What should a bank look for in AI credit decisioning software?
AI credit-decisioning software scores loan and card applications with machine-learning models and automates approve, decline and pricing decisions. Some vendors build a custom model on the lender's own loan history; others sell a decision engine the lender configures with its own models and data sources. Banks buy it on annual licences, sometimes priced per decision. The questions that separate vendors are fair-lending testing, adverse-action reasons, model documentation for validation, and the gap between vendor-reported approval gains and your own portfolio.
Observations from how these products are sold and regulated, not advice.
01
Custom model or decision engine
A custom machine-learning model trained on your loan history is a different purchase from a rules-and-models engine your team configures. The first buys modelling expertise; the second buys workflow and data orchestration.
02
Adverse-action reasons from the model
Every decline needs specific principal reasons. Ask how the vendor derives reasons from a complex model, and whether they have held up in a regulatory exam.
03
Fair-lending testing and less-discriminatory alternatives
Ask what disparate-impact testing the vendor runs, how it searches for less-discriminatory alternative models, and who owns the results. The CFPB's 2026 Regulation B rule removed disparate-impact liability under ECOA at the federal level; state laws and other regulators have not all followed.
04
Vendor-reported lift versus your baseline
Approval-rate and loss-rate improvements in case studies are vendor-reported and measured against each customer's prior model. Ask for a back-test on your own data before signing.
05
Validation package
Model documentation, performance monitoring and change logs determine how long your model-risk team needs to approve the model and each retrain.
Which rules apply to a bank using AI credit decisioning software?
| Rule | Why it applies | Documents |
|---|---|---|
| ECOA / Regulation B adverse action | A lender must give specific principal reasons for a decline within 30 days, whether a model or a person decided; 'failed our score' is expressly insufficient. | ECOA / Regulation B adverse action (15 U.S.C. 1691(d); 12 CFR 1002.9) Regulation B final rule on disparate impact (April 2026) |
| FCRA key factors | A decision based on a consumer report must name the bureau and disclose up to four key factors behind a credit score used. | FCRA adverse action and credit-score disclosures (15 U.S.C. 1681m, 1681g(f)) |
| Model risk management (SR 26-2 / OCC 2026-13 / FIL-15-2026) | Credit models are core models under the 2026 interagency guidance, vendor models included. | SR 26-2 OCC Bulletin 2026-13 FDIC FIL-15-2026 |
| Colorado ADMT Act | Colorado's 2026 law (effective 2027) requires notice and an explanation of adverse automated decisions; a lender's ECOA/Reg B notice satisfies it for the same decision. | SB 26-189 |
| EU AI Act high-risk credit scoring | For lending to people in the EU, creditworthiness assessment is a high-risk AI use under Annex III. | Regulation (EU) 2024/1689 |
Documents link to the AI Regulation Tracker. By use case: Credit scoring & underwriting · Fair lending & discrimination · Model risk management.
How do the AI credit decisioning vendors compare?
| Vendor | What it does | Deployment | Pricing model | Published price | US bank customers* | Founded · ownership |
|---|---|---|---|---|---|---|
| Zest AI | AI credit underwriting for banks and credit unions, plus fraud detection (Zest Protect) and gen-AI lending intelligence (LuLu). | SaaS | Sales-led, per-lender enterprise contracts; not published. | Not published | 2 | 2009 · Private |
| Provenir | Cloud decisioning platform for credit risk, fraud and identity, with a data marketplace, case management and AI models. | SaaS (cloud-native; API-first, real-time and batch) | Sales-led enterprise software licence (SaaS); not published. | Not published | 1 | 2004 · Private |
| FICO | FICO Platform decisioning and analytics software for origination, account management and fraud (Falcon), plus the FICO Score. | SaaS or on-premises | Software: multi-year subscriptions priced on usage (accounts, transactions or decision use cases), often with contracted minimums; not published. FICO Score for mortgages: published per-score and funded-loan fees | Not published (prices shown for FICO Score for mortgage, Mortgage Direct License Program (not the FICO Platform) on the vendor page) | 2 | 1956 · NYSE: FICO |
| Scienaptic AI | AI credit-decisioning models and platform for consumer lending, mainly US credit unions, with fraud checks and an agentic AI layer (iCUE). | SaaS, integrated with the lender's loan origination system (for example Temenos' Loan Origination Solution) | Sales-led enterprise contracts; not published. | Not published | 0 | 2014 · Private |
| Taktile | Low-code decision engine and AI-agent platform for credit underwriting, onboarding, fraud and AML decisions at lenders and insurers. | SaaS (cloud); API integration, with pre-built data connectors | Sales-led SaaS subscription; not published. | Not published | 0 | 2020 · Private, venture-backed |
| Covecta | Agentic AI for banks and lenders: pre-built agents that read documents, draft credit memos and run loan and deposit operations checks. | SaaS on AWS, hosted in the customer's chosen region; agents integrate with or embed in the bank's existing systems, with a choice of LLMs | Sales-led enterprise agreements scoped by use case (b1BANK's is described as multi-year) | Not published | 1 | 2024 · Private, venture-backed start-up |
* Banks among the 100 largest US banks on this site publicly named as customers, each sourced on the vendor's page. Prices are shown only where the vendor publishes them for the AI product itself; a price for another product (a credit score, a notary plan, contact-centre seats) is listed on the vendor's page, not here. More vendors in this category are being researched.
