Attribution has suffered some massive headwinds over the past 3 years, but for some reason – many companies still use attribution as their source of truth, even though often realizing how inaccurate attribution can be.
In the following article, I plan to explain and reveal why using fingerprinting attribution has been far from being an accurate method to measure marketing performance, and why marketers should not trust any fingerprinting reporting.

How Does Attribution Work ?

Attribution is a process to create a link or a match between users’ engagement with ads and a users’ conversion. The goal is to help advertisers identify which ads are effective and which are not, enabling them to optimize their ad spend. 
The most common method of attribution over the past two decades has been last click attribution – an attribution method which assigns 100% credit to the last ad a user engaged with (which in turn means the last publisher and ad platform the user engaged with).

(Mobile attribution process illustration. Source: INCRMNTAL)
 

What is Fingerprinting?

The term fingerprinting  eludes a deterministic user matching based on forensic parameters, however, fingerprinting in attribution is extremely far from being forensic, nor deterministic. 
Fingerprinting attribution tries to match the origin of a user (publisher/ad platform) to the conversion point (app install) by using available data sets to make a best guess. 
Fingerprinting may be used when a publisher does not or cannot access a users’ device ID.

Typically, a fingerprinting match will be based on parameters such as: IP address, device type, battery level (if available), location (if available), and other data available at the point of the ad engagement and app install. 

(Mobile attribution using Fingerprinting mechanics illustration. Source: INCRMNTAL)
 

Fingerprinting accuracy tends to be drastically lower than any other method of attribution, given that none of the parameters used to create a fingerprint match are unique to an individual user.
 

• IP addresses are shared by all users within a network.
• Device location is often not accessible, causing the data to be relative to a region, or even only the country.
• Device type (i.e. “iPhone”) is definitely not unique.
   

Prior to privacy related restrictions and regulations, the process of matching happened using a persistent device identifier. This allowed them to create a deterministic match between clicks and conversions. 
The accuracy of fingerprinting has long dropped, falling below 50%, due to privacy regulations and restrictions imposed on platforms such as iOS and Android, which have reduced access to user level data.

Why Would Anyone Use Fingerprinting at All ?

Despite being explicitly forbidden by Apple, and unofficially “frowned upon” by Google, Fingerprinting was normalized by most attribution providers due to two main reasons: 
 

• Limited (or no) access to the deterministic device identifier
  Access to user level data has been gradually restricted over the last several years. This began with the elimination of the UDID (2012), followed by additional restrictions and limitations such as Limit Ad Tracking (2013), App Tracking Transparency (2020), and the most recent Privacy Manifest (2024). As access to user level data diminishes, attribution platforms were almost forced to come up with alternative tracking methods, often resulting in match rates of less than 50%.
   
• Ad brokering with no access to user data
  An often ignored normality of the mobile ad space was the fact that advertising had often been traded around using media brokers. Brokers did not have access to user level data. This limitation has forced attribution to rely on other available data points (i.e. Fingerprinting)

Fingerprinting continues to lose accuracy as platforms further restrict access to data. However, this hasn’t deterred attribution platforms from offering fingerprinting as a service.


(iOS Attribution method distribution. Source: Kochava)
 

Fingerprinting Was Also Bad for Ad Platforms

Advertising platforms previously relied on attribution postbacks in order to get real time indication of their performance. However, as privacy restrictions eliminated access to deterministic attribution data, ad platforms had to make a difficult choice:

1. Rely exclusively on aggregated campaign level postback for optimization (i.e. SKAdNetwork)
2. Rely exclusively on fingerprinting data, knowing that this is not accurate

Ad platforms with access to first party data, such as Google, Facebook Ads, TikTok, AppLovin, and others – realized that the best option for them was option c: none of the above. 
Rather than relying exclusively on aggregated campaign reporting, or developing a dependency over fingerprinting, knowing that this would be eventually deprecated, ad platforms developed their own ways of modeling performance and generating great results for advertisers. 
We named this method “algorithmic attribution” and have written much on the topic (see part 2).

We’ve also predicted this as “the demand vs. supply shakedown”, though now we are more optimistic about the outcome of this movement. 
 

What Is the Alternative to Fingerprinting?

A marketer is tasked with several responsibilities. Some of these are strategic and do not demand a real time reaction, but some might. Our whitepaper on measurement orchestration, which explores how to harmonize attribution, incrementality and MMM, includes a detailed map of marketers’ tasks, highlighting those that happen in real time. 
 

Fingerprinting is probably the least accurate method of attribution measurement, but its proposed benefit is that it delivers results in real time. An advertiser spending at scale, will want to get real time reporting over creative performance, campaign results, in order to make real time optimization decisions.

Fingerprinting provides this data, however, the deteriorating accuracy means that the data is extremely unreliable. 
 

When it comes to optimization decisions – advertisers should move away from fingerprinting data, to rely on the advertising platforms’ own first party data for real time tasks, using better methods of measurement for strategic decisions. 
While ad platforms data may be ‘biased’ towards the advantage of the platform – for the purpose of tasks which require real time data: Creative A/B testing, Audience A/B testing, placement A/B testing – ‘biased’ data would still serve the purpose.

For Example: 
If an advertiser is looking to A/B test creatives, and one creative has superior performance over another, the level of accuracy in the reporting is secondary to the fact that one creative performs better than another. 
Even with a discrepancy between platform reporting and attribution reporting – directionally – the data is good enough to make a clear decision towards optimization and A/B selection. 
 

The same principle would work for audience, placement, device, and time of day.
 

The Risks and Realities of Fingerprinting Attribution in Advertising

Fingerprinting attribution is not a good alternative to deterministic attribution, as the level of accuracy it provides makes it unreliable. Advertisers who use this as a billing model (i.e. paying a cost per install based on fingerprinting attribution) are headed towards disaster once they realize the discrepancy between what fingerprinting and actual install counts are. 
Transitioning from relying on attribution data to this new paradigm may seem daunting, but it's a change that fortunately needs to be made only once. There is no looking back after that!