Zest AI Wants Your Credit Union to Say Yes More Often

At Great Basin Federal Credit Union, growth arrived with a wonderfully ordinary problem. The institution wanted to serve more places. Its membership territory had expanded beyond one Nevada county into ten additional counties, including three across a state line. More territory meant more applicants. More applicants meant more work. Hiring enough people to keep up would have made expansion rather less charming.

Great Basin adopted Zest AI. Its published customer account reports that, over two years, it processed 165% more auto-loan applications without adding team members, while maintaining its approval ratio. That is a more revealing result than a machine-learning slogan. The institution did more of the work it already wanted to do, with the people it already had.

The familiar AI question is whether a machine can replace a person. Here, the useful question is what keeps a person from getting through Tuesday's pile of applications. Zest's answer begins with risk assessment, but it reaches into staffing, policy, fraud checks and the awkward business of explaining a refusal.

A score has a compression problem

A credit score is convenient because it makes a complicated history portable. Convenience has a cost: different histories can land in the same numerical neighborhood. Zest AI builds machine-learning underwriting models tailored to a lender's borrowers and portfolios, using more information to estimate credit risk. The lender can then connect those assessments to its lending policies and existing loan-origination system.

Auto loans, credit cards, personal loans and home equity are among the advertised uses. A borrower does not go shopping for a Zest loan. A bank, credit union or specialty lender buys the technology and uses it inside its own operation. The institution remains the borrower-facing business.

Zest's proposition joins three concerns that are easy to discuss separately and difficult to solve together: identifying good risks, making decisions quickly and checking whether outcomes treat applicants fairly. The company offers bias-reduction methods, searches for less discriminatory model alternatives, explanations and ongoing model support. A prediction earns its keep when a lending team can use it consistently.

Google meets the credit department

The pairing at the beginning helps explain the product. Douglas Merrill, formerly Google's chief information officer, and Shawn Budde, a former Capital One credit executive, founded the business in 2009. Budde brought experience running subprime credit-card operations. Merrill brought the habits of a company built around extracting useful patterns from enormous quantities of information.

It began as ZestCash, both underwriting and lending. A 2013 account in the Los Angeles Business Journal described its move away from direct lending toward licensing technology to other banks in 2012. The business became ZestFinance; in 2019 it relaunched as Zest AI. The names trace a change in the customer: from the person seeking money to the institution deciding whom to lend to.

There is no need to invent a dramatic collapse to explain the evolution. The documented shift is consequential enough. A company that lends must make its own loans work. A company that licenses underwriting must make its methods useful to other lenders, with their own policies, systems and obligations. Distribution becomes part of the engineering problem.

The convincing number is a before-and-after

Consider Community Choice Credit Union. After using Zest in its auto, personal-loan and credit-card portfolios, it expanded into home equity. In its first 90 days with that underwriting solution, the credit union reported moving from no automated home-equity decisions to 57% automation. More than 300 borrowers received approval on the day they applied.

Washington State Employees Credit Union faced mounting application volumes as it expanded direct and indirect auto lending. Its published case study reports automation of 70% of direct-auto decisions and a 19% average reduction in delinquencies. Explainable reasons for declined applications were part of its selection criteria. Faster approvals alone would have left half the job unfinished.

Idaho Central Credit Union supplies another useful wrinkle. It reported moving from roughly 50% automated decisions to 75%. The starting point was already substantial automation. This was a refinement of an existing operation, with its team putting considerable effort into compliance at the beginning.

These are customer accounts published by the vendor. They describe different portfolios and measurement periods; they should not be combined into a universal return. Their practical lesson is narrower and more useful: specify the bottleneck, record the starting point, and measure the portfolio after the new process goes live.

Fraud is a different question

Someone can appear creditworthy and still be submitting a fraudulent application. Zest Protect, launched in 2024, addresses that separate problem. It assesses application-fraud risks including compromised identity, fraudulent behavior and discrepancies in reported income. Lenders can adjust detection thresholds and receive reason codes, with checks integrated into the origination process.

