Maximize ad relevance after third-party cookies

The Privacy Sandbox aims to keep people's activity private across an open and free internet. To do this, we are collaborating with the ad industry to transition to new private ad technologies and deprecating support for third-party cookies in Chrome in the second half of 2024.

Whether you're a product leader, CTO, CMO, or CEO, it's essential to understand how to support advertising use cases in an increasingly privacy-conscious world and embrace solutions that can optimize both business outcomes and user privacy.

There is no doubt that 2023 will be a critical year to prepare for a world without third-party cookies. In this guide, we'll discuss how the advertising ecosystem can approach ad relevance in a cookieless future:

• What is changing about the data used to show relevant ads?
• How might ad tech offer interest-based advertising without third-party cookies?
• How can machine learning maximize performance using privacy-safe signals?

# What's changing about the data used to show relevant ads?

Interest-based advertising, also known as personalized advertising, is a type of advertising that uses information about an individual's interests and preferences to show them more relevant ads. This type of advertising uses a wide range of data as signals to determine what ad to show, such as what content a user has viewed, the pattern of sites the user has recently seen, or a specific site previously visited.

Today, these signals are primarily powered by cross-site identifiers such as third-party cookies, which are unique to an individual device.

As third-party cookies are phased out, ad tech solutions for interest-based advertising should evolve to take advantage of privacy-friendly signals to show relevant ads. These include first-party data, contextual signals, and platform-provided privacy-preserving APIs, such as the Topics API, FLEDGE API, and Attribution Reporting API, which help support critical use cases for the ad industry while protecting against cross-site tracking.

Interest-based advertising can survive and thrive with innovative technologies developed across the ecosystem. These technologies will help us move toward a world where people's data is better protected, while ads can continue to drive marketing outcomes that support a dynamic and open web.

# How can ad tech offer interest-based advertising without third-party cookies?

Today, interest-based advertisers typically set up the following for campaigns through their ad tech providers:

1. Goal: What is the business outcome the advertiser is trying to achieve with this ad campaign? This tells ad platforms what to optimize for. For example, the advertiser may want to drive sales on their kids' clothing website. Often, these goals are measured through cross-site conversion tags and Attribution Reporting.
2. Audience: Who does the advertiser want to reach? This tells the ad platform who the advertiser thinks is likely a good match for the ad. For example, the advertiser may wish to reach new customers currently in-market for kids' clothing.
3. Placement: What websites does the advertiser want to run ads on? This specifies where ads are allowed to run in terms of ad inventory or categories of ad inventory. For example, advertisers could place their ads across a broad set of websites, or they might select specific websites which have higher likelihood of reaching their desired audience.
4. Budget and bid: How much does the advertiser want to spend in total and for a specific action like impression served, ad click, or ad conversion? This ensures the campaign meets cost requirements for its goals. For example, the advertiser may wish to spend up to $1,000 and pay at most $2.00 CPM to serve 500,000 impressions to targeted audiences and on specified websites.

# Evolving audience creation

In a post-third-party cookie world, ad tech providers will want to adapt how their platforms serve relevant ads. Today, ad relevance is commonly achieved through audiences that an advertiser can use to reach people most likely to be interested in their products and services.

Advertisers commonly use these different audience types:

1. Affinity: Reach users based on what they're passionate about, their habits and their interests.
2. In-market: Reach users based on their recent purchase intent.
3. Remarketing: Reach users who have previously visited an advertiser's website.
4. Audience extension: Reach users of a particular publisher on other websites

After third-party cookie deprecation, ad tech providers can continue supporting these audience types' goals using new approaches, including Privacy Sandbox APIs.

# Affinity

Today, advertisers reach users classified by their affinity (also known as interests), most commonly by leveraging third-party data segments. These audiences are provided by many data marketplaces and distributed for activation across the ad tech ecosystem through channels like demand-side platforms (DSPs) and data management platforms (DMPs).

Segments are typically built by tracking individuals using third-party cookies and then grouping users based on a taxonomy of categories and proprietary methodologies for determining when a user qualifies for a category.

After third-party cookie deprecation, audience selection based on affinity will evolve to use different signals to qualify users for inclusion in any given audience. There will be several ways to do this using the Privacy Sandbox privacy-preserving APIs including:

• Topics API: This API offers a standardized taxonomy of interests and a publicly-known methodology for the on-device classification of interests for a given user based on the types of websites recently visited. Ad tech can call the Topics API to get interests for a given user. The API protects privacy by limiting: the length of browsing history considered, the parties who can access a given topic, the number of categories returned, and more. This API is particularly useful for ad tech without direct publisher relationships or contextual optimization capabilities.
• Topics API with contextual data: A more advanced method involves comparing a user's topics and the context of a page, to estimate additional affinities for users. For example, an ad tech solution may learn that people interested in a particular set of topics (such as outdoor activities) may over-index on visiting specific categories of pages (such as sites about grilling). Ad tech can train a machine learning model to predict that a visitor to an "outdoor activity" website could be interested in grilling even if "BBQ & Grilling" is not returned as a topic through the Topics API. This method is particularly useful for a buy-side ad tech if it has contextual optimization capabilities.
• FLEDGE API: This API enables ad tech to create audience segments by labeling visitors of a web page as members of a particular segment, such as "interested in family adventures." If the ad tech provider has other websites in its partner network that pertain to "family adventures," they can also add visitors for those sites to this same segment.

