Reaching your most valuable audiences: Predicting Conversion Intent

Blog 1 out of 4: capture the behavioral 'rules' that lead to your target

The increase of digital advertising spend, the rise of competition in e-commerce, and the decline of conversion rates, make it important to spend your marketing budget as efficiently as possible. There are many solutions to this, one of them is to focus on targeting your most valuable audiences, for example by predicting their conversion intent. In this and three more blogs, I will describe how to predict which customers are likely to convert, and how to set this up in a successful way in Google Cloud Platform combined with Google Marketing Platform resulting in four posts:

• Blog 1: capture the behavioural ‘rules’ that lead to your target
• Blog 2: feature selection analysis
• Blog 3: training/evaluating the model
• Blog 4: deploying the model

Predicting Conversion Intent

Who would you rather spend your money on: visitors that are likely to convert, or visitors that are just browsing your website. I know what I would do. However, the first question is: how can you identify which customers are very likely to convert? Here comes predictive modelling into place. With predictive modelling I capture the behavioural ‘rules’ that lead to a conversion by using past behaviour of visitors into a model that can be applied to future visitors. For example, it could be that a customer who had a recent visit with many page views is likely to buy something in the coming 15 days.

Capturing behavioural rules

The first step is to set and define the target, in other words, what do we want to predict? I want to predict conversion. However, I need to be more specific than that. For example, what do I consider as a conversion? When should this conversion take place? For an omnichannel fashion brand I decided to use an e-commerce transaction as a conversion, and it should take place within 15 days of the last visit.

Next I define what feature variables can be used to predict the target. These are the behavioural rules we want to capture. You can use data sources such as customer databases, transactional databases and web analytics, like Google Analytics. The most important questions are: is the data available, can it be integrated and can we use it for our specific application. For the omnichannel fashion brand I wanted to use the prediction in Google Audiences, the brand was using Google Analytics, and didn’t have much first-party data in Google Cloud (yet).

The selection of feature variables is very important, since a correct set of variables can have a significant impact on your model performance. When selecting the feature variables you really need to understand what behaviour leads to your target. You can talk to a Customer Journey Specialist to help you create the features. For the omnichannel fashion brand I used the following features: 

• product value of categories seen in last visit
• growth of visits in the last 30 days
• device type, 
• percentage of logged in visits
• percentage of organic/direct visits

What’s next?

Now that you have your data, the features and the target in one table, you are able to do a feature selection analysis to see which features are contributing to your target, and which are not. Irrelevant or partially relevant features can negatively impact model performance. In the next blog post I will describe how to do a feature selection analysis.

ABOUT CRYSTALLOIDS

Crystalloids helps companies improve their customer experiences and build marketing technology. Founded in 2006 in the Netherlands, Crystalloids builds crystal-clear solutions that turn customer data into information and knowledge into wisdom. As a leading Google Cloud Partner, Crystalloids combines experience in software development, data science, and marketing, making them one of a kind IT company. Using the Agile approach Crystalloids ensures that use cases show immediate value to their clients and frees their time to focus on decision making and less on programming.