The machines have won. Now, Let's make them even stronger to stop fraud and keep customers, says Arnab Gupta
Today, companies spend an inordinate amount of time trying to improve the quality of their employees. They lay out millions on programs to "develop their human capital" and "attract and retain the best talent."
This is certainly important. Indeed, as the CEO of a consulting firm, I know that talented and motivated people are absolutely critical to our success. But I respectfully submit that it's not sufficient. Yes, it's important to improve your human intelligence. However, the real winners in the marketplace will be those who succeed at furthering their machine intelligence.
Think of it: We live in a world where computers, not people, are in the driver's seat. In banking, virtually 100% of the credit decisions are made by machines. In marketing, advanced algorithms determine messages, sales channels, and products for each consumer. Online, more and more volume is spurred by sophisticated recommender engines. At Amazon.com (AMZ), 40% of business comes from its "other people like you bought…" program.
Tell It to Iceland
In this brave new world, people handle only the exceptions. Machines do all the heavy lifting.
While these words doubtlessly produce a frisson of fear in many—the specter of Hal in 2001 A Space Odyssey comes to mind—the problem isn't that machines are making so many decisions for us. It's that they sometimes aren't fully up to the job. And because they are so powerful and pervasive, the results can be disastrous.
Take the recent credit meltdown. Computer models underlie close to 100% of the decisions to make too-risky loans. Models, not individuals, told us that a portfolio's risk was balanced appropriately (tell that to Iceland). Unfortunately, these models didn't have enough good data, and good judgment, to figure out that, if everything went to hell in a hand basket all at once—well, everything would go to hell in a hand basket all at once.
Making Machines More Human
And once that cataclysm happened, the hits just kept on coming. At credit-card companies, terrified managers tasked their computers with identifying which customers were the riskiest, and computers over-obliged. It didn't matter if you'd been a great customer for 30 years and were well under your credit limit—if the machine found a whiff of risk, your credit line was slashed.
Now a real human might have looked at someone and said, "Hmm, George was a couple days overdue with his payment last month, and he's used up a lot more of his credit line than normal, but this is the only time he's been late in 15 years, so I think I can trust him." And when a little further digging revealed that George had been on a dream vacation with his family last month (hence the higher balance and slower payment), the human, nuanced decision would have saved a profitable, trusting relationship.
But the machines literally didn't have it in them to figure it out. With some adjustments, however, they could have. One of our clients, a credit-card company, took a second look at the customers it had targeted for adverse actions. By applying sophisticated next-generation modeling approaches, it found almost 20% who really weren't risky at all, saving the company $20 million to $30 million per year and retaining a cadre of loyal customers.
Stopping Credit-Card Fraud
More broadly, we have found that amazing value and competitive advantage can be created by improving the thinking power of machines. Companies like Google (GOOG) and Amazon have already discovered this, and their performance and market valuation reflect it. The billions they invest to make their machines smarter pay off in profits and competitive advantage.
We've found that applying advanced nonlinear modeling and multivariate analytic techniques yield significantly better results for businesses than traditional approaches. They work in a whole range of areas: credit and collections, fraud, marketing and cross-selling, pricing, portfolio valuation, and more. Anywhere companies seek to predict behavior, preference, or affinity is an area of significant opportunity.
An example: Bust-out fraud, when a seemingly good customer suddenly maxes out credit lines and then defaults, is a huge problem in the credit-card industry. To stop it, banks must detect the pattern before the customer runs up huge bills. Traditional linear modeling techniques simply have not been powerful enough to pick up faint early warning signals. But using a neural network approach along with other advanced nonlinear modeling techniques makes it possible to identify much more of the fraud, and do so up to five days earlier, saving businesses millions of dollars each year.
Human Smarts Not Enough
Another example of what machines can do: create stronger and more profitable customer relationships. We've found that powerful recommender engine techniques can help an online/route sales retailer expand its business. A nonlinear model "scores" each family on the salesperson's route, each night, and develops recommendations that then are pushed to each salesperson's handheld computer. These microtargeted recommendations can increase sales 10% to 15% despite today's shaky economy.
And of course, there is Netflix (NFLX) Prize, the global analytics competition from the online movie rental company that attracted 41,000 entrants. To crack the problem, all the contestants had to employ an entire pantheon of new modeling techniques, and develop world-class expertise in blending model results together. The result of my firm's own two-and-a-half year effort is an engine that can improve Netflix's ability to recommend the right movie to the right person at the right time. Given that 60% of Netflix's volume is directly tied to these recommendation—and that good movie selections are a critical factor in customer retention—a 10% improvement is huge.
The difference between good performers and great ones has, historically, often been tied to human capabilities. But looking forward, it may no longer be enough. In this age of complexity and data explosion, vastly improving machines' capabilities—making them faster, more "human-like," and more intelligent—can help organizations take a quantum leap forward.