Quick answer
AI underwriting companies fall into four groups: platforms that read documents and draft the credit analysis, loan origination systems with AI added, credit scoring models, and decision engines. Most are built for the easy file, a borrower with standard tax forms, clean statements and a deep credit bureau file. The harder question is which ones hold up on the difficult file, at a CDFI in the US, an SME lender in the Philippines or a microlender in Mexico. On that test, Kita ranks first.
Key takeaways
Why lenders are automating underwriting now
The largest unmet demand for credit sits where underwriting is hardest. The IFC puts the MSME finance gap at $5.7 trillion, or $8 trillion including informal enterprises, and says 70% of MSMEs in emerging markets lack adequate financing.[2] Nearly 80% of adults worldwide now have a financial account, yet 1.3 billion still lack access to financial services.[3] Many of those borrowers have documents and a real business but no bureau file, which is exactly the file a score-only system cannot read and a manual process cannot reach at a price that works.
Speed matters everywhere. In the Federal Reserve’s 2025 Small Business Credit Survey, 64% of firms that applied to online lenders cited speed of decision or funding as a reason for their choice, and the share applying to online lenders rose from 17% in 2020 to 29% in 2025.[1] The same hard file sits in the US: the CFPB estimates 7.0 million US consumers had no credit record at all at the end of 2020, and the lenders who serve them, CDFIs, credit unions and community banks, underwrite self-employed and small-business borrowers whose income rarely fits a standard form.[4]
Lenders have noticed. In McKinsey’s 2025 survey of credit institutions, 52% had made generative AI adoption a priority in the credit business.[5] The question has moved from whether to automate underwriting to which part of it to automate first.
The four kinds of AI underwriting company
The companies below are not interchangeable. Before comparing them, separate what each kind of product does:
- Document and financial analysis: reads borrower documents, spreads the financials and drafts the credit assessment or memo. This is where analysts spend their hours, especially on SME, commercial and thin-file lending. Kita and, for data only, Ocrolus.
- Loan origination systems with AI: a full system of record for applications, workflow and closing, with spreading and generative AI added. nCino, Abrigo and Baker Hill.
- Credit scoring models: a machine-learning model that scores each applicant, usually from bureau or device data, and auto-decides clear cases. Zest AI, Upstart and CredoLab.
- Decision engines: software where your risk team builds and runs its own rules, models and data calls. Taktile, Provenir and FinbotsAI.
Scoring models and decision engines automate the decision on data a lender already has. Analysis platforms automate the work of producing that data from a borrower’s documents in the first place. Many lenders end up using one of each.
The 11 companies
Kita ranks first; the other ten are grouped by kind rather than ranked against each other, because they solve different problems. Each entry gives what the vendor says it does, what its AI actually does, where it works, how it deploys, and published pricing and implementation times where there are any. Metrics are vendor-reported unless we say otherwise.
Document and financial analysis
Kita
Ranked first: underwrites any borrower, regardless of document format, local context or available credit data
Kita, our platform, is built for lending in challenging environments, in emerging markets and in the most complex parts of US lending: borrowers whose documents arrive as phone photos and handwritten ledgers, whose income does not fit a standard template, and who often have a thin credit bureau file or none at all. It serves microlenders, SME lenders, banks, fintechs and community lenders in Southeast Asia, Latin America and the United States. It is AI-native rather than an older system with AI added, so the work most platforms leave to your analysts, reading the documents, spreading the financials and writing the memo, is the work Kita does.
An application comes in, Kita works the file with the borrower until it is complete, reads every document in any format, spreads the financials and drafts a credit assessment calibrated to your policy, with every figure cited to the page it came from. Your underwriter reviews, edits and decides. On bank statements, our extraction reached 99.3% signal accuracy in a published 62-statement benchmark. Kita is live in production with lenders in the Philippines, Indonesia, Mexico and the United States, across consumer, microfinance and SME lending.[10,11,12]
- AI Underwriter: Spreads the financials, reads the story behind the numbers and drafts the memo calibrated to your credit policy, with every figure cited to its source.
