GIỚI THIỆU VỀ TẬP NÀY
Mike de Vere runs Zest AI, a company that has been applying machine learning to credit underwriting for over two decades, starting with some of the largest banks on the planet and now serving a large share of the credit union market. Since his last appearance on the show three years ago, Zest has expanded well past underwriting into fraud detection and portfolio management, tied together by an intelligence layer and a generative AI companion called LuLu. Mike makes a specific argument in this conversation: machine learning still makes the credit decision, generative AI makes the feedback loop faster, and the real advantage available to community financial institutions is a willingness to pool what they know.
What We Covered
- Zest today, from underwriting to fraud to portfolio management
- Why the intelligence layer is what makes an ecosystem
- Starting with Discover, Citi and Freddie Mac, then moving down market
- LuLu, named after a corgi, and what she actually does
- Safety and soundness as the first use case for most institutions
- Replacing quarterly reports that used to take weeks
- Peer benchmarking versus building your own data lake
- Collective intelligence across 2,000 credit models in production
- Why generative AI has no role in making the credit decision
- Shrinking model refit cycles from 18 months to daily evaluation
- Zest customers versus non-customers on growth, delinquency and efficiency
- Cash flow underwriting, and why generic national models fail
- Zest Protect and fighting AI-powered fraud with AI
- The two objections that come up most in sales conversations
- Takeaways from the IQ AI Lending Forum in Santa Fe
Key Takeaways
- The performance gap is measurable. Comparing Zest customers to non-customers across 2024 and 2025, Mike says his customers grew 16 times faster, ran roughly 20 points lower on delinquency, and were 501 basis points better on efficiency ratio.
- Generative AI belongs around the credit decision, not inside it. Zest still uses supervised, locked-down machine learning models for underwriting, because a regulator will ask you to explain the decision. What generative AI changes is the speed of evaluation, from an 18-month refit cycle to daily.
- Comparison is where the value sits. A lender looking only at its own data lake has visibility on itself and nothing else. LuLu is built to normalize performance data across institutions so a chief lending officer's instinct can be checked against thousands of real policy instances rather than one career's worth of experience.
- Community lenders have a structural advantage they underuse. The credit union industry holds roughly $2.4 trillion in assets. If it acted as one institution, it would be bigger than Wells Fargo, and unlike the big banks these institutions are actually willing to share.
About Mike de Vere
Mike de Vere is the CEO of Zest AI, the AI lending technology company that has been doing machine learning in credit since well before AI became a standard fintech conference track. He came to Zest from a career in data and consumer insights, with leadership roles at J.D. Power, The Harris Poll and Nielsen. Zest now touches $5.6 trillion in assets under management, and by the end of this year expects one in three credit union members to have their consumer loans decisioned with its technology.
Connect with Fintech One-on-One:
- Tweet me @PeterRenton
- Connect with me on LinkedIn
- Find previous Fintech One-on-One episodes
