As the sunset of third-party cookies looms large on the tech horizon, marketers have been looking for alternatives to fill in the gaps. One such alternative involves deploying machine learning into analytics reporting, as seen in Google Analytics modeled conversions.
The modeled conversions feature uses machine learning to estimate conversions from visitor activity. Estimating conversions gives a clearer picture on how your website or app is engaging customers while protecting user privacy and avoiding measurement limitations due to technical restrictions.
How Modeled Conversions Works
Modeled conversions works based on attribution categories in Google Analytics 4 referral traffic. An attributed visit is a visit in which a conversion was “observed” — Google Analytics was able to assign a traffic source to the data. Non-attributed visits are people who come to your website through a browser who have opted out of third-party data measurement.
Modeled conversions examines the data from non-attributed traffic and estimates the likelihood of those visitors to convert. The model aggregates attributed and non-attributed visits, predicting likely impact to campaign metrics — and consequently sales conversions — according to a given report.
To ensure prediction quality, modeled conversions are only reported when there is high confidence that enough traffic exists. This is standard practice as robust machine learning models train on a volume of data to establish prediction accuracy.
Since July, Google Analytics 4 properties have included cross-channel modeled conversions. Historical GA4 data “won’t be impacted,” according to Google. Modeled conversion estimates appear in the core reports, such as Events, Conversions and Attributions. The data is also included as an event-scoped dimension in Explorations.
Where Modeled Conversions Can Influence Marketing Strategy
If modeled conversions sound like an analytic response to the move away from third party cookies, you have the right idea. Modeled conversions address the issue of browsers and technologies that limit measurement based on user preferences. Apple’s well-publicized App Tracking Transparency (ATT) protocol, which I explain in this post, is one of many instances where people can choose to block digital ads. Analytic tools must therefore estimate the value of ad traffic without violating user consent or restrictions from these technologies. Modeled conversions in Google Analytics can help marketers with that estimate. To comply with ATT protocol, for example, Google Analytics will model conversion of ads interaction from ATT-influenced traffic sources.
The machine learning prediction in modeled conversions holds tremendous possibilities for supporting B2B campaigns. Campaign attribution for B2B firms has always been complicated because a sale can take months to close. Sales engagements in these cases introduce multiple touchpoints along the way to win a customer. Google is betting such an environment, with multiple touchpoints, will provide its data-driven attribution with enough data to make accurate estimates. Marketers can use this information to identify which campaigns are leading to increased sales conversions.
Modeled conversions also raise an important point in terms of privacy. Marketers need to be more aware of the privacy policies of the different platforms that appear in referral traffic reporting to ensure personal identifiable information (PII) isn’t being introduced into the data for analysis erroneously. In this case, Google is introducing a protocol that structures what gets modeled or not, to align with customer consent.
Modeled conversions is part of Google’s broader investment into data-driven attribution measurement. Google recently announced its Google Ads would also adopt data-driven attribution as its default setting for conversion measurements.
All of these changes come after Google postponed its phase out of third party cookies and its introduction of FLoC until 2023. Marketers got a little breathing room with that new timeline to evaluate how effective their measurement is and to look for alternatives to the cookies similar to what modeled conversions provide.
Pierre DeBois is the founder of Zimana, a small business digital analytics consultancy. He reviews data from web analytics and social media dashboard solutions, then provides recommendations and web development action that improves marketing strategy and business profitability.