Attribution Model

Attribution Model

What is an Attribution Model?link

Unlike the conventional web marketing analytics tools, which immediately directs Users to websites based on clicks, a mobile App will require an additional process to go through before installation: App Market. Accordingly, a number of impressions and/or clicks are needed to get to actual App Install. During this procedure, the Attribution Model allows Users to check which touchpoint, out of multiple impressions and/or clicks, has actually attributed to App Install in the end. For example, let’s assume that a User has clicked on a Facebook Ad once and also a Naver Ad three times before App Install. In this case, there is a single install, but there are four attribution clicks. So, the single install will be attributed to total number of four clicks. Why so? This is because counting attribution for four different clicks can lead to an overdraft of performance.

Now, there are many ways to acknowledge touchpoints (i.e. clicks, impressions) for the single install. First, all four clicks can be recognized for App Install, each by 0.25 attribution. Also, a linear attribution can be applied, by giving attribution ratio such as 0.1, 0.2, 0.3, and 0.4 each. Another way is to recognize the final click for install by a ratio of 1. While doing so, the most common model to credit attribution to the very final touchpoint, which assigns 100% attribution to it, is called the ‘Last Click’ Mode.

‘Last Interaction’ Model is attractive given that it maintains the performance (install) in an integer format, while assigning credits. For instance, if the install performance of a particular channel is 11.27, then the marketer will have quite a difficult time understanding the number (in this case, there can be even over 20 attributing installs). In contrast, if a particular channel is credited for 10 installs, then understanding will be much easier. Therefore under this model, a complete 1 install will be assigned to a particular touchpoint. Here, the very final click which was assigned with a 100% attribution is referred to as ‘Winning Influencer’. In the example mentioned above, a single click out of four clicks have ‘won’ the remainders to take a 100% attribution.

Currently, Airbridge has an Attribution Model which is based on ‘Last Interaction’ Model. In addition, Airbridge owns a separate Attribution Model which combines 6 principles to the basic ‘Last Interaction’ Model. Airbridge ‘Install’ Attribution Model will be explored in the following with more details.

In fact, not only the installs, but across all conversions that take place within an App can be grouped as follows: (a) Install; (b) Launch(Open); and (c) in-App Events. The reason why Attribution Model is equally need for launch or in-App Events is because it can also lead to revisits via Deep Links, in addition to mobile App installs. Such revisit campaigns based on Deep Links have recently gained traction under the name of ‘Remarketing’. Since an accurate attribution analytics over Deep Links is required for in-App launches and in-App events, Airbridge owns a separate Attribution Model which is designed for launches and in-App events. Read more on Airbridge’s Attribution Model for ‘Launch(Open)’, and ‘In-App Event’ in the following.


What is a Lookback Window?link

A lookback Window is a requirement and also a supplementary concept to the Attribution Model, which defines the time of the Attribution period upon the touchpoint that was made shortly before the actual conversion (Install, ‘Launch(Open)’, In-App event). In other words, it is about to which point in time a touchpoint would be considered as an attribution. For example, a User has clicked on an ad banner through Google AdWords 365 days ago, and installed the app today. Assuming that there was no clicks in between, and also under the ‘Last Interaction’ Model, that particular click from 365 days ago, will be credited for the install attribution. But it is quite difficult to acknowledge a very old last touchpoint for an install, ‘launch(open)’, or in-app event.

Thus, a Lookback Window will limit the time frame to which certain touchpoint has attributed to a conversion in effect. Lookback Window can be adjusted by Users later on, and for now, Airbridge applies default Lookback Window, depending on same-User identification methods (Airbridge will soon release Lookback Window adjustment feature as a whole, and also by channels).


Airbridge ‘Install’ Attribution Modellink

On Airbridge, the currently adopted Attribution Model is an ‘Ad as an Accuracy-based Last Interaction Model’, for determining winning influencers to an ‘install’. Simply put, a priority order will be assigned to multiple touchpoints along the attribution path.

