Blog

Announcements, analysis and opinions on industry trends around the mobile programmatic world.

We are looking for a Designer & Content Creator in Brazil

We are seeking a talented and creative individual, an entrepreneurial and highly driven  Designer & Content Creator to join our team. The Designer &...

Read More


    Fingerprinting and Attribution: Shining a light in the dark

    Tiago Vila Verde
    Posted by Tiago Vila Verde

    Every marketer that runs mobile app advertising campaigns knows that some partners outperform others, and often by a great deal. Partners use different traffic sources, algorithms and media buying strategies that lead to different results. However, as important as these factors are, they depend on a more important element: app attribution settings.


    Most mobile app advertisers work with an app tracking vendor such as Adjust, Appsflyer, Branch or Kochava. These vendors allow their clients to define the attribution criteria for their partners’ media buying activities - and here is the key factor: the criteria established to attribute events will have a strong and direct impact on the performance numbers of the partners, regardless of what traffic sources, algorithms or media buying strategies they use.

     

    THE IMPACT OF ATTRIBUTION SETTINGS

    Let’s think about a concrete case. Imagine there is an advertiser that works with Partner A and Partner B for two months. The first one buys a lot of cheap impressions, with low impact creative formats (such as a 320x50 display ad). The other buys high quality and more expensive inventory with higher engagement ratios due to the type of creatives used: video, native and interstitials. Now let’s say the advertiser sets the attribution rules to one day post view and seven days post click during the first month, and uses only seven day post click (no post view) during the second month. These would be the results:

    Screenshot 2020-02-24 at 17.42.55 

    We can clearly see that in spite of both partners having kept exactly the same campaigns in both months, simply by changing the attribution settings the results are totally different: both of them show a higher CPI in the second month, and while partner A looked more efficient with the attribution settings in the first month, in the second it seems that partner B performs much better.

     

    DIFFERENT TECHNIQUES TO ATTRIBUTE CONVERSIONS

    Now that we have seen the importance of attribution rules, let’s drill down further to understand how tracking tools apply the attribution criteria defined by each advertiser.

    There are different ways of app tracking vendors to perform the matching between an impression or a click with an event tracked (install, lead, purchase, etc.). The most common way is to do the matching using the device ID as a reference: the vendor tracks the device ID that generated the event and looks for a match between that ID and all the IDs associated to impressions and clicks generated by partners. If there is a coincidence, the conversion is attributed to that partner.

    Although the device ID matching model is quite accurate and is used as a first option, there are cases in which it can’t be applied, as that data is not always available. For example, for mobile web traffic – where there is no SDK integration – the device ID cannot be shared by the traffic source (publisher, SSP, ad exchange, etc.). So, in order to avoid losing visibility over the contribution of its advertising partners efforts for this type of traffic, tracking tools allow advertisers to use other methods to match events and campaigns.

     

    FINGERPRINTING

    One of these methods is called fingerprinting: a technique that relies on other kinds of data (such as IP, user agent, OS version, carrier, etc.) to tie a conversion to an impression or click. It’s worth noting that fingerprinting and device ID matching are compatible. Fingerprinting is not as accurate as the device ID matching approach however: different devices share the same characteristics and you may think it is the same user when in reality it is not. So, it should only be used as a fallback, and only in case the device ID is not available. There are specific cases in which Fingerprinting can be especially inaccurate:

     

    1. Fingerprinting relies on the IP and, while a user navigates in a cellular connection as opposed to WiFi, their IP is given by the mobile carrier. Here, on one hand, it can change very often for the same user and on the other it can be shared by thousands of users.
    2. Fingerprinting relies on the user agent, which includes the model and operating system of the device. In cases such as iPhone use, there are not many models out there and often the user agent includes only the fact that it is an iPhone.

     

    In summary, fingerprinting can be a great help if you want to reach a wider audience. However, you should follow campaign results very closely and measure the impact it has on them within your organic users, because it can be very tricky.

     

    EXAMPLE OF HOW MISLEADING FINGERPRINTING CAN BE 

    Let’s imagine an advertiser has activated the fingerprinting option in their attribution tracking tool as a fallback (I confess this is a test we actually ran, but let´s imagine). He is working with a media buying partner to promote its app and that partner only buys traffic with device IDs because he knows results will be most accurate tracked with that setting.

     

    However, the partner is told that results are not competitive in some specific publishers and the advertiser agrees to have the partner perform the following test: Not to send the device ID to the tracking tool, and forcing it to apply fingerprinting. Although the campaign was running exactly the same way two things happened once the device ID stopped being sent:

     

    1 – The CPI decreased 85% - when the device ID was shared the CPI was $10.34 and it decreased to $1.55 during the test.

    2 – The number of clicks required to get an in-app purchase assigned to the campaign decreased 125 fold. That means, with exactly the same budget, you could increase your “results” dramatically.

     

    Even if the partner didn’t do anything different with its campaigns, the fact of not sharing the ID and “benefiting” from the inaccuracy of the fingerprinting method gave the impression that the partner´s results were a lot better, bringing them inline with what the advertiser was seeing elsewhere.

     

    FINDING THE RIGHT OPTION 

    So, how can advertisers protect themselves and find the right attribution settings that indeed represent the real contribution of their advertising partners?

     

    Advertisers need to take action and put different tests in place in order to gain data they can trust. Here is where incremental metrics come into play: they allow you to not only see  the number of conversions “attributed” to a campaign, but the number of those conversions that are additional (incremental) thanks to your campaign - In other words, incremental metrics eliminate from the “results” those that would have happened independently of the campaign. More and more advertisers are starting to focus on incremental eCPI and incremental eCPA in opposition to the traditional eCPI and eCPA. This is significantly reducing the risk of wasting part of their budgets and is increasing effectiveness.

     

    CONCLUSION

    Some advertisers underestimate the importance of defining the right attribution settings. It has a direct impact on the campaign results that marketers use to base their analysis on and make decisions, so they should make sure they pick the right option for them.

    Fingerprinting is an attribution technique that allows advertisers to match in-app events to marketing campaigns for all the traffic that does not have a device ID available. However, it has some important limitations, especially in terms of accuracy.

    Marketers applying this method should take measures to understand the real numbers behind their partners’ marketing efforts – and using incremental metrics is fundamental to making efficient decisions and maximizing results.

     

     

    tiago.vila.verde
     
    Tiago Vila Verde
    Head Of Analytics
    at Mediasmart Mobile

     

    If you want to know more about Mobile Attribution have a look to our Introduction to Mobile Attibution.

     

    Topics: attribution, mediasmart