Privacy Playbook: Ensure accurate measurements

Chapter 2. A pregnant Black woman presents charts and graphs. In this chapter, you’ll learn how to: 1. Measure conversions more accurately with first-party data. 2. Rely on machine learning to fill gaps in the customer journey and unlock insights

You are reading part 2 of a 3-part privacy playbook. Jump to section 3. Drive performance by keeping your ads relevant or 1. Build direct relationships with your customers.

Illustration of two people holding a key. Insight: People want increased control over how their data is used.

Measure conversions more accurately with first-party data

Once a first-party data foundation from practices like sitewide tagging has been established, you can enable your measurement solutions to work together and provide comprehensive reporting.

For example, enhanced conversions for web allows your site tags to access first-party data to fill in measurement gaps, delivering a more accurate view of how people convert after engaging with your ads.

^{Source: Google Internal Data, 2021.}

Here’s how it works

Hashed data, or user-provided data from your website, is shared with Google, which then matches it to signed-in Google Accounts, enabling conversions on Search and YouTube to be attributed to ads without compromising user privacy.

This, in turn, provides more observable data, allowing you to strengthen conversion modeling, measure Conversion Lift, and optimize campaigns with Smart Bidding.

Case Study: Enhanced conversions for web helps Tennis Express serve up privacy-safe growth. A smartphone displays images of tennis-related sportswear showing the Paris Collection from Primeblue.

Approach

Tennis Express, a U.S.-based sports apparel retailer, wanted to grow its first-party data and started by implementing a solid tagging strategy. It began by identifying untagged site URLs and grew sitewide tagging adoption to over 98% of pages using Google Tag Manager. Then, it activated enhanced conversions for web to increase the accuracy of conversion tracking and fuel Search campaign performance.

Results

The percentage of conversions that came from first-party, rather than third-party, cookies increased to 89%. This strong tagging foundation and activation of enhanced conversions for web also increased conversions across Search campaigns by 114% year over year.

Case Study: MandM Direct combines first-party data with machine learning to see up to 20% more conversions. A smartphone shows a child in winter gear and rain boots and an adult in athletic clothes and shoes.

Approach

Online clothing retailer MandM Direct wanted to explore privacy-safe techniques for preserving campaign measurement. It started by implementing sitewide tagging on its website and adopting Google’s new analytics platform, Google Analytics 4. Next, it strengthened the accuracy and scope of these tools by adopting enhanced conversions for web and using Consent Mode so its site tags could rely on conversion modeling when users opted out of cookies.

Results

Implementing enhanced conversions for web helped MandM report 3% and 20% more conversions on Search and YouTube respectively — plus a 15% increase on top of that by using Consent Mode.

Tag customization to increase measurement accuracy

The accuracy of conversion measurement can also be improved with Consent Mode,* which enables the customization of how tags behave based on user consent choices.

_{*Consent Mode is applicable to advertisers operating in the European Economic Area and the United Kingdom.}

Here’s how it works

On websites using consent banners, users can customize which cookies they will allow. These choices have an impact on how much data you can measure in your Google Ads account. Consent Mode addresses this challenge by communicating the consent status of a user to Google Ads and improving the accuracy of your data through reporting and modeling.

For example, when users don’t consent to cookies, Consent Mode will use conversion modeling instead to fill gaps that can’t be directly linked to ad interactions.

On average, conversion modeling can recover more than 70% of ad-click-to-conversion journeys, so you can still accurately measure the performance of your media in a privacy-safe way.⁵

Approach

As more people turned to online shopping and home delivery for their essentials, e-commerce company Nemlig saw a large increase in visitors to their site, resulting in a longer page-load time. The company also operates in the European Economic Area, where website tags must adjust according to users’ cookie consent choices.

The team worked with their Google Marketing Platform partner IIH Nordic and decided to move to server-side tagging to help them accurately report on customer insights from their website while maintaining a speedy shopping experience. And because server-side tagging natively supports Consent Mode, Google tags in Nemlig’s server container automatically updated to respect the consent choices from users.

Results

Nemlig saw a site-speed improvement of 7% and observed 40% more 90-day conversions for new customers than before. In addition, Nemlig’s analytics are now far more trustworthy, with online reported orders reflecting registered orders on Nemlig’s back-end system more accurately.

Learn more about Google Marketing Platform

Rely on machine learning to drive more accurate results and unlock insights

Even if you’ve built a strong foundation of first-party data, gaps in the customer journey might occur. For example, varying consent choices come into play when people move across devices, use different browsers with different restrictions, or move from online to offline touchpoints.