- Kita Capture: Reads any document in any format, including phone photos, handwritten ledgers and local document types it has never seen, and flags tampering.
- AI Credit Officer: Runs the borrower back-and-forth over WhatsApp, SMS, email and chat in 30+ languages, chasing missing documents until the file is complete.
- Intelligent LOS: Runs as your full loan origination system, or embeds into HubSpot, Salesforce or the LOS you already use.
- Document Risk Score: Returns numeric risk scores from any document a borrower uploads, over API, ready to feed the scorecard or model you already run.
- Company
- Built in Silicon Valley. Backed by Y Combinator (W26) and BoxGroup, which led a $4.5M seed round in 2026.
- What the AI does
- Reads and verifies every document, completes the file with the borrower, spreads the financials and drafts a cited credit assessment. The lender’s underwriter makes every decision.
- Built for
- Microlenders, SME lenders, banks, fintechs, credit unions and CDFIs.
- Regions
- United States, Southeast Asia and Latin America. Live with lenders in the Philippines, Indonesia, Mexico and the US.
- Deployment
- Full origination system, or modules connected to the LOS, CRM and core you already run.
- Pricing
- Not published. Annual license plus per-application pricing, quoted to your volume.
- Implementation
- Configured by Kita’s team around your products and credit policy. Configuration changes ship in under 24 hours.
- Worth knowing
- Kita returns a credit assessment, not a decision: your underwriter signs every loan. It hands closed loans to your servicing system rather than servicing them. ISO 27001 certified, with SOC 2 Type II in engagement.
- Best for
- Lenders whose borrowers do not fit a template, in any market, who want a cited, auditable credit assessment on every file in minutes instead of days.
Ocrolus
Document capture, cash-flow analytics and fraud detection for US lenders
Ocrolus turns bank statements, pay stubs and tax forms into structured data, cash-flow and income analytics, and document fraud signals. It reports 400+ customers and about 750,000 credit applications a month, mostly small-business funders and mortgage lenders. Its 99%+ accuracy figure relies on human review behind the automation. In 2026 it added automated condition generation for mortgage lenders.[13,14,15]
- Cash-flow analytics: Revenue, debt capacity and income calculations from bank statements.
- Fraud detection: Flags tampered and altered documents before they reach an underwriter.
- Company
- Founded 2014, New York.
- What the AI does
- Classifies and extracts documents, analyzes cash flow and income, and detects document fraud. It does not draft a memo or make the credit decision.
- Built for
- US small-business funders, mortgage lenders and consumer and fintech lenders.
- Regions
- United States.
- Deployment
- API and dashboard alongside your LOS, with integrations including Encompass and Taktile.
- Pricing
- Not published.
- Implementation
- Not published.
- Worth knowing
- Ocrolus is a data layer, not an underwriter: lenders pair it with a decision engine or LOS and still write the analysis themselves. It holds SOC 2 Type II and ISO 27001.
- Best for
- US small-business and mortgage lenders that need bank-statement and income data to feed their own decisioning.
Loan origination system with AI
nCino
Enterprise banking platform on Salesforce, with AI agents added from 2024
nCino is a public cloud banking platform for loan origination, account opening and portfolio management, built on Salesforce, with about 2,700 customers in 25+ countries. Its Automated Spreading reads tax returns and financial statements, its Banking Advisor generative AI (generally available since June 2024) writes credit memo narratives and summaries, and its consumer lending solution auto-approves, declines or refers loans against the bank’s policy rules.
In November 2025 it announced Digital Partners, role-based AI agents, starting with an Analyst agent for credit analysis; the others are rolling out through 2026.[16,17,18,19,20,21]
- Automated Spreading: Extracts tax returns, audited and company-prepared statements into spreads.
