iOS 14 & SKAdNetwork: The most robust solution | AppsFlyer

Get ready for iOS 14 with AppsFlyer

iOS 14 is changing the way the industry measures attribution. Stay up to date with recent news and learn more about our solutions.

Our iOS solutions

PRIVACY-CENTRIC ATTRIBUTION

Our attribution suite provides 100% coverage with privacy built in.

SDK FOR IOS 14

Unlock full iOS 14 readiness, supporting APIs, ATT framework, SKAdNetwork integration and more.

SKADNETWORK

Holistic SKAdNetwork management allows you to measure, visualize and optimize easily.

WEB-TO-APP UA

Tap into the web for new growth opportunities with measurable campaigns that offer seamless web-to-app journeys.

Leverage the industry’s most powerful attribution suite

Don’t compromise on accuracy, coverage, or end-user privacy. Get the most out of the industry’s most advanced attribution technology, designed to deliver campaign-level data without infringing on user privacy. Privacy-Centric Attribution is an elementary part of AppsFlyer’s iOS 14 solution suite, giving advertisers the tools to continue to measure attribution accurately, while maximizing end-user privacy.

iOS 14 advanced privacyAdvance to Aggregated Advanced Privacy

The Aggregated Advanced Privacy framework puts you in the driver’s seat, limiting user-level attribution data available to both advertisers and partners, while maintaining the freedom of AppsFlyer’s customers to make their own choices and configurations.

SK360: The most robust SKAdNetwork solution in the industry

  • OPTIMIZE

  • ANALYZE

  • PREDICT

  • PROTECT

  • CONNECT

Strategize and optimize your conversion value schema

Map, manage and experiment with your conversion value schema in our self-serve dashboard. With advanced control over KPI measurements, you can make the most out of the SKAdNetworkframework.

Learn more >>

All of your SKAdNetwork insights in one place

View simulations and visualizations of your critical performance KPIs. Explore advanced visualizations of performance KPIs, including ROI, CPI, ARPU, ROAS and more. With extensive insights, data trends and deep-dive metrics, you can make informed decisions about your iOS campaigns with confidence.

Learn more >>

Take informed action on iOS campaigns based on short-term insights

COMING SOON!

Take action quickly, overcoming the SKAdNetwork timer limitations. Based on early signs of user engagement, our predictive analytics engine lets you put mobile attribution on “autopilot”.

Learn more >>

Keep your data and money safe from all types of mobile fraud

Make sure you’re getting real, accurate data on your campaign performance with data accuracy validation for SKAdNetwork. Protect your ad spend with 360-degree fraud coverage in the new iOS reality.

Learn more >>

Connect with your favorite integrated partners

AppsFlyer continues to deliver seamless, self-serve integration with your chosen advertising partners. End-to-end cooperation ensures that postbacks, conversion value schemas and data are delivered to AppsFlyer and your chosen partners smoothly and easily.

Learn more >>

Take advantage of your existing assets with Web to-App UA

Mobile websites are an excellent channel to educate, engage and drive growth among users, while maintaining full control over the user experience with the brand. Paid web campaigns that drive high-intent traffic to your website and convert to app installs can be efficiently measured with AppsFlyer, with full visibility into every step of the funnel.

Web Campaign-to-App

Resources

App Clips Guide

A comprehensive guide for developing your first App Clip and elevating the iOS UX

Read more

SKAdNetwork Data Study

SKAdNetwork doesn’t capture all non-organic attribution data. Our findings show that SKAdNetwork alone is not enough

Read more

Zeroed IDFAs

Why are 35% of all iOS devices showing zeroed IDFAs before ATT enforcement?

Read more

FAQs

What is Probabilistic Modeling?

Probabilistic Modeling is a statistical technique for estimating campaign performance, which can’t be used to uniquely identify a user and/or device. Probabilistic Modeling leverages machine learning to estimate campaign performance without compromising on privacy. Unlike fingerprinting, which seeks to maximize captured data points from each user to create a unique identifier that can be used to track users over an extended period and across websites, AppsFlyer’s probabilistic modeling seeks to do the exact opposite; minimize the captured data points and prevent the ability to create a unique persistent or permanent identifier that can be used to identify a user and/or device and is not used for targeting or profiling. 