Private · founded 2009
Zest AI
Sells AI-automated underwriting models, priced and built per lender, to banks, credit unions and specialty lenders, and cross-sells fraud detection (Zest Protect) and a generative-AI lending-intelligence assistant (LuLu) into the same customer base.
Banks among the 100: Zions, First Hawaiian
Private · founded 2004
Provenir
Sells banks, credit unions, fintechs and non-bank lenders a configurable decisioning platform, licensed as cloud software, on which the lender runs its own credit, fraud and identity strategies and models, with third-party data supplied through Provenir's marketplace.
Banks among the 100: SoFi
NYSE: FICO · founded 1956
FICO
Sells banks two things: the FICO Score, licensed per score mostly through the credit bureaus, and decisioning software (FICO Platform, Falcon fraud, originations and account-management solutions) on multi-year subscriptions priced by accounts, transactions or use cases.
Private · founded 2014
Scienaptic AI
Sells lenders, mostly US credit unions, machine-learning credit models and a decisioning platform plugged into their loan origination system, and has organized itself as a CUSO in which credit union clients hold equity.
Banks among the 100: none named publicly yet
Private · founded 2020
Taktile
Sells banks, fintechs and insurers a low-code platform on which their own credit, fraud and compliance teams build and change automated decision flows and AI agents, with third-party data and human review built in; it does not supply a ready-made credit score.
Banks among the 100: none named publicly yet
Private · founded 2024
Covecta
Sells banks, building societies and specialist lenders a library of pre-built, domain-specific AI agents that do the manual document and analysis work around lending and deposit operations, integrated alongside existing core and origination systems rather than replacing them.
Banks among the 100: ServisFirst Bank
How much does AI underwriting software cost?
Most credit-decisioning vendors do not publish prices; contracts are annual licences, sometimes with per-decision or per-application fees, plus model build and integration. The comparison table shows a published price only where the vendor's own site lists one.
Can an AI credit model give adverse action reasons?
It must. ECOA and Regulation B require specific principal reasons for every decline, the same as for a human underwriter, and FCRA adds key-factor disclosures when a credit score is used. Vendors derive reasons with explainability methods; the lender stays responsible for their accuracy.
Do banks or credit unions use AI underwriting more?
Credit unions and community banks are the core customers of several vendors in this category. Among the 100 largest US banks, the vendor pages on this site list the banks that have been named publicly as customers.
Which AI credit decisioning vendors are compared here?
Zest AI (AI credit underwriting for banks and credit unions, plus fraud detection (Zest Protect) and gen-AI lending intelligence (LuLu)); Provenir (Cloud decisioning platform for credit risk, fraud and identity, with a data marketplace, case management and AI models); FICO (FICO Platform decisioning and analytics software for origination, account management and fraud (Falcon), plus the FICO Score); Scienaptic AI (AI credit-decisioning models and platform for consumer lending, mainly US credit unions, with fraud checks and an agentic AI layer (iCUE)); Taktile (Low-code decision engine and AI-agent platform for credit underwriting, onboarding, fraud and AML decisions at lenders and insurers); Covecta (Agentic AI for banks and lenders: pre-built agents that read documents, draft credit memos and run loan and deposit operations checks). Each has its own page with products, pricing, ownership and the banks among the 100 largest US banks that are named as customers. More vendors in this category are being added as their pages are researched and sourced.
When a bank picks an AI credit decisioning vendor, you'll read it here first.
when a bank picks an AI vendor, or a vendor ships for banks, it is in the next morning's brief · six sourced stories · 7 am ET · free
plus every tracker, bank and agent page update, the morning after · leave any morning