- Banking Advisor and Digital Partners: Generative AI for memo narratives, application summaries and policy questions, sold as consumable “Intelligence Units”.
- Company
- Founded 2011, Wilmington, North Carolina. Public on Nasdaq (NCNO).
- What the AI does
- Spreading, generative memo narratives and summaries, and policy-rule auto-decisioning inside a full origination platform.
- Built for
- Banks and credit unions from community institutions to the largest US banks, plus independent mortgage banks.
- Regions
- 25+ countries, mainly the US, UK, Europe and Australia.
- Deployment
- Full platform replacement, running on Salesforce.
- Pricing
- Not published. AI is sold separately as Intelligence Units.
- Implementation
- Under six months on average for regional and community banks; more than 12 months for the largest enterprise clients, per its 10-K.
- Worth knowing
- The AI is a paid add-on to a large platform. If you are not replacing your origination system, nCino is a big project to get to it.
- Best for
- Banks and credit unions already on nCino, or replacing their whole lending stack, that want AI inside the system of record.
Abrigo
Lending, credit risk and CECL software for US community institutions, now with agentic AI
Abrigo, formerly Sageworks, sells lending, credit risk, allowance (CECL) and anti-money-laundering software to more than 2,400 US community banks and credit unions. Its loan origination system includes financial spreading, global cash flow, credit memos and auto-decisioning for loans that meet policy. Its Lending Assistant add-on (2025) drafts editable credit narratives and pulls data from unstructured financial statements and debt schedules.
In July 2026 it launched its Agentic Platform Experience, which handles document collection, data review, exceptions and quality control across the loan life, with an audit trail and a human in the loop.[22,23,24]
- Lending Assistant: Writes credit narratives and checks submitted documents against what was requested.
- Agentic Platform Experience: AI agents across pipeline, underwriting, closing and servicing; Abrigo estimates a 40%+ cut in manual work.
- Company
- Founded 1998, Raleigh, North Carolina. Formerly Sageworks.
- What the AI does
- Spreading, drafted credit narratives, document checks and policy auto-decisioning inside a full LOS, with agentic workflows added in 2026.
- Built for
- US community banks and credit unions.
- Regions
- United States.
- Deployment
- Full LOS, with AI as modular add-ons.
- Pricing
- Not published.
- Implementation
- Not published.
- Worth knowing
- The agentic platform only reached general availability in 2026, so its production track record is short.
- Best for
- US community banks and credit unions that want AI added to a familiar spreading, memo and CECL stack from one vendor.
Baker Hill
Cloud LOS for US community banks and credit unions, quick to deploy
Baker Hill sells loan origination, risk and analytics software to community banks and credit unions. In November 2025 it launched UN/FY, which it says pre-fills applications, verifies documents, scores risk continuously and decides small-business loans in seconds. Its Intelligent Documents product extracts US tax forms (1040, 1065, 1120, 1120S and schedules) into pre-filled spreads. Statement spreading, global cash flow and credit memos are standard features of the LOS.[25,26,27]
- UN/FY: Data gathering from cores and partners, document verification and fast small-business decisions.
- Six-week go-lives: Seven Midwest institutions went from contract to live lending in six weeks in 2025.
- Company
- Founded 1983, Carmel, Indiana. Private, acquired by Flexpoint Ford in 2023.
- What the AI does
- US tax-form extraction into spreads, document verification and small-business auto-decisioning. Credit memos are LOS templates rather than AI-drafted.
- Built for
- US community banks and credit unions, small-business and commercial lending.
- Regions
- United States.
- Deployment
- Full LOS.
- Pricing
- Not published.
- Implementation
- Six weeks from contract to live lending for seven Midwest institutions in 2025.
- Worth knowing
- Its document AI covers US tax forms only, and we found no generative AI memo drafting.
- Best for
- US community banks and credit unions that want a fast-to-deploy LOS for small-business and commercial lending.