Probabilistic modeling measures the app developers’ owned creative and campaign details, not data from the apps in which the ads are served. Moreover, in most cases, the app in which the ad was served is unknown.

Where fingerprints are used to create detailed profiles of users to enable precise targeting (at any point in time and on any site), probabilistic modeling is employed with the sole purpose of estimating aggregate campaign performance of paid and owned media (such as websites, social media platforms, emails, and user referrals).

While fingerprinting is used as a proxy to aggregated campaign performance, we have found that more accurate measurement can be achieved on an aggregated level in a non-deterministic world. Simply because the machine learning goal is to maximize the accuracy of the campaign’s aggregated performance, not the end-user attribution accuracy.

In other words, while the goal of fingerprinting attribution is to try and minimize false-positives and negatives, probabilistic modeling can provide much better aggregated campaign performance accuracy and much better user privacy. This is why lookback windows cannot be defined.

What is the difference between Fingerprinting and Probabilistic Modeling?

Fingerprinting:

Fingerprinting (often referred to as browser fingerprinting) is a term associated with the process of collecting a broad range of computer and browser information through a user’s web browser to uniquely identify a user and/or device. Fingerprints are used to identify a device when other persistent identifiers such as cookies cannot be read or stored by a website. A fingerprint is created by combining various data points from a device and/or browser that are accessible to every website when visiting their page. This could include the browser version, installed browser extensions, and plugins (including their versions), hardware properties, list of fonts, canvas and WebGL, HW benchmarking, language, time zones, OS version, screen characteristics, and menu bars.

Statistically, there is a very small chance that two or more devices will have identical set-ups and configurations, you can view how unique your configurations and setup are here. Fingerprints are especially strong given they can detect a device over an extended period even if certain parameters of the fingerprint are altered, with IP addresses changed, or even behind a VPN.

While fingerprinting techniques were initially created as a method for banks to detect fraud and prevent identity theft, today, they are used to track users across websites in order to compile long-term records of individuals’ browsing histories, and deliver targeted advertising or targeted exploits to users; thus, raising serious privacy concerns. For this reason, browsers (Safari, Chrome, and Firefox to name a few) have recently begun making changes to limit the amount of data they expose to websites, and to make users look more alike, creating a kind of “herd immunity”.

The good news is that mobile browsers provide much less data, and already provide the aforementioned “herd immunity” against the type of fingerprinting common on desktops. iOS devices especially are much less fragmented, making them greatly immune to fingerprinting.

AppsFlyer’s Probabilistic Modeling:

Probabilistic modeling is a statistical technique to estimate campaign performance and can’t be used to uniquely identify a user and/or device. It leverages machine learning to estimate campaign performance without compromising on privacy. Unlike fingerprinting, which seeks to maximize captured data points from each user to create a unique identifier that can be used to track users over an extended period and across websites, AppsFlyer’s probabilistic modeling seeks to do the exact opposite; minimize the captured data points and prevent the ability to create a unique persistent or permanent identifier that can be used to identify a user and/or device and is not used for targeting or profiling. 

Probabilistic modeling measures the app developers’ owned creative and campaign details, not data from the apps in which the ads are served. Moreover, in most cases, the app in which the ad was served is unknown.

Where fingerprints are used to create detailed profiles of users to enable precise targeting (at any point in time and on any site), probabilistic modeling is employed with the sole purpose of estimating aggregate campaign performance of paid and owned media (such as websites, social media platforms, emails, and user referrals).

While fingerprinting is used as a proxy to aggregated campaign performance, we have found that more accurate measurement can be achieved on an aggregated level in a non-deterministic world. Simply because the machine learning goal is to maximize the accuracy of the campaign’s aggregated performance, not the end-user attribution accuracy.

In other words, while the goal of fingerprinting attribution is to try and minimize false-positives and negatives, probabilistic modeling can provide much better aggregated campaign performance accuracy and much better user privacy. This is why lookback windows cannot be defined.

