Wise Practitioner – Predictive Analytics Interview Series: John Foreman of MailChimp
By: Eric Siegel, Founder, Predictive Analytics World
In anticipation of his upcoming conference keynote at Predictive Analytics World Boston, “Problems, then Techniques, then Toys. Keeping Your Predictive Analytics Right-side Up,” we asked John Foreman, Chief Scientist at MailChimp, a few questions about his work in predictive analytics.
Q: In your work with predictive analytics, what behavior do your models predict?
A: At MailChimp, we use predictive modeling across the application to improve the experiences of our users. Some examples:
- We predict users who are unlikely to send spam, and we allow them to begin sending email through the system without manual account vetting (manual vetting slows people down by a day)
- We predict users who are likely to send spam, and we shut them down before they send in order to protect our email-sending ecosystem
- We predict users who are on a free account but who are likely to pay in the future. We then give them the same customer support given to currently paid users
- We predict users who are most certainly not bots and we remove reCAPTCHA entirely from the app for them
- We predict the knowledge base articles that a user is most likely interested in when they contact customer support
- We predict the best time to send an email address marketing content and provide that to users in our Send Time Optimization (STO) system
- Given a small segment of email addresses, we predict other email addresses on a user’s list that have the same interests to facilitate better segmentation and targeting
- We predict demographic data on email addresses
These are just some examples of the different models in play.
Q: How does predictive analytics deliver value at your organization? What is one specific way in which it actively drives decisions?
A: Predictive analytics is a key part of our user on-boarding and compliance process. MailChimp has over 6 million customers, and without predictive modeling, the company would be left linearly scaling the headcount of customer support and compliance. Predictive models enable us to automate the easy jobs, allowing our compliance personnel to hunt down the worst of worst in terms of bad actors. This lowers our headcount, saving us a great deal of money. We are able to manage 6 million customers with less than 300 people total at the company.
Furthermore, our user-facing predictive products (Send Time Optimization & Segment