2022-11-15 | Article | Insights
Until a few years ago, working with Google Ads (GAds, formerly Google AdWords) was manual, fragmented, and time-consuming. A good campaign manager had to be experienced with Excel in particular. Much has changed with the automation capabilities that have emerged since then and the greater integration of individual Google tools. Google Ads now takes a lot of work off the advertiser's hands at the operational level. On the upside, the requirements have become all the greater at the strategic level in order to make the best possible use of the new opportunities. Now, it is particularly important to make optimal strategic use of the existing automation potential for your company. In addition to the obvious and meanwhile well known automations that Google Ads offers, such as automated bidding strategies, automations can also be created by agencies and advertisers. This can be done, for example, at the interface between different Google tools to create holistic opportunities for advertisers. One example are advertiser-specific conversion probability predictions (CPP), which can be created using aggregated customer data, Google Analytics, and BigQuery and automated for use in Google Ads campaigns.
To set up a conversion probability prediction model, the CRoss Industry Standard Process for Data Mining (CRISP-DM) can be used and extended by an activation phase. This way, such a project encompasses seven main steps.
01 Within the first step, a business understanding is created while conducting analysis of the website structure and the different user groups visiting the website and their interests and behaviours.
02 In the second phase of the project, a data understanding is created by extracting Google Analytics raw data via the BigQuery console. In the Vertex AI Workbench, you can then conduct data cleaning and begin with the identifying of patterns with the help of an exploratory data analysis.
03 Then, the data needs to be prepared in a suitable form for the model which involves cleaning missing values, feature engineering, and feature selection.
04 While aiming to predict the probability of a website user to convert, a classification model approach is needed. There are several options ranging from logistic regression to tree based machine learning algorithms.
05 For evaluation purposes of the model performance, unseen test data must be passed measuring the outcome with a confusion matrix to evaluate how often the model classifies a user correctly and which labels are most often confused for that label in this test data set where the outcome is known.
06 Vertex AI pipelines can be used to manage and monitor the deployment workflow which consists of several components such as generating a daily session- or user level BigQuery dataset or a data import from BigQuery to Google Analytics to build segments and audiences for Google Ads.
07 In the final step of the process, the data is prepared for activation in Google Ads. For this purpose, the data is imported into Google Analytics 360 via "Query-time" import. Please note that this feature is currently not available for user-scoped custom dimensions in GA4. In a nutshell, query-time import is a temporary mapping and can be applied to historical data. This ensures that the conversion probability value/group can be stored in a custom dimension and immediately mapped to an existing client identifier. This can be used for reporting purposes and also to build the final audience groups for activation.
There are several use case possibilities to activate the audiences built based on the conversion probability of the website visitors. For example, users with low conversion probability can be excluded from remarketing campaigns with display ads to retarget only valuable users/ potential customers and thus increase the efficiency of your campaigns. If you prefer to target all users instead, you can evaluate options for using different ads based on the conversion probability of a user segment. To this end, you can perform another analysis aimed at understanding which creative strategy works how well with which user group, based on different conversion probabilities, by segmenting your audience by different thresholds.
Besides the remarketing approaches, you can use your highest CPP-audiences to target similar segments on the Display Network, the Search Network, YouTube, Gmail, or in apps. Similar segments is a targeting feature in Google Ads that uses your data segments to help you expand your reach to new potential customers who have characteristics similar to your existing customers or people who have visited your website - in this case those with a high conversion probability. They are automatically created and updated in real time once you have set up at least one eligible list in the Google Ads Audience Center library and allow you to generate more reach targeted efficiently for conversions.
Furthermore, you can explore Video action campaigns as an alternative to display remarketing and search campaigns to activate your CPP-based audiences. Video for Action is a great way to create a powerful storytelling campaign with video or present products or services that require more explanation than a display or text ad can offer while being still optimized for conversions offering an efficient way to reach customers in high-performing places on and off YouTube - all within one campaign. By combining inventory from the YouTube home feed, YouTube watch pages and Google video partners, and more, video action campaigns can help you drive improved performance for your campaigns by reaching also similar customers to your highest CPP-audiences.
Much has changed in Google Ads over the years - the tool has evolved from a search marketing tool to a sophisticated and holistic media activation platform comparable to a demand-side platform with much lower barriers to entry allowing for complex and ambitious scenarios such as using custom conversion probability audiences with various formats across different stages of the customer journey funnel. Now more than ever, agencies and advertisers are required so set clear goals, develop long-term activation strategies, and make informed decisions about how to leverage the power of the tool to make it work for them.