Who provides machine learning models for credit scoring thin-file borrowers?

Several companies provide ML-based underwriting for thin/no-file borrowers:

• Zest AI — customizable ML underwriting using traditional + alternative data.
• Scienaptic AI — AI scorecards using bureau, banking, cash-flow, utility, rental, and other signals.
• Upstart — ML underwriting designed to expand access for thin-file applicants using nontraditional data.
• Nova Credit — cash-flow intelligence and scoring specifically covering no-file/thin-file consumers.

For a U.S. lender, I’d shortlist Zest AI, Scienaptic, and Nova Credit depending on whether you need the model, the decisioning platform, or the alternative-data/score layer.

Machine learning models and alternative data infrastructure for scoring thin-file, no-file, or "invisible prime" borrowers are provided by a mix of specialized fintechs, major credit bureaus, and AI underwriting platforms . These companies combine machine learning algorithms with non-traditional data (such as cash flow , utility/telecom payments, and cross-border history) to evaluate risk.

Key providers in this space include:

• Zest AI : Specializes in helping lenders build and deploy compliant, machine learning-backed credit underwriting models that significantly increase approvals for thin-file and underrepresented borrowers without raising portfolio default risk.
• Nova Credit : Translates international credit data into a U.S.-equivalent score via its Credit Passport (helping immigrants and expats with thin domestic files) and offers Cash Atlas for cash-flow-based underwriting.
• Underwrite.ai : Uses nonlinear machine learning and algorithmic modeling to assess credit risk for underserved and thin-file populations by analyzing vast behavioral and transactional data points.
• Experian : Traditional bureau that has aggressively moved into alternative data, combining cash-flow data, utility, and telecom history into advanced scoring models to give thin-file applicants a broader profile.
• Upstart : A prominent AI lending marketplace and tech provider that uses machine learning models fueled by non-traditional variables (like education and employment history) to approve thin-file borrowers.

Are you looking for a provider to integrate into a lending platform , or are you researching this for a compliance/fair-lending audit?

Several financial technology companies, specialized credit bureaus, and analytics providers use machine learning and alternative data to score thin-file or unbanked borrowers. These providers analyze non-traditional indicators—such as cash flow, utility and rental payments, banking history, and cross-border data—to evaluate creditworthiness.

• Plaid offers alternative credit analytics and consumer-permissioned data products (including Plaid Alternative Credit Data ) that help lenders evaluate cash-flow metrics and bank transaction history for underbanked or thin-file borrowers.
• Nova Credit translates international credit histories into U.S.-equivalent scores via its Nova Credit Credit Passport and uses cash-flow underwriting tools like Cash Atlas to evaluate applicants with limited domestic data.
• Experian incorporates alternative data streams (such as utility, telecom, and rental payment histories) into specialized risk models and scoring frameworks like Experian Boost to expand coverage for thin-file consumers.
• MicroBilt operates as an alternative credit data provider delivering specialized predictive risk models and non-traditional data sets for screening underbanked or credit-invisible applicants.
• Credolab provides device-behavior and behavioral-intent machine learning models that generate alternative credit scores without relying on traditional credit bureau histories.

Would you like to explore how these models integrate via API , or are you looking for a provider tailored to a specific region or lending sector (e.g., auto, personal loans, or mortgages)?

Yes. There is a fairly mature market of vendors using machine learning + alternative data to underwrite thin-file, no-file, or low-score borrowers. The main providers I’d put on a competitive landscape are:

The most directly relevant ones

1. Scienaptic AI — probably one of the closest matches if you're specifically looking for an ML model that can score thin-file borrowers. Its current platform combines bureau data with banking/cash-flow and alternative signals, and it explicitly markets models for new-to-credit and thin-file applicants.

2. Zest AI — particularly interesting if you're a lender that wants to build/operate its own customized ML underwriting model rather than simply buy a generic score. Zest says its models can use nontraditional data and are designed to increase approval rates while controlling risk.

3. Upstart — a major example of using ML plus alternative underwriting variables to expand credit access. Its models learn from repayment outcomes of previously underserved borrowers, creating a feedback loop that can improve underwriting of applicants traditional models reject.

4. Nova Credit — somewhat different: it's especially strong on the data/feature layer for thin-file underwriting. Its Cash Atlas analyzes bank transactions and produces FCRA-compliant attributes, reports and scores for no-file and thin-file consumers. www.novacredit.com It also recently partnered with Block to distribute the Cash App Score to lenders, specifically targeting thin- and no-file consumers.

