The Future of Web Tracking

One of Apple’s latest iPhone updates, iOS 14.5 — released in April — added a revolutionary AppTrackingTransparency framework that has the potential to transform the digital advertising industry. This framework was introduced to enhance privacy in web browsing. This move by Apple enables users to choose whether they want web applications to track their data and browser history, across applications and websites. The news came as a shock to social networking service providers, who are expecting nothing short of a major setback as a result of this shift.

Google Replaces Third-party Cookies with New Technologies

Today, the major players that leverage AdTech are Google and Facebook. However, Google did not object to Apple’s latest privacy feature. Rather, this multinational tech giant is taking the necessary steps to develop a new web tracking system altogether, for Chrome. They intend to avoid privacy risks by blocking third-party cookies. These are the cookies set by third-party websites on the page that the user visits.

For instance, if a company uses Google or Facebook ads or if they want a like/share button on their site, the company would need to embed code on their site. This code would also store cookies, making it a third-party cookie. This allows them to gain access to Facebook or Google.These cookies gather information regarding the user’s habits, search history, etc. This data is then used to serve the user with advertisements. Google Ads also follows a similar process and is the leader in digital advertising.

Source: ClearCode

Interestingly, Google won’t be replacing existing third-party cookies with alternative identifiers for advertisements<(IDFAs). Google is in the process of developing multiple privacy-preserving and open standard mechanisms/ technologies such as Privacy Sandbox and Federated Learning of Cohorts (FLoC) instead.

Migrating to Federated Learning of Cohorts (FLoC) and Turtledove

FLoC is a proposed browser standard that will enable advertisers to target users based on their interests while protecting their privacy and personal information. However Google claims, FLoC will protect the user’s identity, from advertisers, by assigning them to a cohort, which is simply a group of users.

Social media tycoons like Facebook and Twitter may also migrate to a similar model. With the help of illustrations, we will discuss various methods and architectures designed for web tracking.

Traditional Architecture

Traditional Machine Learning employs data pipelines that use a central server to process data. In this architecture, local devices gather data and the sensors route this data to a central server (on-premise or cloud) that also hosts a trained model. The trained model processes the data and produces predictions. Which is then delivered to the devices.

Traditional architecture (Source: Google Cloud Tech)

Traditional architecture (Source: Google Cloud Tech)

Now, the disadvantage of this architecture is that it hampers the ability of the model to learn in real-time. The back and forth communication can also affect user experience by causing network latency/ connectivity issues, affecting battery life, etc.

Disadvantages of the traditional architecture (Source: Google Cloud Tech)

Disadvantages of the traditional architecture (Source: Google Cloud Tech)

Locally Trained Client Model Architecture

In this architecture, a client model is independently trained with the help of client data on the device itself. However, there isn’t enough data on a client device to help train its own model.

Locally trained client model (Source: Google Cloud Tech)

Locally trained client model (Source: Google Cloud Tech)

Collaborative and Decentralized Learning with FLoC

The core principle of FLoC is collaborative and decentralized learning. And in this method, the user’s data is never sent to the central server. How does this work?

This architecture includes an initial model that is distributed to the clients. Here, the clients are chosen based on their activity and other key points. Followed by which, each chosen client trains the initial model locally using client data, within the device itself, and produces a local model. This locally trained model is then sent to the central server. However, in this method, the local data that is used to train the model does not egress from the device. Instead only weights, biases and other parameters are sent to the central server.

Master/ combined model in FLoC (Source: Google Cloud Tech)

Master/ combined model in FLoC (Source: Google Cloud Tech)

The central server, then, creates a combined/ master model with the help of all local models produced by each client. This is not a one-time process, rather it is a continuous loop, where the master model becomes the initial model again, produces a local model, so on and so forth.

FLoC Architecture (Source: Google Cloud Tech)

FLoC Architecture (Source: Google Cloud Tech)

Now, the goal of FLoC is to preserve interest-based advertising, keeping user privacy intact. To achieve this, the FLoC API hashes users’ browsing history and assigns them to a particular cohort. To ensure privacy in this process, each cohort is given an ID, which is paired with aTurtledoveAPI to generate more personalized ad algorithms.

Core Tenets of Turtledove
  1. The browser shall gather behavioral data.
  2. Advertisers shall use this data to target users, but shall not combine it with any data collected while serving ads to the user.
  3. Ad networks shall not store data regarding users’ interests.

The Future of Digital Advertising with FLoC

By proposing to implement Federated Learning of Cohorts (FLoC), Google is essentially suggesting that they will not track users’ data. They will only use patterns to create cohorts while the client data remains on the device itself. This is a more optimised way of analyzing user data by seeking out signals/patterns without compromising their privacy.

Google is moving towards a privacy-focused approach with regards to web tracking. However noble this might sound, this shift is obviously set to benefit the tech giant. Users are extremely concerned about their online privacy. It’s no surprise that privacy-conscious browsers like Firefox, Safari, and Brave are gaining popularity. Therefore, unless Google tackles this issue head-on, they could be taken out as the leader in the browser market as well as the advertisement industry.

While FLoC allows websites to acquire user data without affecting their privacy, it also establishes a strong foothold in the future of digital advertising. Google has made certain that only a few ad networks would be able to develop tools to mine user data while maintaining privacy. They’re also rendering standard online user tracking methods obsolete.