Machine learning can be used to fill those measurement gaps and improve your marketing. Conversion modeling, for example, has been and will continue to be a key feature in Google’s measurement solutions. Modeling uses observable signals to provide a more complete picture of your performance in a privacy-safe way. It’s important to note that because each advertiser’s customer set comes with its own unique behaviors, there isn’t a one-size-fits-all model. That’s why, whenever possible, we integrate conversion modeling directly into Google Ads products so the modeled data — including install, in-app action, and conversion value — will appear in your conversions-reporting column. This gives you insight into conversions that otherwise would not have been recorded, for instance, due to platform restrictions limiting the use of third-party cookies or other identifiers.

Here’s how conversion modeling works for each advertiser

Step 1

We separate ad interactions into two groups: one with observable links between an ad interaction and conversion, and the other where that link cannot be observed.

In Observed interactions, gold “ad interaction” squares are connected in rows by chain links to silver “conversion” squares. In Unobserved interactions, the gold and silver squares appear scattered in a random pattern.

Step 2

We divide the observed group into subgroups that share nonsensitive characteristics like device type, browser, country, conversion type, and so forth.

Linked gold “ad interaction” and silver “conversion” squares are gathered into subgroups by nonsensitive category type.

Step 3

Within each of these subgroups, we calculate conversion rates.

Step 4

Next, we take the ad interactions and conversions from the unobserved group and assign them to one of the existing subgroups based on shared characteristics.

For example, subgroup 1 may all be in France, use Chrome as their browser, and are on an iPhone. We see in the unobserved group similar characteristics in the ad interaction and conversion data, except for one of these dimensions that is the one we are trying to predict against (for example, browser type). So, we align these groups appropriately based on their similarities.

Unobserved interactions, represented by a random array of gold “ad interaction” squares and silver “conversion” squares are moved into a column of Subgroups labeled Observed.

Step 5

Using the known conversion rates from the observed population, in conjunction with machine learning, we can model which unlinked ad interactions belong to which unlinked conversions.

Note: Our data scientists continuously make improvements to the algorithm to ensure accuracy and account for scale. Additionally, we proactively test and validate models using techniques (like holdback validation) to improve accuracy. This allows us to regularly measure biases and inaccuracies as we continue to tune our models.

A column of linked gold and silver squares labeled Observed on the right help a machine learning model in the middle link the random Unobserved squares become joined by links in a column on the left.

Step 6

Once the ad interactions and conversions have the appropriate links between them, we aggregate them and share them in your reporting. We only include modeled data in reporting when we have a high degree of confidence that conversions actually occurred as a result of ad interactions. This rigor ensures we avoid systematically overreporting, so you can confidently work with the reported data to improve your own results.

All the linked gold and silver interaction squares are aggregated into a model and converted into reports.

Learn more about how conversion modeling works

Activating privacy-safe measurement solutions, like enhanced conversions for web and Consent Mode, in tandem with modeling improves your digital foundation by capturing more conversion data. And the more data we have to model, the better our models become. Smarter models can help drive better performance by enabling products like Smart Bidding in Google Ads to access more complete information — all with user privacy at the forefront.

In addition to more complete conversion measurement and optimization, modeling can also help you gain new customer insights from your behavioral analytics data. For example, Google Analytics 4 provides a holistic customer view across both web and app, and uses advanced machine learning models to surface customer insights from your first-party data. You can then use those insights to improve your marketing. To get the most out of the tools, be sure to link your Google Analytics and Google Ads accounts so you can maximize the benefits from automation when bidding on Google Analytics conversions in your Google Ads campaigns.

Finally, using data-driven attribution in Google Ads and Google Analytics 4 can take your analysis a step further by using advanced machine learning to determine how each marketing touchpoint contributed to a conversion without compromising user privacy. Like all of Google’s measurement solutions, we respect people’s decisions about how their data should be used in the attribution process, and have strict policies against covert techniques, like fingerprinting, that can compromise user privacy.

To help you take advantage of better attribution in today’s changing privacy landscape, data-driven attribution is now the default attribution model for all new conversion actions in Google Ads.

Learn more about the different types of machine learning models

Discover what the future looks like

The Privacy Sandbox is a collaborative, open-source, industry-wide effort to develop new technologies that support the key advertising use cases businesses rely on, without the need for third-party cookies. Marketers will still be able to get the performance and insights needed to grow without having to track individuals across the internet and on Android apps.

For instance, instead of measuring what people do online in a way that could identify them, marketers can keep them anonymous by putting limits on data when their actions are reported and adding random data, or noise, into the report.

Here’s the proposed solution for measurement

A web browser will match a conversion that happens on an advertiser’s website with an ad that was clicked or viewed on the web. The browser will only report information in a way that doesn’t expose people’s identities — for example, aggregating the data and limiting the amount of information shared about each conversion.

11. 2139_Privacy-Playbook_Chapter_2_Inline

By joining the Privacy Sandbox initiative, Android shares Chrome’s vision of enhancing user privacy while supporting key advertising use cases across the web and mobile app ecosystems.