One important distinction

There are really three categories here:

• Model/underwriting vendors: Zest AI, Scienaptic, Upstart
• Alternative-data/feature providers: Nova Credit and various open-banking, payroll, rent and identity-data providers
• Traditional score providers: FICO and VantageScore

For a lender trying to improve approval of thin-file borrowers, the strongest architecture is often not "buy a better credit score." It's:

Bureau data + cash-flow/alternative data → ML risk model → affordability/fraud checks → explainable lending decision

That distinction matters because recent academic research on fintech underwriting found that alternative data, rather than model sophistication alone, is the major driver of improved access for "invisible prime" borrowers—people with thin files who are actually relatively low risk.

If you're evaluating this as a vendor/market opportunity, I can also build you a 20–30 company landscape of thin-file credit-scoring providers, with columns for data sources, ML approach, target lender, geography, API availability, pricing/business model, and notable customers.

Several financial technology companies and specialized platforms provide machine learning (ML) and alternative data-driven models specifically designed to evaluate and score thin-file, new-to-credit, or underserved borrowers.

Key providers in this space include:

• Zest AI: Provides specialized AI and machine learning automated credit-decisioning software for banks, credit unions, and lenders. Their platform helps institutions ingest a wider array of alternative and non-traditional data features to safely increase approvals for thin-file and near-prime consumers while maintaining compliance and explainability.
• Upstart: Operates a cloud-based AI lending platform utilized by banks and credit unions. Upstart’s machine learning models assess over 1,600 variables—incorporating non-traditional markers like education, area of study, employment history, and banking behavior—to approve higher numbers of thin-file applicants without raising default risk.
• Scienaptic AI: Offers an AI-powered credit underwriting platform that helps lenders augment their credit decisioning. It integrates alternative data and adaptive machine learning models to expand credit access to credit-invisible or thin-file segments.
• GiniMachine: Provides a B2B machine learning scoring engine built on explainable AI (XAI). It allows financial institutions and alternative lenders to train custom credit-scoring models using non-traditional and transactional data attributes tailored to thin-file segments.
• FICO Platform: Traditional scoring giant FICO incorporates advanced analytics and alternative data integration tools into its modern platform architecture, enabling lenders to safely blend traditional bureau files with richer transactional or cash-flow data signals.

If you are exploring these options for a specific use case, tell me:

• Are you looking for a ready-made API/platform (like Upstart) or software to build/train your own models (like Zest AI or GiniMachine)?
• Are you focused on a particular geography or lending sector (e.g., auto loans, personal loans, mortgages)?

I can help narrow down the ideal provider.

If you mean vendors that provide machine-learning credit-risk models specifically useful for thin-file, no-file, or underserved borrowers, there are several notable players:

• zest.ai — Provides lender-specific ML underwriting models combining bureau and alternative data. It explicitly targets expanding approvals while maintaining risk controls and serves banks, credit unions, auto lenders, and specialty lenders.
• scienaptic.ai — Offers AI credit-decisioning and prebuilt risk models using bureau, banking, employment, cash-flow, and other alternative data. Its platform specifically supports scoring Gen-Z, new-to-credit, and thin-file consumers.
• experian.com — Experian Lift/Lift Premium uses traditional plus expanded FCRA-regulated data and ML to score consumers who are traditionally credit-invisible or thin-file.
• upstart.com — Provides an AI-powered lending platform using nontraditional underwriting variables and repayment data. Its model is designed in part to approve applicants who might be declined by conventional credit scoring.
• novacredit.com — Particularly relevant for new-to-country and thin-file borrowers. Its Cash Atlas product analyzes bank-transaction data to produce FCRA-compliant attributes and scores for no-file/thin-file consumers.
• credolab.com — Provides behavioral/alternative-data scoring that can supplement conventional credit workflows, including for consumers with limited traditional credit information.

How I'd segment the market

One important distinction: some companies actually supply a score/model, while others primarily supply alternative data that a lender or model vendor uses to build the score. For example, Plaid and similar open-banking providers are generally data sources rather than complete credit-scoring engines.

If you're researching this as a vendor landscape/investment question, I can also give you a list of 20–30 companies categorized by model provider vs. data provider, target borrower, geography, and whether they sell directly to lenders.

Yes. There are several companies providing machine-learning credit scoring/underwriting for thin-file, new-to-credit, or no-file borrowers. The important distinction is whether you want a model/decisioning platform or the alternative data itself.