The following table presents an easy way to view the significant differences between traditional browser fingerprinting and probabilistic modeling:

 FingerprintingAppsFlyer’s Probabilistic Modeling
Equivalent to unique and/or persistent IDYesNo
Can be used to uniquely identify a deviceYesNo
Can be used to track users across sites/apps (cross-site tracking)YesNo
Relies on vast amounts of user-device and browser dataYesNo
Can be used to create profilesYesNo
Can be used to target usersYesNo
Can identify a device even when users hide behind VPNs or alternate IPsYesNo
DeterministicYesNo
PrivacyCan be leveraged in intrusive waysFriendly
Is AppsFlyer’s Probabilistic Modeling aligned with Apple iOS 14 guidelines?

AppsFlyer is a first-party software-as-a-service used by app developers and advertisers as an extension to their technology stack, similar to a CRM. AppsFlyer allows developers to manage, analyze, and secure their first-party end-user data, while complying with privacy regulations and platform policies, such as the ones recently introduced by Apple.

AppsFlyer’s Probabilistic Modeling is a statistical technique to estimate the aggregated campaign performance and can’t be used to uniquely identify a user and/or device. It leverages machine learning to estimate campaign performance without compromising on privacy. Unlike fingerprinting, which seeks to maximize captured data points from each user to create a persistent or near-permanent unique identifier which can be used to track users over an extended period and across websites, AppsFlyer’s probabilistic modeling seeks to do the exact opposite; minimize the captured data points and prevent the ability to create a unique persistent or permanent identifier that can be used to track a user.

While fingerprinting is used as a proxy to measure campaign performance, we have found that more accurate measurement can be achieved on an aggregated level in a non-deterministic world. Simply because the machine learning goal is to maximize the accuracy of the campaign’s aggregated performance, not the end-user attribution accuracy.

In other words, while the goal of fingerprinting is to try and minimize false-positives and negatives, probabilistic modeling can provide much better aggregated campaign performance accuracy and much better user privacy. This is why lookback windows cannot be defined.

Probabilistic modeling is about answering a very simple question: did the consumers find value in my paid, earned, and owned media such as websites, social media platforms, emails, user referrals? Probabilistic modeling measures the app developers’ owned creative and campaign details, not data from the apps in which the ads are served. Moreover, in most cases, the app in which the ad was served is unknown.

Similar to SKAdnetwork, AppsFlyer’s aggregated advanced privacy prevents cross-site tracking and the ability to uniquely identify a user or device. Moreover, AppsFlyer’s aggregated advanced privacy takes SKAdNetwork privacy levels as a minimum benchmark for its privacy-preserving thresholds. To that end, we believe that probabilistic modeling combined with our aggregated advanced privacy, is aligned with Apple’s iOS 14 guidance, and the reason why we made it the default setting. With that said, we recommend our customers to review the Apple app developer agreement and guidelines, and decide how to approach their aggregated advanced privacy settings, their partners’ integrations, and data collecting policies to ensure that their app is compliant with iOS 14 guidelines.

What is Aggregated Advanced Privacy (AAP)? Does it comply with Apple guidelines?

Similarly to SKAdnetwork, AppsFlyer’s Aggregated Advanced Privacy (AAP) aims at preventing cross-site tracking and preventing the ability of uniquely identifying a user or device. AppsFlyer’s Aggregated Advanced Privacy takes SKAdNetwork privacy levels as a minimum benchmark for its privacy-preserving thresholds.

While deterministic user-level attribution is an accurate proxy to aggregated campaign performance, we have found that more accurate measurement can be achieved on an aggregated level in a non-deterministic world. Simply because the machine learning goal is to maximize the accuracy of the campaign’s aggregated performance, not the end-user attribution accuracy.

In other words, while the goal of user-level attribution is to try and minimize false-positives and negatives, aggregated methods are not bound to user-level constraints, and therefore can provide better accuracy and much better user privacy.

AAP uses a combination of statistical probabilistic modeling, machine learning to create a proxy set in order to estimate the aggregated campaign performance LTV and ROI. In the near future, we plan to use deterministic attribution extrapolation, data sampling, and cohorts extrapolations to measure aggregate campaign performance, LTV and ROI. We are also planning to add more privacy thresholds to improve privacy to make cross-site tracking technically impossible and improving on the SKAdNetwork privacy benchmark.We believe that our AAP solution complies with Apple’s iOS 14 guidance, hence why we made it the default setting. Moreover, SKAdnetwork introduces ad-fraud risks, and AAP is essential for protecting app developers from potential ad-fraud.