Attribution

Normalizing Measurement of App Installs

Lucas Brown

As the saying goes, it’s important in life and business to “give credit where credit is due”. This is certainly no different for performance marketing and advertising in which only those who deliver measurable results – i.e. the almighty conversion –  are credited accordingly. However, the question is: How, from a measurement perspective, do you compare and credit your various advertising partners if they are using various standards, e.g. a different attribution length?

The attribution length is the maximum time period between the last-click of an ad and the corresponding conversion of interest, in this case the install. Why is this important and how can it make such a big difference in performance marketing? Well, consider the following two scenarios.

You have agreed (as part of an insertion order with an advertising partner or part of the service agreements with self-service platforms) with your advertising partner (Ad partner 1) for them to use an attribution length of 7 days which allows them to have up to 7 days lapse between a click on their ad and a generated install for which they should be credited. Say an install occurs for which they had a click 5 days ago. In this instance, we would attribute said advertising partner with the install and they, of course, would likewise claim credit for the install. However, if an install occurred and their click happened 8 days ago, we would not attribute them with the install and they should not claim credit for that install (though there is certainly nothing stopping them from doing so and if you do not have an unbiased third party attribution solution, how are you even to know?).

In addition to your agreement with Ad partner 1, you also run marketing campaigns with another advertising partner with whom you use an attribution length of 15 days.  While the same attribution logic (within attribution length = credit for conversion, outside of attribution length = no credit for conversion) applies to both Ad partners, you cannot directly compare these two advertising partners’ performance as they do not have to hold to the same attribution lengths and therefore achieve very different performance results.

As you can see, while an agreed upon attribution window creates a standard to which both you and your advertising partner must hold, there is currently no “standard” attribution window length in the mobile app industry to which ALL must hold. For performance marketing and advertising, the lack of a standard attribution length makes it difficult to compare the performance of one advertising partner to another, and let’s face it, that is primarily what performance marketing is all about!

As the various advertising partners we have integrations with have different attributions lengths for measurement, our reported attribution must be able to reflect their differences and therefore we set the default attribution lengths for each one of them individually .

With our release of Multi-touch attribution, we “normalize” the measurement of installs by measuring the both the number of installs and the “assists” that occur after the advertising partner’s attribution length with a max attribution length of 30 days. We report these contributions that occur outside the advertising partner’s default attribution window as “Non-windowed Contributions”.

To learn more, read our article on Multi-touch Attribution with Attributed Installs, Install Assists, and Non-windowed Install Contributions.

Consider the following use case in which we compare two ad partners with and without our Multi-touch approach.

Ad Partner 1 has an attribution length of 1 day. Ad Partner 2 has an attribution length of 28 days. Let’s say you, the advertiser, purchased the same amount of clicks for the same cost on both of these advertising partners. If we calculate install rate or CPI based solely on Attributed Installs, you would end up with the following statistics.

 

Ad Partner 1 Ad Partner 2
Clicks 2000 2000
Attributed Installs 300 500
Cost 500 500
Install Rate 15% 25%
CPI $1.67 $1.00

 

Based on the above, Ad Partner 2 seems like they are the more effective advertising partner as they have a higher install rate and thus a lower cost per install. But remember, Ad Partner 1 has a much shorter attribution length, so it is not viable to compare Ad Partner 1 to Ad Partner 2 as they have different attribution lengths.

This is where Install Assists and Non-windowed Contributions come in extremely handy as they enable you to accurately compare advertising partner to the other, apples to apples. If we base Install Rate and CPI on Total Contributions (Attributed Installs + Install Assists + Non-windowed Contributions) then you can accurately compare one to the other using the same attribution window.

 

Ad Partner 1 Ad Partner 2
Clicks 2000 2000
Attributed Installs 300 500
Install Assists 10 10
Non-windowed Install Contributions 300 2
Total Contributions 610 512
Cost 500 500
Install Rate 15% 25%
CPI $0.82 $0.98

As Total Contributions includes all interactions an advertising partner had with new users within 30 days, and Non-windowed Contributions measures those attributed installs and assists outside of the advertising partner’s attribution length but within 30 days, you can compare multiple advertising partners easily and accurately with one standard.

