It seems like you can’t spend more than a minute or two reviewing online articles or social media without stumbling on something about artificial intelligence (AI) or machine learning. It’s the hottest thing since the iPhone. But while AI is heralded as amazing new technology, it is in fact an amazing 30+-year-old, proven technology. Like Dorothy in the “Wizard of Oz,” you don’t really have to look much further than your own backyard to find AI––specifically, in your own wallet, mobile or otherwise.
Machine learning and AI are a prominent part of banking and processing payment cards, not just reserved for self-driving cars and computers that can win at Jeopardy. In banking, machine learning not only improves predictions, it is effective for improving subsequent decisions. Machine learning has a presence in many of our analytic models at FICO that are widely used in the financial industry. Here are a few examples.
The FICO® Falcon Fraud Management solution is the industry leader, protecting two thirds of the world’s payment card transactions against fraud. In far less than the blink of an eye (40-60 milliseconds, to be precise) Falcon scores each authorization that a merchant submits for approval.
Here, Falcon uses adaptive analytics, a type of machine learning in which self-learning models work with the Falcon consortium models to improve the prediction of future fraudulent behavior based on fraud attacks in production. These models score transactions based on recent known fraud and non-fraud transaction data, dramatically improving the models’ sensitivity to changing fraud patterns, and keeping up with the fraudsters.
Making credit decisions
Many consumer credit decisions (such as applications for credit cards and loans) are made using FICO® Origination Manager. In this solution, machine learning helps significantly improve the overall predictive power of our models by determining the risk assessment of applicants through AI. The solution then uses those insights to inform an optimal model design and segmentation for a traditional scorecard model.
Using this AI derived knowledge, and subsequently constructing traditional scorecard models, allows for strong improvement in model performance while sustaining the traditional advantages of scorecard models including transparency and explainability.
The FICO® Score is used by 90 of the top 100 largest US lending institutions for their risk assessment needs. A FICO Score is generated using multiple scorecards, with each scorecard tuned to assess risk for a specific consumer segment—for instance, consumers with serious delinquencies. Machine learning is used as a benchmarking tool for the scorecards that are manually developed, allowing the team to garner additional insights much more quickly. These insights help to inform variable generation and segmentation schemes.
On a side note, I recently read a fascinating book on the social implications of big data, Weapons of Math Destruction. In it I was very pleased to read what author Cathy O’Neil says of the FICO Score:
“Fair and Isaac’s great advance was to ditch the proxies in favor of the relevant financial data, like past behavior with respect to paying bills. They focused their analysis on the individual in question – and not on other people with similar attributes. E-scores, by contrast, march us back in time. They analyze the individual through a veritable blizzard of proxies.”
FICO is proud that our history of empirical models based on quantifiable hard behavioral attributes and outcomes. To be recognized for it in this way is extremely gratifying.
The FICO® Marketing Solutions Suite uses patented FICO analytics, rules management and optimization technology to enable large-scale personalized offer formulation. For example, one leading North American grocer uses Marketing Solutions Suite to score thousands of potential offers across millions of customers on a weekly basis, to determine the best set of offers for each customer within the loyalty program.
Machine learning is used in this solution to bolster predictive results. The automated tree-ensemble models produce scores on the tens of thousands of products to offer to retail clients, and effectively specify the optimal set of offers to provide. The benefit of automated modeling has truly been seen in the time to market for these solutions, reducing the effort from months to days.
Where are we going?
Machine learning and AI aren’t only for making better predictions. In a time when real-time interactions are prevalent, decisions are being made on the fly. Through the use of machine learning algorithms, decisioning can be done instantaneously in our world of digital channels. We are already focusing on this area through our mobile analytics initiatives.
Where I am going?
If you’re a New Yorker, this week I’ll be in your backyard. I’m delighted to be keynoting at Predictive Analytics World for Financial in New York City on October 26, and at the Advisen Cyber Risks Insight Conference, also in NYC, on October 27. There I’ll be talking about the new FICO® Enterprise Security Score, which I’m incredibly excited about. Come see me! Or just follow me on Twitter @ScottZoldi.