Connecting the Dots: Introduction to Causal and Machine Learning
I have taken the route of defining one of our most complex problem spaces in this series. Behaviors are very hard to predict reliably over time and a large group of people. I went over the reasoning in detail over several posts, so I will not rehash that.
These are high-value use cases in our field. Nike wants to increase sales leveraging the sneaker collector community as a more directed marketing and evangelization arm. Tinder wants to keep new users on their platform. Peloton must re-engage subscribers and bring them back to the bike, so they don't cancel their subscriptions.
Intel needs to influence its suppliers to ease its current supply chain issues. Retailers needed to influence their customers to be less price-sensitive as inflation became inevitable two years ago. Luxury brands must position themselves as desirable and relevant to the millennial generation to survive.
You will often be in the position of influencing your business to transform into an AI-first business for its own survival. Lift the hood or trunk if you're fancy, and most of our models include behavioral data. Spend some time thinking about how much of your data is generated by a person doing something, how much inference you serve about people, and how many use cases involve someone deciding or choosing a behavior.
I will primarily use customer-facing examples because the link to people is obvious. In reality, every time we build a model that learns from people, explains people, or serves data to people, the same principles apply. Behavior is a choice. In all three scenarios, we are modeling or attempting to influence decisions.