The above scenario makes it abundantly clear how important attribution windows are and the impact they can have on the perceived performance of your advertising partners. You must keep these distinctions in mind when evaluating your advertising partners as well as determining the efficacy of your campaigns.

MAT Integrated Advertising Partner Attribution Lengths

We conducted a survey of the 343 advertising partners (as of end of January 2014) we are currently integrated with including various banner ad networks, video ad networks, DSPs and ad exchanges. We found that 17% were conducting install measurement using a 1 day attribution length; 34% with an attribution length between 1-7 days; 8% with an attribution length between 8-14 days; and 41% with an attribution length between 15-30 days.

As about half of our integrated publishers have attribution lengths above 15 days and the other  half below 15 days, it is important to understand the impact as outlined above on how Non-Windowed Contributions and calculating Total Contributions can allow you to measure performance of one to another – apples to apples.

Potential Downsides to Longer Attribution Lengths

Simply put, advertising partners with longer attribution lengths are attributed with more installs. There are two potential reasons that this occurs:

  1. The user downloaded the app but waited several days (e.g. 20 days) to open the app. Since third-party analytic providers can only track installs on first app open (not on download), the install would be tracked on the 20th day after the actual download.
  2. The user had some other influence to download and install the app. For example, an advertising partner shows an ad to a user and gets them to click (in some cases mistakenly). 20 days later, a friend of the user who uses the mobile app encourages them to download and install the mobile app. When the user installs the app we have no way of knowing about the friend’s persuasion and we would only have record of the last click and the install would be attributed accordingly.

Many advertising partners with longer attribution lengths believe that reason #1 justifies the longer attribution lengths. In doing so, the campaigns being ran with longer attribution windows initially seem better than campaigns being ran with advertising partners using shorter attribution lengths as a longer window allows for more time to generate a conversion.

Many of our clients (companies with apps) believe reason #2 most accurately describes reality and thus try to negotiate shorter attribution lengths with their advertising partners. Some of these clients believe friends of friends are a significant driver for new user acquisition, while other clients have significant television and radio campaigns that cannot be easily included in real-time attribution of their mobile app installs. As these influences cannot be taken into consideration during attribution and we can only link the responsible advertising partner to a conversion, these mobile app installs would technically be false-positives since the attributed advertising partner was not, in fact, responsible.

The jury is still out on whether longer or shorter attribution lengths result in more effective and accurate measurement. As our own internal poll showed, there is currently a great divide in the industry as regards the topic of attribution length as both sides of the equation (i.e. app developers/advertisers and their advertising partners) want to ensure that credit is being given where it is due.

Increased Probability of Duplicate Installs

A longer attribution length also increases the probability that two or more advertising partners will entice the same user to click on their ads. Thus without a centralized attribution platform implemented, like our MobileAppTracking product, you would end up paying two or more advertising partners for multiple “installs” when really there was only one mobile app install via one user. To learn more about how our product prevents you from paying for duplicate installs, please read about our last-click attribution model.

Consider the following scenario. A user clicks a campaign banner hosted by Ad Partner 1, but does not install the mobile app. Three weeks later, the same user clicks a campaign banner hosted by Ad Partner 2 and installs the app. The potential issue with a longer attribution length is that both Ad Partner 1 and 2 can claim credit for the conversion (and you pay both), when really Ad Partner 2 was the one that enticed the install. Therefore, decreasing the attribution length decreases the chance that you will pay multiple advertising partners for the same conversion.

By using a central attribution platform with a single SDK like MobileAppTracking, you can avoid paying out multiple times for a single conversion. As the single SDK of the attribution platform will automatically measure all installs and attribute the conversion to the last recorded click, the publishers’ SDKs is no longer needed and can no longer record conversions on its own for which they then assume responsibility.

