Eric Siegel opened Predictive Analytics World with a view of using predictive analytics in enterprise risk management. Eric began by giving some examples of “macro” risk – single, catastrophic risk events. But his focus, and the focus of predictive analytics, is on “micro” risk – risk-based micro decisions in my terminology. These are risk decisions focused on a single customer, a single claim, a single loan. Examples include the risk of a customer becoming a loss by making a claim, leaving your customer base, failing to make payments on a loan.
Eric Siegel opened Predictive Analytics World with a view of using predictive analytics in enterprise risk management. Eric began by giving some examples of “macro” risk – single, catastrophic risk events. But his focus, and the focus of predictive analytics, is on “micro” risk – risk-based micro decisions in my terminology. These are risk decisions focused on a single customer, a single claim, a single loan. Examples include the risk of a customer becoming a loss by making a claim, leaving your customer base, failing to make payments on a loan. Only if these risks can be exposed, seen, can they be managed. And this means that predictive analytics practitioners must “kick down the door” of risk management and get them to focus on aggregated micro risks not just headline risks.
Eric then gave some definitions of predictive analytics
- Analytic methods that score each customer
- Deployment of predictive modeling/induction
- Data-driven micro-risk management
Eric likes the last one, and I would agree.
More Read
5 different areas can have predictive anlaytic risk management:
- Sales and Marketing
Example risks are that making a retention offer will remind them they want to change or that they will accept an offer but would have stayed anyway. You could charge a higher price and drive a customer away or a lower one when they would have purchased at the higher. While I regard these as mostly opportunity-centric decisions, Eric sees them as opportunities to pick the customer treatment with the lowest risk.
Eric gave a quick overview of uplift modeling (covered in this post about a previous keynote by Eric) - Fraud
Fraud is one of the two most established uses of predictive analytics to reduce risk and covers healthcare fraud, credit card fraud and much more. - Insurance
- Healthcare
Insurance and healthcare are both industries with lots of actuarial work going on – trying to predict risk for populations of drivers, patients etc. Many are now extending this with predictive models of risk of falling sick, of making a claim etc. One of the presenters is from Heritage Provider Network who are launching a $3M prize for the best prediction of who will be admitted to hospital (powered by the folks at Kaggle). Like the Netflix prize, Eric expects to see multiple teams come together and build sophisticated ensemble models. - Credit
Credit is in many ways the home of predictive analytics in risk management – credit risk is by far the most common meaning of the word “risk” in analytics. Predicting credit risk for a specific product and a specific customer is key to managing credit risk.
Obviously Eric’s perspective here is very similar to mine – check out Risk by risk – a decision-centric approach to risk management for instance. He also talked about IBM’s Watson a couple of times, particularly in the context of building ensemble models, and you can check out my post on IBM Watson here, and he gave a quick overview of lift curves .