Credit scoring model
Zest AI
Custom machine-learning credit models for US consumer lending
Zest AI builds client-specific machine-learning credit models from bureau data, sometimes combined with a lender’s own data, and auto-approves or declines applications against thresholds the lender sets. Middle-band files go to the lender’s rules and manual review. It serves nearly 300 lenders, mainly credit unions, across auto, credit card, HELOC, personal and small-business loans, and reported 77% bookings growth in the first half of 2026.[28,29]
- Reason codes and fair lending: Reason codes for every decision mapped to adverse action notices, automated model risk documentation, and searches for less-discriminatory model alternatives.
- LOS integrations: Native integrations with Temenos, MeridianLink (fraud) and Origence arc OS.
- Company
- Founded 2009, Burbank, California.
- What the AI does
- Builds the credit model and auto-decisions consumer applications. Does not read borrower documents, spread financials or draft memos.
- Built for
- US credit unions, community banks and specialty consumer lenders.
- Regions
- United States.
- Deployment
- Plugs into your existing LOS.
- Pricing
- Not published.
- Implementation
- As little as four weeks, per Zest AI, after a two-week proof of concept.
- Worth knowing
- Its models run on US credit bureau data, so it fits poorly where bureaus are thin, and it does not help with commercial files that need documents read and financials spread.
- Best for
- US credit unions and banks that want a fair-lending-ready model to auto-decide consumer and auto loans.
Upstart
AI lending marketplace connecting consumers to US banks and credit unions
Upstart is a public AI lending marketplace connecting consumers to more than 100 banks and credit unions that lend on its models. Its personal-loan model uses over 2,500 variables, and 91% of loans on the platform in 2025 were fully automated, with no human intervention by Upstart. Each lending partner sets and approves its own credit policy. Through its Referral Network, partners receive matched applicants, with about two-thirds approved instantly.[30,31,32]
- Referral Network and white label: Partners either receive matched borrowers or run a “powered by Upstart” program on their own site.
- Consumer products: Personal loans, auto retail and refinance, HELOCs and small-dollar loans.
- Company
- Founded 2012, San Mateo, California. Public on Nasdaq (UPST).
- What the AI does
- Consumer credit risk and pricing models, fraud checks and instant verification. No document analysis, spreading or memo drafting.
- Built for
- US banks and credit unions growing consumer lending.
- Regions
- United States.
- Deployment
- A borrower acquisition and decisioning channel alongside your existing systems, not an LOS.
- Pricing
- Not published.
- Implementation
- A digital lending program in as little as 60 days, per Upstart.
- Worth knowing
- Upstart is a marketplace as much as a tool: its top three lending partners originated 83% of its loans in 2025, per its 10-K. It is consumer-only.
- Best for
- US credit unions and community banks that want more personal, auto or HELOC volume within a credit box they set.
CredoLab
Behavioral credit scores for applicants the bureau cannot score
CredoLab builds a credit-risk score from privacy-safe device and behavioral metadata captured during a digital application, so a lender gets a score for applicants with no bureau history. It says its models are locally calibrated across 50+ countries, Indonesia is its largest market, and it has scored 200M+ people for 325+ clients.[33,34]
- SDK and marketplaces: Runs invisibly inside the lender’s app, and is sold through the FICO, Mastercard, Taktile and Provenir marketplaces.
- Privacy posture: ISO/IEC 27001 certified and collects no personally identifiable information, per CredoLab.
- Company
- Founded 2016, Singapore.
- What the AI does
- A behavioral credit score from device data. No document reading, spreading, memo or decision engine.
- Built for
- Digital consumer lenders, buy-now-pay-later providers and banks with low bureau coverage.
- Regions
- Global, strongest in Southeast Asia; also Latin America and Africa.
- Deployment
- SDK and API inside your existing application flow.
- Pricing
- Not published.
- Implementation
- Not published.