Wauna Federal Credit Union's account describes adding Protect after adopting Zest underwriting. The aim was to distinguish suspicious applications from legitimate members without sending every file through another obstacle course. Its experience illustrates why fraud screening and credit prediction belong in the same workflow, even though they answer different questions.

This is where the product suite starts to make commercial sense. An underwriting model can recommend confidence; an unresolved identity check can still stop the application. Removing one delay exposes the next. Selling the adjacent tool gives Zest another product to sell and gives the lender a chance to preserve the automation it bought in the first place.

LuLu gets a seat at the meeting

The intelligence products tackle what happens around and after a decision. Zest's reporting covers portfolio performance, applicant outcomes and marketing opportunities. A lender can examine credit migration, pricing and the people who applied but never booked a loan. This is information for running a lending business, beyond deciding a single application.

LuLu adds a natural-language interface. Pulse draws on public financial and economic datasets, including bank and credit-union call reports and mortgage-disclosure data, to help users compare peers and explore market conditions. Strategy, launched in May 2025 initially for MeridianLink customers, brings institution-specific analysis and policy simulations into the conversation.

A team can test a proposed policy change and estimate what it might do to approvals, automation and risk. Estimates require judgment: Strategy's product page specifies that unfunded-loan analysis uses reject inference, and that the tool may not be used to make decisions on loan applications. Asking a useful question and authorizing a loan remain different activities.

“It almost feels like we're cheating. We have such a leg up, and I love it.”Jennifer Denoo, CEO, Great Basin Federal Credit Union, on LuLu

The customers bought into the company

Zest sells enterprise software and supporting expertise. Its route to buyers includes integrations and partnerships with loan-origination providers such as MeridianLink and Origence. This places it inside the systems financial institutions already use, rather than requiring every customer to construct a new lending operation.

The capital behind that strategy is substantial. Insight Partners invested $15 million in 2020. A growth round of more than $50 million followed in 2022, co-led by Insight and CMFG Ventures. In December 2024, Insight made a $200 million growth investment, with fraud protection and generative AI among the expansion priorities.

Then, in November 2025, five customers led another financing: SchoolsFirst, Members 1st, ORNL and Truliant credit unions, alongside Citi through Citi Ventures. A purchase order says a product cleared procurement. An equity investment says the customer also wants a stake in the supplier's future. It is a telling relationship, though hardly a substitute for evaluating the software.

For a prospective buyer, the useful cost exercise extends beyond a software contract. Estimate the work of integration, validation, training and continuing oversight against the value of additional funded loans and staff time released. A faster decision has economic value only when the resulting portfolio performs acceptably.

Small institutions, large expectations

Zest competes with other AI decisioning providers, including Scienaptic, as well as an institution's existing score-based rules and internal modeling. Upstart occupies an adjacent part of the market with an AI lending marketplace. Zest's particular combination is tailored underwriting, fairness tools, operational support and a growing set of products around the decision.

Its work with Commonwealth Credit Union makes the small-institution ambition concrete. In February 2026, they announced CU Lending Collective, a credit union service organization designed to bring scoring models and operational guidance to smaller credit unions. Hosted deployment, monitoring and model documentation are part of the offering. The obstacle is as much the capacity to run the technology as the capacity to buy it.

There is something worth copying here even without buying Zest: expand one portfolio, measure its behavior, and let the evidence justify the next step. Community Choice did that before moving into home equity. A lender with unreliable data, an untested policy or nobody responsible for monitoring needs to repair those foundations before increasing automated decisions. More speed would otherwise make its mistakes arrive sooner.

The company reported 77% growth in first-half 2026 bookings and more than 1,500 active models by August. Bookings are sales commitments, not reported revenue. Its customers represent trillions in assets; those assets belong to the institutions. The precise scale matters less than the underlying wager: that a community lender can use better evidence without surrendering its local purpose.

The borrower never needs to know the software's name. Getting a considered answer while the car is still available, or while a home-improvement job is still possible, is quite enough.