The power of prediction in machine learning

Supervised learning and predicting the future

Machine learning paired with prediction is incredibly powerful. You can train a model, make predictions, and take action as required.

Examples where a business may want prediction, include:

  • A SaaS business wants to predict which customers are most likely open to upselling.
  • An ecommerce business wants to better target products to its customer base.
  • Any company trying to drive customer retention.
  • Fraud detection.
  • And much more!

Here’s a sample output where we were predicting customer churn based on a customer success data set. We used the same data set in our first blog to learn what the impactful predictors are. Based on this sample, you can see that customer 3000 and 3002 are predicted to leave the product.

This is a screenshot of the prediction results. The left column shows their IDs ("CompanyID"). The middle column shows their predicted outcome (yes or no). The right column shows the confidence of the prediction, which ranges from high, medium, to low.

Taking prediction a step further

Within prediction, there’s a concept of scenario modeling or “what if”. What if XYZ occurred—would the results of my original prediction change? What-if scenarios are really important for identifying what actions or changes could result in different outcomes, and determining which course of action may be best.

Using the same customer churn example from above, we can choose to do a what-if style prediction in two ways:

  1. For a single row—this is useful when you want to know what impact a specific change would have on that row.
  2. For all rows—this is useful when you want to see how a change impacts the entire data set.

Continuing with the example we’ve been using, we learned in our last blog that platform stability (uptime) was the most important predictor for a customer staying. So, what happens to our initial predictions if product stability improves?

This screenshot shows how the predictions change after using Dacture's "What If" feature. The left table shows the original predictions. The right table shows how the predictions have changed. The first and third row have changed from "yes" to "no", indicating that those customers will no longer churn. The confidence of the prediction has also changed for those rows, from high to medium.

You can see that increasing uptime has led to customers 3000 and 3002 moving from a Yes to leaving the product to a No, which indicates they are more likely to stay. That means that if this SaaS company were to invest in their infrastructure and stability, it would significantly improve their customer retention.

When we looked at the impactful predictors for our data set, we saw that Feature 4, which is just a miscellaneous SaaS feature, was also fairly impactful in predicting whether a customer stays with the product. Maybe we want to know if Feature 4 is a sticky feature (a feature that keeps customers using the platform). We selected customer 3004 and did a what-if scenario based on a single row. So what if customer 3004 starts using Feature 4?

This screenshot shows the "what if" results when performed on a single row. In this example, the predicted outcome is now "yes" and the confidence of prediction is now high.

Well, they’re predicted to stop using the product. Based on that, it’s fair to say that Feature 4 has some issues and is driving customers away.

As always:

  • If you want to learn more about what Dacture can do your for organization, discuss use cases, or see a demo, you can schedule something with us
  • You can email us at if you have questions

Get notified when Dacture is live

Please leave your email address below and we will notify you when Dacture is live