In matches between computers and chess masters, the machine usually wins.
In trying to best determine a car shopper’s creditworthiness, Tom Anderson puts his money on automation.
It’s not that people aren’t important. They are in terms of programming computer systems to do as told in deciding whether to reject a loan application or approve it, and at what rate.
Such systems are hard to beat once people set them up right, says Tom Anderson, CEO and managing partner of Exeter Finance. He joined the company in November.
The lender recently rolled out a new so-called automated-decisioning system. It can make car loan determinations in 20 seconds.
Anderson, 51, is a finance industry veteran, having held CEO posts at four previous companies. He holds an undergraduate degree from Dartmouth and a master’s degree in management from MIT.
He speaks with WardsAuto about the new system and the man vs. machine aspects of automotive lending. Here’s an edited version.
WardsAuto: How do you set something up that makes a decision that quickly. Is it incredible algorithms or what?
Anderson: A bunch of factors come into play. Technology has gotten better. Information and the access to it have increased dramatically. There is having the right people to build the appropriate algorithms and track, manage and stay on top of it. That’s important in a risk business.
If you do it right, it’s incredibly valuable. We’ve invented a lot of money in it. It’s paying off. If dealers are selling more cars and you are making customers happy with what they’re walking away with, you tend to do more business.
WardsAuto: A while ago, many lenders said they relied on auto-decision systems to refuse loans, but not to approve them. They didn’t have enough faith in the system to let it do approvals without human involvement. Where are we now with that?
Anderson: I’ve been in financial services a long time. I’ve done around 100 studies looking at the performance of human underwriters alone; human underwriters supplemented with algorithms and automation; and automation alone.
In 100% of the cases, automated decisioning outperforms the other two. Bottom line: human underwriters destroy value.
Those people telling you the world isn’t or wasn’t ready for automated decisioning are stuck in the old world. I hope they stay there because I can kick their patootie using this.
It’s clearly easier to use technology and data today, because they’ve gotten better than five years ago. But in 1991, I built automated models for a consumer-lending company, and back then people said, “You can’t do it, it doesn’t work.” I heard all the stories. But we smoked everyone else. Those other people are just protecting the old way and don’t know what they are talking about.
WardsAuto: Again, it was some years ago when they were saying that. Maybe the technology wasn’t there then.
Anderson: Believe me, the technology was there.
WardsAuto: So when the human gets beat by the machine, how is the human failing?
Anderson: This is a slight oversimplification, but not by much. There’s a lot of data. You can develop statistics and increase the accuracy of your predictions of somebody’s risk. Humans then override what statistics say.
There’s a reason casinos make money. Because statistics are in their favor. That doesn’t mean every single time they are right, but on average, the casino wins.
When humans intervene, thinking they’re smarter than the statistics are, they make exceptions. Those exceptions rarely win in the long run. It doesn’t mean they’re wrong on every single one. But there is a reason statistics say that is not the right path.
Humans are incredibly valuable in developing insights about what might be predictive, and therefore what can test statistically. But on average to allow human (underwriters) to make exceptions or to deviate destroys value. They can underprice or overprice, say “yes” when they should say “no” and vice versa.
WardsAuto: What was the product development for this new system? Did you sit down and say we need something like this as a point of difference in the market?
Anderson: A team of folks recognized the speed and precision of upfront decision- making is important in the market. So they started down that path. Blackstone, our largest shareholder, also supported it.
I’d love to take credit for it, but it was started before I got here. Maybe five years from now, I’ll rewrite history and take credit for it.
WardsAuto: Was the idea to have a competitive edge?
Anderson: One was to have an advantage in the marketplace. Two was to have an advantage in managing the risk. We’re in the risk business. We make loans and have got to be repaid.
WardsAuto: Do you have data showing how often the system approves a proposed loan opposed to rejecting it?
Anderson: We have a credit policy which sets what’s outside (what) we are willing to play in, and we’d obviously disapprove what’s outside of that. Within it is about correctly pricing the risk. It’s almost less about disapproving those within the range opposed to pricing the risk appropriately.
Our percent of approvals actually has gone up, not down, and the performance of our loans is better than before.
Sorry, No Exceptions Allowed
WardsAuto: Some dealers say lenders relying a lot on automated-decision systems don’t understand some of the circumstances behind a person who has a checkered credit history or a relatively low credit score, but has a logical explanation for that. They say getting on the phone with a lender and explaining extenuating circumstances often makes a difference.
Anderson: This is back to what I was talking about with humans vs. the technologies.
More than 50% of the time, those quote-unquote extenuating circumstances work out negatively. So, our system does not allow that. There’s a reason for that. Even though they plead a compassionate case, and you go “OK, that’s logical,” factually, they are wrong more than half of the time.
By saying yes, you find yourself losing money and consumers defaulting. The answer is that the system does not allow bending of the rules. Factually, that’s bad for the lender and the consumer in the end.
We do try to keep track of what those reasons and circumstances are and if there is a hypothesis that makes sense we will test it statistically. If it ends up being true, we just build it into the models.
WardsAuto: What would be an example of that?
Anderson:There are always new insights that come up. Just to paint a picture, let’s say you start with a model only using the FICO score.
Then someone says, “That’s nice, but this person just got a new job paying $10,000 more a year. You start thinking maybe income is a factor in addition to the FICO score, and you start tracking income relative to performance. And of course you find income is a predictor. So you add it to your model. Now your model has two variables, FICO score and income.
Then someone might say, “Wait, I realize their income isn’t that high but they are only borrowing 70% of the value of the car, and the loan relative to the value of the car is important.” And you say, “Gee, that sounds like it might make sense.” So you track that, look at it and find loan-to-value does matter. So you add that to the model.
I’m not suggesting that what people think doesn’t matter. The point is, you can find out factually. And if it factually holds true, you can put it in the model, opposed to making one-off exceptions. If you test things and it turns out it doesn’t predict it, you don’t want to make exceptions for it.
WardsAuto: So you’re adjusting the model as things progress? Are there refinements being made?
Anderson: Constantly. We monitor it on a daily basis.
WardsAuto: Millennial customers are particularly interested in the quickness of an auto purchase, one that doesn’t get bogged down by delays. How important is that young group of buyers to fulfilling the need for speed at the dealership?
Anderson: It’s an interesting question. Consumers are demanding faster decisions. I’m sure if you broke it down, you could say a big percentage of that change is driven by the younger population.
But across the board, there is clearly a trend and a shift in how people buy. Speed and simplicity are incredibly important, and increasingly so.
I recently bought a car. It aggravated the heck out of me because I was stuck there for two hours. I said, “I know this business. Here are the answers: No, no, no, yes, yes. Give me the keys.”
WardsAuto: I assume the delay wasn’t because your loan application was rejected?
Anderson: Thankfully not. That would have been really embarrassing.
