Machine learning is a great technology to optimize aspects of your business – whether it be subscription conversions or e-commerce purchases. But, across many of our partners, we’ve seen that implementing machine learning so that the technology is directly optimizing for a particular business metric can be a challenge.

In this blog post, we will discuss how common machine learning use cases, like predicting a future user action, can have unforeseen challenges associated with them.

Calculating the Probability of User Converting

Let’s start with a basic example – calculating the probability of a user converting to a subscription product. In general, machine learning is a great technology to predict a future action a user will make. However, there are considerations when using machine learning to determine the probability of someone converting.

Most machine learning models do not provide a probability of converting but instead a “model score” for each user (only a small subset of machine learning models like, for instance, logistic regression output user probabilities). If we want to calculate the probability we would need to build a solution which converts model scores into user probabilities (note that Cortex enables customers to directly specify that models output probabilities and not model scores while still providing our customers with the full set of powerful machine learning models).

Using Machine Learning Probabilities

Armed with probabilities for every user (and not model scores), your business can solve for use cases like determining whom to send an email. I.e. you can calculate (probability of purchasing when sent an email) X (average order value of a purchase) and ensure this figure is greater than the cost of sending the email. Model scores which are not probabilities (i.e. the outputs of most machine learning models) can not be used in this way.

Example: Average order value of a purchase = $5, Cost to send email = $0.10

At Vidora, one of our top priorities is to make sure our customers are successful in deploying and driving value from their machine learning deployments. Ensuring that Cortex features directly align with our partners’ business metrics and business value is an important consideration for our product roadmap. We have several features over the coming months which further enable our partners to directly align machine learning with their business goals.

If you are interested in learning more about how machine learning can help optimize for business value, please reach out to us at

Want to Learn More?

Schedule a demo and talk to a product specialist about how Vidora’s machine learning pipelines can speed up your ML deployment and ultimately save you money.