Conversion Rate Optimization with Machine Learning
It seems as though machine learning is only attainable by the chosen ones - data scientists with names ending in PhD or large corporations with deep pockets. This is because machine learning typically requires access to large datasets, substantial processing power, and a team of mathematical savants. These prerequisites have historically limited the potential of machine learning applications.
In reality, recent developments have enabled marketers to apply machine learning techniques to commonly used data analytics (e.g. Google Analytics). All you need to bring is your curiosity.
Mirror, Mirror On The Wall...
Marketers can glean insights and make better decisions by investing a little time and effort into applying machine learning techniques to Google Analytics.
Start by asking, if we had a crystal ball, which online visitors are the fairest of them all? The answer is simple: the fairest of them all are the ones who have converted into paying customers. The problem is, out of your thousands, if not millions of visitors only tiny percentage will actually convert. Meaning that most of your marketing efforts are lost to those who are not interested or ready.
If you knew which online visitor had a higher probability of converting, you could realign your budget so that you could spend more on leads who are worth nurturing. This is where machine learning can help you rank your online visitors.
Bibbidi Bobbidi Boo
Now that we’ve identified the mystery, the next step is to learn from your history. More accurately, you'll want to teach your sales history to your machine. Your historical sales data will enable you to train your machine to find prospects with similar features and attributes. This is because machines learn from observing real-world interactions.
Does The Glass Slipper Fit?
Machine learning is like a prince searching for Cinderella - he takes the glass slipper (historical sales data) and compares the fit to every possible suitor (online visitors). It may sound daunting, but fortunately, Google has developed BigQuery ML (Machine Learning) that will help us find our Cinderella.
Putting Humpty Dumpty Together
BigQuery ML (beta) enable analysts (including marketers) to create and execute machine learning models in using standard SQL queries without the need for Python or Java programming experience. There are three types of models that BigQuery ML currently supports (as of 2018):
- Linear Regression: used for predicting a numerical value
- Binary Logistic Regression: used for predicting one of two classes (e.g. identifying whether an online visitor is a current customer or not)
- Multiclass Logistic Regression for Classification: used to predict greater than two classes (e.g. classifying online visitors on their likelihood (low, medium, or high) of converting
For Conversion Rate Optimization, we can use Binary Logistic Regression (logistic_reg) to determine which online visitors who are likely to convert (or not). There are three basics steps to sort your online visitors: Create Model, Predict, and Export.
The create model (CREATE MODEL) will allow marketers to apply one of the three models (mentioned above), while the predict model (ml.predict) will apply the machine learning model to your online visitors (found in your Google Analytics). Finally, you’ll want to export your results and target these users with a Search Engine Marketing campaign.
Using the sample dataset provided by Google, your results should look like this:
Happily Ever After
The beauty of this technique is that marketers can now glean insights and make better decisions without relying on a data scientist. By investing a little time and effort into applying machine learning techniques to Google Analytics, you can hone your next marketing campaign to target prospects who are thirsty.