Predictive targeting: How machine learning can help
Lots of companies hire data scientists, only to discover that management and data scientists don’t tend to communicate very well. Unlike the management team, the data scientist doesn’t have the full picture of the company. He might build something he thinks is valuable for the business, but in the end the system is less than optimal because his perspective is limited. As management, you need to be able to brainstorm ideas and bring them to an experienced data science team that can help you realise your goals.
How does this actually work?
The data you’re collecting includes everything a user has ever clicked on. So you know what she did in the past, and you want to know what she’s going to do in the future. When you feed this data through a machine learning algorithm, it turns patterns into predictions. You now have a predictive system. And you magically know what your users are going to do.
It’s simple – predictive targeting takes all those different clicks and turns them into predictions for your products.
From targeting to conversion
Once the algorithm tells you that a customer is really interested in your product right now, you send him your campaign email. If you do it right, then your conversion rates increase dramatically.
So how do you do it right? You want to send the email to users with the top 1% highest scores for interest in this particular product. You contact only those users who are most likely to convert, and you get much higher conversion rates than you would with simple rule-based targeting.
And this happens automatically. You take half a billion user events, let the machine churn over them, and test the system offline. Then you can accurately predict what your conversion rate is going to be. The machine finds patterns that are far too granular for a human analysis. Because a machine can take account of everything, it identifies new patterns that really matter for your conversion rates. That’s what leads to radical improvements of 200–300%.