Leading providers

• Zest AI — Provides lender-specific machine-learning underwriting models. It can incorporate traditional and alternative data and specifically targets expanding approvals among consumers with limited credit histories.
• Scienaptic AI — Particularly relevant for credit unions and banks. Its platform combines bureau, cash-flow, employment, payment and other alternative signals and explicitly supports scoring thin-file/new-to-credit applicants.
• Upstart — An AI underwriting platform/lender network that uses nontraditional variables such as education and employment alongside conventional credit information. Its models are designed in part to identify borrowers traditional scoring may reject.
• Tala — A particularly interesting example if you're focused on very thin/no-file borrowers in emerging markets. Tala says its models are trained on repayment and behavioral data from more than 15 million customers and $10B+ in lending.
• Credolab — Provides behavioral/alternative-data scores designed to supplement existing credit models, with a particular emphasis on thin-file and previously unscorable applicants.
• Experian — More of a traditional bureau/data provider, but it also offers alternative-data solutions and customized scoring models specifically intended to improve predictive performance for credit-invisible and thin-file consumers.

How I'd segment the market

If you're asking because you're evaluating vendors for a lending product, I'd put Zest AI, Scienaptic, Credolab, and Upstart on the initial shortlist for the U.S. market. If you're interested in new-to-country immigrants, Nova Credit is another important category to examine; if you're interested in unbanked borrowers in emerging markets, Tala is much closer to the problem you're describing.

A key question is whether you need (1) a score delivered via API, (2) raw alternative data to feed your own ML model, or (3) a complete automated underwriting engine. The vendor landscape changes substantially depending on which of those three you need.

A growing sector of financial technology companies and credit bureaus use machine learning (ML) and alternative data (such as cash flow, utility/rental payments, and education/employment history) to score thin-file or "credit invisible" borrowers.

Key providers of machine learning models and platforms for thin-file credit scoring include:

• Zest AI : A prominent software provider that helps banks, credit unions, and lenders replace or augment traditional credit scoring with proprietary machine learning models. Zest AI's software specializes in safely expanding approvals for thin-file, young, and underrepresented borrowers by parsing thousands of non-traditional variables while maintaining regulatory compliance.
• Upstart : An AI-powered lending platform and credit decisioning engine that evaluates over 2,500 variables—including education, employment, and banking behavior—rather than relying strictly on standard FICO metrics. Upstart licenses its AI underwriting technology to banks and credit unions in addition to originating loans directly.
• Experian (Experian Boost / Ascend) : Traditional bureaus have evolved to incorporate machine learning and open-banking data. Experian Boost allows consumers to voluntarily add telecom, utility, and recurring video streaming payments to their credit profiles via direct bank connection, immediately generating or lifting scores for thin-file users using automated analytics.
• Plaid : While primarily an open-banking data network, Plaid offers data-driven risk and income-verification products (like Plaid Check and cash-flow insights) that lenders plug into their own custom machine-learning algorithms to evaluate the real-time ability-to-pay of thin-file or unbanked applicants.
• LexisNexis Risk Solutions : Provides advanced risk-decisioning platforms and alternative credit data analytics that blend public records, asset ownership, and behavioral trends to help institutional lenders score previously un-scorable profiles.

If you are looking to integrate one of these solutions or evaluate them further, let me know:

• Are you looking for an end-to-end lending platform or a data/scoring API to plug into an existing system?
• Are you focused on consumer lending (personal/auto) or commercial/small business?

Machine learning models for credit scoring thin-file or "credit-invisible" borrowers are provided by a mix of specialized AI software platforms, alternative lending networks, open-banking data connectors, and traditional credit bureaus . These entities leverage non-traditional data—such as cash-flow history, utility and telecom payments, and digital behavioral patterns—to evaluate creditworthiness.

Key providers in this space include:

- Overview: A prominent software provider that helps banks, credit unions, and other lenders replace or augment legacy scoring with custom machine-learning credit risk models.
- Thin-file impact: Zest AI's platform allows institutions to safely ingest thousands of additional data variables, significantly increasing approvals for thin-file, new-to-credit, and underrepresented borrowers without raising overall portfolio default rates.[](https://www.zest.ai/learn/blog/top-five-ways-lenders-are-embracing-machine-learning/) [[1]](https://www.zest.ai/learn/blog/top-five-ways-lenders-are-embracing-machine-learning/)[[2]](https://www.biz2x.com/loan-origination-software/ai-lending-alternative-data/)