FLEDGE protects user privacy by keeping assignment to audience segments on-device, and not sharing back to ad tech whether the same user belongs to multiple interest groups. This limits cross-site tracking. This API is particularly useful for an ad tech with a network of site partnerships.

With these methods, ad tech can offer scaled affinity audience segments without relying on cross-site user identifiers. Ad tech don't have to limit themselves to one method, and may differentiate based on their publisher relationships, advertiser relationships, and machine learning capabilities.

# In-market

Presently, advertisers reach users classified as being "in-market" (also known as having "purchase intent") by using third-party cookie segments similar to how they access audiences based on "affinity." Whether a user is classified in-market for a product like "cooking gear" or simply interested in cooking depends on the proprietary taxonomies and methodologies of ad tech providers.

After third-party cookie deprecation, privacy-preserving APIs will provide new signals to inform 'in-market' audience creation. Some alternative methods include:

• Topics API: Similar to using this API for affinity audiences, using it for in-market involves returning a topic that can approximate purchase intent for a given user based on an on-device, publicly known methodology, and taxonomy.

The standardized three-week lookback window for generating these topics protects user privacy by limiting the total amount of data made available to ad tech providers. However, different categories of products and services have different consideration cycles ranging from days to months, making this API useful for advertisers whose customer purchase cycle aligns with the Topic's lookback window.

• FLEDGE API: As with the affinity use case, this API gives ad tech platforms the ability to create their own segments, such as "in-market auto buyers." If the ad tech provider has other websites in its partner network that pertain to "in-market auto buyers," they can also add visitors for those sites to this same segment while maintaining cross-site user privacy. FLEDGE is particularly useful for an ad tech provider when there is a direct publisher/advertiser relationship that allows data partnership, and a need for greater customization than Topics would allow.
• Topics API + Attribution Reporting API: By combining Topics and the Attribution Reporting API, you can expand the lists of topics that map to specific conversions (such as purchases), which creates additional ways to reach an in-market audience.

For example, analysis or machine learning systems may uncover that users who saw an ad about scuba gear, and bought it, very often have "Beaches & Islands" and "Fishing" topics associated with them. An ad tech solution could translate this insight into improved reach to users "in-market for scuba gear" by selecting users with those two topics. Attribution Reporting protects user privacy in this instance by providing noisy conversion data about associations of topics with conversions.

This approach makes sense when ad tech providers don't have much contextual data but have machine learning or robust data science and analysis capabilities.

• Attribution Reporting API with contextual data: Ad tech solutions can leverage contextual categorization of the pages where ads are shown, categorization of advertisers and products, and data from Attribution Reporting to uncover trends or patterns in the types of sites people favor when in-market to buy certain types of products and services. For example, this combination of data may lead to insights like learning that people who are on web pages about family activities are highly likely to be in-market to buy outdoor apparel too.

These methods are just a few of the many ways ad tech providers can creatively scale and customize audience segments without relying on cross-site user identifiers. They could also integrate more signals like first party data and other combinations of privacy-preserving APIs for even greater results. Thus, ad tech providers can differentiate themselves by taking different approaches to audience building, securing unique data, and developing superior machine learning capabilities.

# Remarketing

Advertisers can re-engage users who have previously visited their website through remarketing. Currently, this involves placing a third-party cookie on a browser at the time of a website visit and then bidding to show ads to that browser when the cookie is observed on another website. Ad tech providers can create different remarketing segments for a given website based on user activities taken throughout the website.

Without third-party cookies, ad tech providers will be able to use the FLEDGE API to support remarketing use cases:

• FLEDGE API: Ad tech providers can create customized remarketing segments for a site by creating interest groups dependent on user activity. In prior use cases with FLEDGE, ad tech providers were building very large audiences from multiple websites. In this use case, only one website is trying to re-engage a past visitor, and without the privacy protections built into FLEDGE, this use case might lead websites to single out individuals. While allowing effective audience remarketing, this API protects individual privacy by setting k-anonymity thresholds to ensure a sufficient number of individuals are eligible to see the ad.

Even without third-party cookies, the Privacy Sandbox enables advertisers to use their first-party data for remarketing at scale, across third-party websites.

