason why mixing data sets together, and then later going through to attempt to break them down by country can be very inefficient and inaccurate.
- Performance inaccuracies: In addition to the inaccuracies identified above in relation to install numbers, there’s the problem of not knowing the performance of additional installs that would be added when merging data. Getting to a valid total cost per channel might roughly work based on extrapolation and by ignoring the problems outlined above, but as soon as you get into campaign level, where marketers usually make their decisions, it is rendered more or less unusable. Getting related performance metrics such as day 7 and day 30 revenue for a report down to campaign level would also be inaccurate.
How Adjust’s side-by-side reporting empowers marketers in light of limitations
Depending on the individual circumstances including an app’s business model, the number of installs per campaign, ad spend, distribution across channels, number of campaigns in total and the overall ability to identify signals of importance with 24 hours, it’s down to the marketers and developers to determine which data sources are most trustworthy and which indicators from both data sources might correlate. If a correlation can’t be seen immediately, this is cause to dig into the data further.
Considering all of the limitations that exist, Adjust currently recommends using side-by-side reporting via our Automate feature. With a range of SKAdNetwork-related KPIs (conversion, revenue, and events), you can build comprehensive reports for SKAdNetwork campaigns and leverage the templates to see the information side-by-side with your Adjust attribution data. This enables you to compare iOS campaign data from both sources in an easy to manage tabular report.
Keeping the data sets split is the best way forward for ensuring accuracy. However, if deduplication and the merging of ATT level and SKAdNetwork data is the direction you and your team choose to take, we do still have you covered. We give clients control over their deduplication efforts as part of our dedication to providing 100% transparency and our SDK attribution real-time callbacks let you trigger an identifiable event that you can leverage within your conversion value configuration. For more information, reach out to your Adjust contact person.
We’re currently investigating ways to build reporting that shows a more complete picture while also ensuring that the attribution data is extremely accurate. In the meantime, we recommend continuing with side-by-side reporting to ensure the problems outlined above are avoided, and that reporting and data-sets remain clean. For more information on iOS 14.5+ you can visit our resource center here.
Craving monthly app insights? Subscribe to our newsletter.