Data Science & AI

Separating AI Hype From Reality

Bruce Upbin

March 12, 2019

Artificial intelligence is taking on aspects of the irrational exuberance from the previous decade. Last year, venture capitalists backed 1,028 AI startups compared to 291 in 2013. The number of AI startups in Israel alone grew to 1,150 last year from 512 companies in 2014. Then there’s the conference circuit. At a recent MIT conference, one professor warned that it’s getting harder to distinguish the real AI advancements from “snake oil.” Evidence of that? A recent story in the Financial Times cited a report claiming that that two-fifths of Europe’s AI start-ups do not use any AI programs in their products.

Amazon did a good job of spoofing AI hype in its Super Bowl ad, which featured rejected ideas for its AI-enabled Alexa home assistant. An AI dog collar accidentally sends dozens of giant bags of dog food to Harrison Ford’s house. Forest Whitaker struggles to hear his podcasts playing out of his AI toothbrush… because it’s in his mouth. It’s only funny because we all know we don’t need AI to tell us what kind of beer we like.
But while AI may be ripe for satire today, let’s acknowledge it’s already made a very real impact. At ZestFinance, we’re seeing every day how lenders are able to lower their charge-off rates and expand their pool of good borrowers by using AI. Our ZAML suite of software tools employs machine learning (ML) — an application of AI that uses mountains of data and sophisticated math to help lenders make better decisions and predictions. Instead of just looking at a FICO score or a handful of variables, lenders using ML can increase the number of variables used by up to 100x, producing a more predictive model that has helped banks reduce charge-offs by 30% or increase approval rates by 15%.

It’s easy to make fun of the kind of AI that tries to tell us what kind of beer we like, but let’s acknowledge the very real impact AI is having in the financial services industry.

ML models not only analyze thousands of factors, but they also analyze the interactions between them, creating more accurate scores for loan applicants. This helps both the lender and the applicant because it expands the idea of who can, and should, receive a loan. So-called “thin-file” clients, applicants who may not have a credit card or who have never applied for a loan before, are often rejected simply because there aren’t enough traditional data for a lender to make a good decision about the loan. ZAML can ensure that thin-file applicants who are actually in a good position to make repayments get the loans they need to buy cars and homes and start new businesses. For one auto lender, ZAML tripled the approval rate for thin-file applicants with no increase in risk.

The use of AI in regulated industries such as finance and health care requires the same transparency and model monitoring that lenders demand today for operating models safely. If the population of people applying for loans is different than the population the model was trained on — due to changing economic conditions or a new marketing campaign — a lender might not know that the model is making decisions based on outdated assumptions until the impact is felt on the bottom line. Zest’s ML-based monitoring helps lenders see potential risks early so they can adjust or rebuild their models accordingly. And our industry-leading explainability tools allow modelers to explore, interpret, and document at every step of the way.

While credit underwriting might not be as flashy as, say, Harrison Ford’s pets or Forest Whitaker’s toothbrush, it’s an industry that directly affects millions of people and it offers a glimpse into what AI will be able to accomplish as the technology matures.

Photo by Franki Chamaki on Unsplash

Thank you for subscribing!
Something went wrong while submitting the form.