At a glance: AppsFlyer models aggregate data that basic SKAdNetwork reporting is unable to provide.
SKAdNetwork (SKAN) reporting can be limited because:
- SKAN relies on conversion values (CVs) that are limited to 6 bits, resulting in 64 possible values (0-63).
- Apple sometimes withholds the actual CV (and returns data as "null") to preserve user privacy.
- Data is only available from within an activity window of between 12-72 hours and full lifetime value (LTV) isn't known.
SKAN modeled data solves these issues:
- AppsFlyer uses machine learning, Bayesian statistics, AppsFlyer-reported data, and SKAN data to model hidden data points.
- Advertisers get a more complete picture regarding app and campaign LTV performance
Note: When you change your conversion value schema, it takes 2 days for modeled data to become available.
Modeled data types and availability
|Null conversion values||
|Revenue LTV||Overview dashboard Single Source of Truth view||Revenue LTV modeling is currently limited to 7 days|
|Geo||Overview dashboard Single Source of Truth view|
Null conversion values
Sometimes the conversion value (CV) has a null value because Apple withholds the actual CV to preserve user privacy. In AppsFlyer, a null CV is counted as an install. Consider:
- Null and zero have different meanings:
- Zero: The user installed the app but didn't perform any measurable action during the activity window. Consequently, the metric is accurate and isn't modeled.
- Null: We don't know what the user did; Apple withheld the CV.
- Null CVs can distort your performance metrics. To overcome the distortion and provide a truer measure of app performance, null conversion values are modeled to conversion values between 0-63.
Modeled conversion values:
Use the distribution of SKAN installs with conversion values to model null installs. The algorithm used to model data is dynamic and takes into account several variables. See the following example of how it might work:
- There are 100 installs, 60 null, 10 with CV=1, 5 with CV=3, and 25 with CV=5.
- This means that of the 40 installs with conversion values, 25% have CV=1, 12.5% have CV=3, and 62.5% have CV=5.
- The 60 null installs get the same modeled distribution: 15 get CV=1, 7 get CV=3, and 38 get CV=5.
- Are included in the SKAN dashboard and API metrics. The original, unaltered metrics, that don't include modeled data, are available alongside these metrics and are marked as unmodeled.
- Use data from the following granularity levels: App, media source, campaign, and adset. If modeled data using those levels isn’t available, the value N/A displays.
Modeled metrics available based on the conversion value:
- Event occurrences
SKAN reporting on its own can only provide revenue based on the SKAN measurement window and CV configuration for revenue in Conversion Studio. SKAN modeling is able to model the revenue up to 7 days (168 hours) post-install.
For SKAN version 3 and earlier, the modeled data extends beyond the timeframe of the activity window.
For SKAN version 4+:
- Data freshness: Revenue LTV becomes available before the second SKAN postback arrives (the postback delay is between 24-144 hours).
- Measure more: You can use the coarse value configuration of the second and third postbacks for other metrics, like events and retention, and still get revenue LTV data by modeling.
- Data enrichment: If you use the second postback to measure revenue, data from the postback is used to improve the modeled revenue LTV data and provide revenue data at a more granular level than the 3 levels provided by the postback.
- Revenue LTV metrics display for apps with revenue configuration in SKAN Conversion Studio, for revenue that comes from in-app purchases and ad revenue.
- Revenue LTV requires at least 14 days of revenue data. Meaning, if you configure Overall revenue on November 1, 2022, the first install date for which modeled data is available is November 14, 2022. The data displays 8 days after that (November 22, 2022).
- Revenue LTV (currently limited to 7 days) and other derived metrics are available in the SSOT dashboard (in the widgets, charts, and table) on day 8 post-install as follows:
- Revenue D7
- ARPU D7
- Gross profit D7
- ROI D7
- ROAS D7
SKAN postbacks don't include a geo dimension. Therefore, SKAN data can't be grouped by geo using just SKAN postbacks. However, AppsFlyer is able to model SKAN install geos to provide the geo dimension and grouping for SKAN data.
- Use machine learning tools that take into account the geo-distribution for campaign installs attributed via other AppsFlyer methods.
Are available via the Geo dimension in the SSOT dashboard.
Note: If you group data by both Geo and Attributing method, you can see the data for each geo split between AF model and SKAN uniques. SKAN uniques are modeled.
- The accuracy of AppsFlyer geo-modeling is approximately 90%. If the geo can't be modeled, the SKAN install is grouped under None. None means the geo is unknown.