  1. Ad touchpoint is prioritized over a non-organic ad touchpoint (Non-Organic Prioritized).
  2. Across touchpoints, clicks are given a priority over impressions (Click-Prioritized).
  3. On the touchpoints in regard, the priority is given in accordance with its technological accuracy (Accuracy-Prioritized).
  4. Under the same level of technological accuracy, the priority will be given in the order of Last Clicks (Last Click).
  5. When timestamp data cannot be collected, as under the two cases (Google AdWords UAC, Facebook Ad), those with the last click will be given the priority, among those under same level of accuracy (Timestamp-Prioritized).
  6. All touchpoints must be made within the Lookback Window (Attribution Windowed).

When the rate of attribution is derived according to the priority above, the first touchpoint in rank is referred to as a Winning Influencer. Basically, the channel, which has the touchpoint assigned as a success factor, will get the install count in the very end.


‘Install’ Same-User Identification and Lookback Windowlink

Airbridge employs the following five types of technology to identify if a User is the same User before and after the install. Moreover, a fit Attribution Window depending on each matching technology will be assigned for Lookback Window in default, as of now.

CategoryTechnology TypeAccuracyLookback Window per ClickLookback Window per Impression
Deterministic MethodAndroid ReferrerApprox. 100%
(Loss -10%)
0-30 days (min. 7-day)-
ID Matching (IDFA, GAID)Approx. 100%0-30 days (min. 7-day)-
Platform MatchingApprox. 100%0-30 days (min. 7-day)-
Deeplink Deferred Install MatchingApprox. 100%0 ~ 30일 (min. 7-day)-
Cookie Matching
(iOS Safari 9.0 Above)
Approx. 100%
(likely to have Loss)
0-30 days (min. 7-day)-
Probabilistic MethodProbablisitc Matching
(IP Address + User-Agent)
(Fingerprinting)
Approx. up to 85%0-96 hrs (min. 24 hrs)-
  • Android Referrer: When clicking on the ad to send User to Android’s Google Play Store, Android Referrer query string parameter will be sent along with it. This method can distinguish wif the User is the same User before the app install and after the initial launch(open), based on Android system (Applicable for all Android Users).

  • ID Matching: This is a method of checking if a User who has made ad impressions/clicks and a User who launched the App for the first time via ADID and IDFA, which are Google and Apple’s ad IDs for each user device, are the same User (Applicable to Users of Android and iOS who had approved for use)

  • Platform Matching: For Google AdWords and Facebook, same cross-platform User identification is made through App Conversion Confirmation and App Events API. Accordingly, Google AdWords and Facebook advertisers verify that they are the same User through communication with respective servers.

  • Deep Link Deferred Install Matching: Users are sent to Google Play Store for install. In this case, when Deep Link is added to the “URL” query string parameter in address, after install, "Continue" button will be displayed instead of "Open". If you click on “Continue”, a Deep Link will be called, from which the User information can be checked, whether the first User who launched the App is the same User who clicked on the link (This method cannot be used when moving to the Google Play Store using Intent, such as JavaScript or tag on Chrome).

  • Cookie Matching: This is a verification method to see whether cookies left on web from ad impressions/clicks via Safari View Controller are identical to the mobile App from the first launch and verifying that they are the same (Applicable for iOS 9 Users who click on the link via Safari).

  • Probabilistic Matching: This method gives a probabilistic guess based on the combination of non-unique values ​​(i.e. device information, IP), on whether the User who has made clicks/impressions on the ad and the User who launched the App for the first time is the same User (Applicable to both Users of Android and iOS, but yields priority to three methods as mentioned above, given that it is a probabilistic method).


Test Scenariolink

Scenario 1link

When there is a conversion path as follows, depending on Airbridge ‘Install’ Attribution Modelthe winning touchpoints for the install becomes ‘Ad clicks for 3 times/ID Matching’, and the following explains why.

