Use Case - Churn prediction model

I'm gonna show you how you can use count to build your really great churn prediction model. So in the canvas, you can see I've added in various different data sets that I think could be helpful as predictors of churn. I've added in a customer list from our CRM, all the support tickets we've had over the last eighteen months. I've added in some communications via email or in app messages as well, product usage data by customer, anything we've got from NPS surveys, and our revenue and contract information again from our CRM. So all these tables could have come from our database, or I could have just exported them and imported them as a CSV. And alongside with these different data sources, I've actually added a bit of methodology, how I want the agent to go about the work. So I've asked it to look at these different data sources and compare the activities and usage of these different these different data sources to the historical churn customers to give us a way of understanding which of these indicators could help us predict churn. And then I want to apply that scoring model to the current customers who are live to help us get a score per customer of their likelihood of churn, and then give me a sense of what potential revenue impact could be and where we should focus. I've also asked it to flag any assumptions it's finding or any data quality issues that it comes across where we can't actually get good data from this, either to exclude it from the model as a version two or to go away and improve the data quality in the right places. So now I've got all this set up. I'm gonna add an agent into the canvas and I'm gonna give it those tables and that sticky as context. And I'm gonna ask you to go build me the the model. Hey, Count dot ai. Could you use these tables and this methodology to go build me a churn prediction model, please? Great. So the agent's now gonna go off, look at those different data sources, look at the methodology, and go build us a verse first version of its churn prediction model. Okay. So let's see what the agent's done. So as you can see, it's given us an executive summary. You can see that there are hundred and sixty five active accounts requiring immediate intervention with a potential one point four million ARR risk. We can see the historic churn rate. We can see that it thinks it's got a ninety two percent accuracy on its model predictions based on historical data. And it's given us back a various assumptions and data quality flags. I it's found different date formats. It's found that there's some typo and variants in support tickets, and that two hundred and six customers lacked any custom success management person attached to them. So one of things I love about the agent is not only can we follow it through its whole methodology and its thinking on the right hand side menu here, But we can actually not only read this output, but also see the actual code that it's written to to build the assumptions of the model. So if we want to make any small tweaks ourselves, we can. Or we obviously can just ask the agent to make make some changes directly in the in the chat prompt here, maybe to change some of these assumptions around data quality. But assuming that we're then happy with the model, we can do a few things. We can obviously bring in our team. We can share this whole canvas with our team and work with them to improve the model or turn it into something which we wanna use as an ongoing report. Or I could even, look to present this as a presentation. I wanted to talk people through the methodology as well. But you can see this already goes from a set of very powerful datasets into a fantastic churn prediction model, which we could iterate from or just start using straight away.