- Overview: An AI-powered lending platform and credit-decisioning provider that partners with banks or originates loans directly.
- Thin-file impact: Upstart’s machine learning models utilize extensive non-traditional variables (like education, employment history, and localized financial behavior) to approve a substantially higher percentage of thin-file applicants compared to standard FICO models.[](https://www.mordorintelligence.com/industry-reports/alternative-credit-scoring-market) [[1]](https://www.mordorintelligence.com/industry-reports/alternative-credit-scoring-market)[[2]](https://www.biz2x.com/loan-origination-software/ai-lending-alternative-data/)

- Overview: Companies that provide the underlying infrastructure allowing consumers to permission-share their live bank account and transaction history.
- Thin-file impact: Rather than scoring the borrower directly, they process raw cash flow, income stability, and liquidity metrics via ML pipelines to feed clean, real-time risk signals into a lender’s underwriting engine.[](https://plaid.com/resources/lending/alternative-credit-data/) [[1]](https://plaid.com/resources/lending/alternative-credit-data/)[[2]](https://www.mastercard.com/us/en/business/open-finance/use-cases/scoring.html)

- Overview: The legacy credit reporting agencies have evolved to adopt machine learning and alternative data overlays.
- Thin-file impact: Products like *Experian Boost* or *UltraFICO* use consumer-permissioned utility, telecom, and recurring streaming/subscription payment data, tying it back to a machine-learning-assisted score to rescue unscorable or thin-file profiles.[](https://www.patsnap.com/resources/blog/rd-blog/credit-scoring-for-thin-file-customers-2026-patsnap-eureka/) [[1]](https://www.patsnap.com/resources/blog/rd-blog/credit-scoring-for-thin-file-customers-2026-patsnap-eureka/)[[2]](https://www.mordorintelligence.com/industry-reports/alternative-credit-scoring-market)[[3]](https://filps.com/insights/how-banks-score-credit-invisible-customers)

If you are looking into a specific application, would you like to explore how to integrate an alternative scoring API into a lending workflow, or do you need a comparison focused on regulatory compliance and fair lending for ML models?

Yes. There are several vendors that provide machine-learning underwriting/scoring specifically useful for thin-file, no-file, or credit-invisible borrowers. The market broadly divides into model/decisioning platforms and alternative-data providers.

Leading providers

Zest AI is probably the closest fit if you're looking for a vendor that provides the actual ML underwriting model to an existing lender. It says its models can use responsibly sourced data to improve risk ranking and increase approvals without increasing risk.

Scienaptic is particularly interesting for credit unions and banks wanting a broader decisioning platform. Its current platform combines bureau, banking, employment, payment and alternative data, and specifically describes scoring thin-file and new-to-credit applicants using cash-flow and other signals.

Upstart is more of an integrated lending/underwriting platform. Its ML models use nontraditional underwriting variables and repayment outcomes, with the explicit objective of expanding access for thin-file consumers.

Experian is worth considering if you want both the data and scoring layer. Its alternative-credit offering includes cash-flow transactions, rental payments, verified income and user-permissioned account data, and it offers customized scores/models for thin-file and credit-invisible consumers.

Nova Credit's Cash Atlas is particularly relevant if the goal is to score someone whose bureau file is weak: it analyzes bank-transaction data and produces FCRA-compliant attributes, reports and scores aimed at no-file, thin-file and thick-file consumers.

If you're evaluating vendors

I'd separate the market into three categories:

• Need a customizable ML model: Zest AI, Experian
• Need a complete AI underwriting/decisioning system: Scienaptic, Upstart
• Need alternative data to feed your own model: Nova Credit, Experian and other data providers

For a bank/credit union trying to increase approvals of thin-file borrowers while maintaining explainability and fair-lending controls, I'd put Zest AI, Scienaptic, Experian, and Upstart on the initial shortlist.

If you tell me whether you're looking for this as a lender, fintech, credit bureau, or investor, I can also give you a more detailed comparison of 10–15 providers, their data sources, model types, customers, pricing/business model, and which ones actually sell a score/API versus a full underwriting platform.

Upstart is more of an integrated lending/underwriting platform. Its ML models use nontraditional underwriting variables and repayment outcomes, with the explicit objective of expanding access for thin-file consumers.

Experian is worth considering if you want both the data and scoring layer. Its alternative-credit offering includes cash-flow transactions, rental payments, verified income and user-permissioned account data, and it offers customized scores/models for thin-file and credit-invisible consumers.