Along with a longer attribution length is a probable increase in the number of potential Install Assists vs Non-windowed contributions. If two advertising partners enticed the same user to click on their ads and had longer attribution lengths, one would get credit for the Attributed Install and the other an Install Assist. If they had shorter attribution lengths, one might be credited for an Attributed Install and the other a Non-windowed Install Contribution or even both get credited with Non-windowed Install Contributions. While the industry has yet to define any standards around the value of these additional contributions, it goes without saying that the length of the attribution window has a significant impact on how advertising partners are actually attributed with these contributions and thus the credit they receive for them.

Determining Attribution Lengths

As an advertiser buying on a CPI (Cost Per Install) or CPA (Cost Per Action) basis, you generally agree and are bound to pay the advertising partner based on installs for a fixed or negotiated attribution length. If you buy on a CPM (Cost Per Thousand) or CPC (Cost Per Click), the advertising partner may still measure effectiveness of their campaigns based on installs using a fixed or negotiated attribution length.

If you do not send data to the advertising partner based on the agreed attribution length, then you may be in breach of your agreement with them. This could be considered fraudulent activity as well.

Moreover, if you opt to set the attribution length to a timeframe different than the one previously agreed upon between you and the advertising partner, the number of installs you attribute to the advertising partner (the ones we attribute for you in MobileAppTracking) versus the number of installs the advertising partner claims will be different. You will then have to deal with the discrepancy by sending the data to the advertising partner for analysis.

Therefore it is extremely important that the attribution for your advertising partner in MobileAppTracking as set to the agreed upon length. If you want to change the set attribution length in the platform, you MUST first negotiate the length with your advertising partner. Note: The attribution length can not be changed with the likes of Google Adwords, Chartboost, and a few others.

Setting Attribution Length in MobileAppTracking

When a unique identifier is passed into a MobileAppTracking tracking link on click, attribution can be completed by matching the unique identifier on click to the unique identifier on install; this is called Unique Identifier Matching. When attribution is done by Unique Identifier Matching we allow the attribution length to be set as the Device Attribution Length.

The Device Attribution Length can be set for a specific advertising partner. This attribution length will take precedent over the account device attribution length preference. Learn how to set the Device Attribution Length for your account or advertising partner.

If you have agreed with one of our integrated advertising partner on a shorter Device Attribution Length, then it can be set accordingly in MobileAppTracking by publisher.

Conclusion

You can analyze your advertising partners’ total effectiveness and compare them to each other accurately by using a normalized attribution length. When analyzing normalized installs and determining attribution lengths, remember the following:

  1. Calculate Total Contributions which is the sum of Attributed Installs, Install Assists, and Non-windowed Install Contributions.
  2. Calculate Install (Conversion) Rates and CPI (Cost per Install) using Total Contributions which will allow you to compare apples to apples within a 30 day period.

  3. If you prefer to measure installs at a shorter attribution length, then check with the advertising partner to see if it can be negotiated. Only some integrated advertising partners can attribute on a custom length. Please do not short-change your advertising partner by setting a shorter length that is not negotiated and/or agreed upon.
  4. You can setup custom attribution lengths in MobileAppTracking based on what is agreed upon with the advertising partner. The default attribution length for each of our advertising partners is already set in the platform.
Author
Lucas Brown

Lucas and his twin brother Lee began their first businesses before starting high school in their hometown of Elma, WA. They quickly found this world suited them, launching other successful businesses in college and reporting their first million dollars in revenue as sophomores in their dorms. Developing unique solutions for their own ad network, Lucas had the foresight to white-label their technologies to empower other businesses, which quickly became the birth of TUNE. As Chief Product Officer, Lucas leads agile product teams that strive to quickly meet the needs of an ever growing client base. As a founder, Lucas and his twin brother Lee supported the company in the beginning, growing to profitability by 2010. Lucas' expertise in complex business models, integrations, implementations, and his constant drive for innovation make him a cornerstone of this Seattle based startup. Lucas graduated from Babson College in Boston, MA.