Zest AI builds machine learning underwriting models for banks and credit unions. Its software scores loan applicants using far more variables than a traditional credit score, with explainability and fair-lending documentation built in. Sold to lenders, not consumers.
Zest AI builds machine learning underwriting software for lenders. Instead of leaning on a three-digit credit score alone, its models weigh hundreds of variables from an application and a credit file to predict whether a borrower will actually repay. The pitch to a bank or credit union is simple: approve more of the applicants you're currently declining, keep losses flat, and produce documentation strong enough to hand to a regulator. The company started in Los Angeles in 2009 as ZestFinance, founded by former Google CIO Douglas Merrill, and rebranded to Zest AI when it shifted from lending money itself to selling underwriting technology.
One thing to be clear about up front: Zest AI sells to financial institutions, not to consumers. If you're a borrower hoping better AI will get you approved somewhere, there's nothing here to sign up for. Directory listings that label it free are wrong on both counts; it's neither free nor a consumer product.
The company is very much alive in 2026. It took a $200 million growth investment from Insight Partners in late 2024, made CNBC's list of the world's top fintechs two years running, and by its own count has more than 1,500 models in production.
The core product builds custom underwriting models trained on a lender's own portfolio data combined with credit bureau data, then plugs those models into the lender's loan origination system so decisions happen automatically. Around that core sit a few pieces that matter in practice:
Zest AI's sweet spot is credit unions and community or mid-size banks that want machine learning underwriting without building a data science team. It has unusually deep roots in the credit union world, and auto lenders use it heavily too. Larger banks with their own quant teams are more likely to build in-house and use vendors like this selectively.
There's no public price list. Zest AI sells through enterprise agreements, and the real cost includes a model build, integration with your origination system, and validation work before anything goes live. Plan on a months-long procurement and implementation cycle, not a self-serve trial.
We like that Zest AI concentrates on the unglamorous parts of the problem. Plenty of vendors can train a gradient-boosted model on loan data; far fewer ship the adverse action codes, fair lending reports, and model governance paperwork that let a compliance officer sleep. That focus, plus the credit union track record, is why it keeps winning deals, and the swing toward fraud detection was a sensible expansion.
The honest caveats: nearly every published lift number ('approve X percent more borrowers at flat risk') comes from Zest itself or its customers, so treat them as directional. Pricing is opaque, the integration lift is real, and when your examiner asks hard questions about the model, the accountability sits with you, not the vendor. Smaller lenders should also weigh whether their loan volume is large enough for a custom model to beat a well-tuned scorecard before committing to the spend.