# Audience extension

Advertisers sometimes want to reach more of the same audience they see from a particular publisher, but when those users are on other websites. Audience extension is a process that extends publisher first-party audiences by finding them on other sites to increase frequency or delivered reach of the same audience.

By using audience extension, a publisher can provide an advertiser with an audience segment, such as affinity (such as golfers) or demographics (such as an age range), and allow the advertiser to find that audience on other sites. Audience extension is also used when an advertiser wants to increase awareness of their products by reaching consumers when they shop on a retailer's website and elsewhere on the web.

Ad tech providers can use the FLEDGE API to extend audiences for publishers without third-party cookies:

• FLEDGE API: Ad tech providers can create custom audience extension segments for a site by creating interest groups dependent on user activity such as reading a particular section of a website (e.g. travel section). This process is effectively similar to remarketing and offers the same privacy protections. It makes sense for advertisers who value the 1P audience data of a publisher but cannot get enough ad inventory on that publisher website for that audience.

# How can machine learning maximize performance using privacy-safe signals?

With the deprecation of third-party cookies, advertisers may want to consider how machine learning and privacy-safe signals can be used to drive the best outcomes.

# Drive advertiser outcomes through automation

Most ad tech providers offer varying degrees of manual and automated campaign optimization.

The most manual solutions require advertisers to specify desired audiences, placements, and bids, and then stay within those inputs. Manual setups provide robust control to advertisers but may provide sub-optimal results if the advertiser needs to know all the performant audiences and placements or cannot predict the theoretical optimal bid for each impression, given all the variables involved.

The most automated solutions ask advertisers to specify their desired business outcome (such as $2 cost-per-action/sale), using machine learning to identify the audiences and placements that perform well for that advertiser and the right bid to achieve the desired goal. In this setup, there are few or no constraints on the ad tech solution except a budget and goal. Audience selection by the advertiser may be treated as a "suggestion" or "starting point," but machine learning will look for patterns among all available data that may be indiscernible to humans.

Machine learning uses these patterns to optimize performance by adding more relevant audiences and adjusting bids based on the predicted performance of those audiences. The Privacy Sandbox is one of many sources of signals that will be available to inform machine learning after third-party cookie deprecation.

Machine learning can maximize ad performance by continually testing and learning all the best audiences, placements, and bids across time, campaigns, and even advertisers. It's worth noting that sophisticated analytics performed by skilled teams can also discover similar correlations.

Reducing the need for advertisers to manage audiences, placements, and bids will simplify advertiser workload and enable machine learning systems to drive the best possible outcomes. Ad tech investments in automated solutions—in addition to benefiting advertisers—can also help transition away from third-party cookies.

# Additional signals for machine learning

Ad tech providers have always factored in multiple signals when deciding whether to bid to serve an ad. In a world without cross-site cookie tracking, ad tech will benefit from using every available privacy-safe signal in machine learning to predict business outcomes, such as clicks or conversions.

The following privacy-safe signals are sometimes undervalued but can contribute significantly to ad relevance in a future without third-party cookies:

1. Ad creative features: Analyzing ad creative at a component level (e.g., text, images, design) may help predict performance with specific audiences or on certain pages, such as the ad's subject matter or whether it includes a lot of text.
2. First-party data: Publishers, marketers, and retail networks are increasingly building first-party identifiers and segments, such as seller-defined audiences. Knowing a user's behavior over time on a given site allows you to better predict what ads work best for that user or segment on that site, without cross-site profiling. A publisher's first-party data can help improve bidding across all their sites. These site-specific bid improvements can cumulatively increase performance across a campaign.

Ad tech providers can unlock the best results by combining all available tools, such as machine learning and privacy-safe signals from privacy-preserving APIs, along with contextual data, creative data, and first-party data.

# Conclusion

After third-party cookies are phased out, it is essential that the advertising industry continues to deliver relevant ads and that consumers receive the privacy protections they expect. We know that building with new tools, like those from the Privacy Sandbox, requires effort, and we will continue to support the industry throughout this transition.

Moving forward, we encourage you to:

1. Invest in integrating privacy-preserving APIs such as Topics, FLEDGE, and Attribution Reporting into your ad tech solutions, to support common interest-based advertising use cases after third-party cookies go away.
2. Test Privacy Sandbox APIs in conjunction with other privacy-safe signals, including first-party publisher data, to understand future performance and inform strategy.
3. Maximize performance by enabling machine learning to use all available privacy-safe data, with as much freedom to learn and optimize as possible.

The ad tech industry can perform many core targeting and bidding functions using the Privacy Sandbox APIs. However, there are numerous benefits from incorporating additional privacy-preserving signals beyond those APIs, and deploying all of these signals together.

Innovation is in the digital advertising industry's DNA. By evolving existing approaches to ads relevance, we can successfully transition from third-party cookies to a more private and more performant web.