3 6 1 01 en

  1. Only the touchpoints that fall under Lookback Window are considered (Attribution Windowed) → ‘Ad Clicks for 1 time’ is disqualified
  2. Out of the touchpoints, those with high rate of matching accuracy are given a priority (Accuracy-Prioritized) → ‘Ad Clicks for 4 times’ is disqualified
  3. Within the same range of matching accuracy, touchpoints with last clicks are given a priority (Last Click → ‘Ad Clicks for 2 times’ is disqualified
  4. A winning touchpoint for install is ‘Ad Clicks for 3 times’

In case a conversation path as mentioned above had taken place, and if the Lookback Window is shorter so that all three of the ad clicks were not included, despite it being a low rate of matching accuracy, the winning touchpoint becomes ‘Ad Clicks for 4 times’ within the Lookback Window.


Airbridge ‘Launch(Open)’, ‘In-App Event’ Attribution Modellink

Meanwhile, while determining winning influencer for ‘Launch(Open)’ and ‘In-App Events’, Airbridge adopts the ‘Last Deep Link Execution Priority Model’. That is to say, when under the same Lookback Window, Deep Link revisits will be credited with higher attribution than Winning Influencer.

  1. The launch of Deep Link itself instantly becomes a part of performance by that click in regard. In contrast, for non-Deep-Link general executions and in-App events, the following method will be observed to assign Attribution rate.
  2. Between Deep Link Launch(Open) and existing Winning Influencers, a priority will be given to Deep Link Launch(Open). Simply put, when a conversion is made, whether Deep Link has been launched or not will be assessed within the Lookback Window. During this procedure, the last Deep Link Launch(Open) must have a sooner timestamp to the lastest install (Deep Link Prioritized).
  3. All Deep Link Launches must be made within Lookback Window (Attribution Windowed).
  4. Ad Deep Link Launch will be given a priority over non-ad Deep Link Launch (Non-Organic Prioritized).
  5. Within a common Deep Link launches, the last click touchpoints will be given the priority (Last Click).

‘Launch’, ‘In-App Event’ Same User Identification & Attribution Windowlink

Airbridge currently employs the following Lookback Window on ‘Launch(Open)’ and ‘In-App Events’ over the lastest Deep Link that was executed.

Subject of ComparisonLookback Window
The last Deep Link touchpoint (excludes Deep Link Deferred Install Matching)0-30 days (min 3-day)
The last Winning Touchpoint on a last install (includes reinstalls) 0-30 days (min 30-day)

### Test Scenario

Scenario 1link

When there is a conversion path as follows, depending on Airbridge ‘Install’, ‘In-App Events’ Attribution Model, Winning Touchpoints for the ‘Launch(Open)’ and In-App Events becomes ‘Ad clicks for 4 times/ID Matching’, and the following explains why.

3 6 1 02 en

  1. The Winning Touchpoint candidate group for In-App Events or Launch(Open) are Winning Touchpoint and Deep Link Launch that fall under Lookback Window, for install. → ‘Ad Clicks for 2 times’ is disqualified
  2. When there is a Deep Link Launch, a priority will be given over a Winning Touchpoint (Deep Link-Prioritized) → ‘Ad Clicks for 1 time’ is disqualified
  3. Across Deep Link Launches, the last clicked Deep Link will be given a priority (Last Click → ‘Ad Clicks for 3 times’ is disqualified
  4. A Winning Touchpoint for in-App Event and Launch(Open) is ‘Ad Clicks for 4 times’

Scenario 2link

When there is a conversion path as follows, depending on Airbridge ‘Install’, ‘In-App Events’ Attribution Model, Winning Touchpoints for the Launch(Open) and In-App Events becomes ‘Ad clicks for 1 time’, and the following explains why.

3 6 1 03 en

  1. The Winning Touchpoint candidate group for In-App Events or Launch(Open) are Winning Touchpoint and Deep Link Launch that fall under Lookback Window, for install (Attribution Windowed) → ‘Ad Clicks for 2 times, Ad Clicks for 3 times’ is disqualified
  2. A Winning Touchpoint for in-App Event and Launch(Open) is ‘Ad Clicks for 1 time’, which is the Winning Touchpoint for install

If you have any questions on the Attribution Model, please email us via ‘1:1 Inquiry’ in the bottom right corner, or write to us at Airbridge Support. We will be in touch with you promptly.