- Worth knowing
- Device-data scoring depends on app permissions: since May 2023 Google Play has barred personal-loan apps from reading photos, contacts and call logs. It does nothing for SME files.
- Best for
- Digital consumer lenders in emerging markets who need a score for applicants the bureau cannot see.
Decision engine
Taktile
Low-code decision engine for lenders with their own risk team
Taktile is a decision platform where risk teams build, test and run their own credit, fraud, onboarding and anti-money-laundering decisions, deploy their own models and plug in 30+ data providers. In September 2025 it added AI agents for small-business underwriting that extract data from PDFs, check a business’s website and summarize financial statements. Its customers include lenders in Mexico and Brazil.[35,36]
- Own your logic: Strategies, experiments and champion-challenger tests run by your team rather than a vendor.
- SMB AI agents: Document extraction, website verification and financial summaries feeding the decision flow.
- Company
- Founded 2020, Berlin and New York.
- What the AI does
- Runs the lender’s rules and models, with AI agents that extract documents and summarize financials. No proprietary score, spreads or memo.
- Built for
- Fintechs, alternative lenders, banks and insurers.
- Regions
- Global, including Mexico, Brazil, Europe and Japan.
- Deployment
- Decision layer connected by API to your existing systems and data providers.
- Pricing
- Not published.
- Implementation
- “Weeks, not months,” per Taktile.
- Worth knowing
- Taktile is a toolkit, not an out-of-the-box underwriter: you bring the credit policy, models and data, and a team to build and maintain the flows.
- Best for
- Fintechs and lenders with an in-house risk team who want to own and iterate on decision logic across credit, fraud and onboarding.
Provenir
AI risk decisioning for high-volume lenders in 60+ countries
Provenir combines a configurable rules engine, its own machine-learning credit and fraud models, simulation tools and a data marketplace of 120+ providers. It runs more than 4 billion decisions a year for banks, fintechs, auto lenders and telcos in 60+ countries, including emerging-market lenders in Latin America, Africa and Southeast Asia. In February 2026 it added a natural-language AI assistant that can automate document review.[37,38,39]
- Decisioning plus data: Rules, models, fraud and data sources in one platform, with traceable, explainable decisions.
- Emerging-market reach: Offices in Singapore, Dubai, São Paulo and Mexico City alongside New Jersey and London.
- Company
- Founded 2004, Parsippany, New Jersey. Private.
- What the AI does
- Rules, ML credit and fraud models and an AI assistant over structured and bureau data. No spreading, memo or LOS.
- Built for
- Banks, fintechs, auto lenders, buy-now-pay-later providers and telcos.
- Regions
- Global, 60+ countries, including Latin America, Africa and Southeast Asia.
- Deployment
- API-first layer on top of your existing systems.
- Pricing
- Not published.
- Implementation
- New use cases “live in 4 weeks,” per Provenir.
- Worth knowing
- Provenir decides on structured data. It does not read a business’s documents, spread its financials or write a memo, so it fits high-volume consumer and SME lending more than relationship credit.
- Best for
- High-volume lenders, including fintechs and telcos in emerging markets, that need real-time credit and fraud decisions on their existing stack.
FinbotsAI
Explainable credit scorecards built from your own data
FinbotsAI’s creditX builds application, behavior and collection scorecards from internal, external and alternative data, and runs them through a built-in business rules engine. It claims data to live credit models in 10 days.[40]
- Scorecards in days: Model building and deployment that FinbotsAI contrasts with nine to twelve month internal builds.
- Explainability: Model and decision explanations; says it completed Singapore’s AI Verify framework.
- Company
- Founded 2017, Singapore.
- What the AI does
- Builds and runs credit scorecards with a rules engine. No document reading, spreading or memo drafting.
- Built for
- Banks, microfinance institutions, fintechs and SME lenders.
- Regions
- Headquartered in Singapore.
- Deployment
- API-based layer alongside your existing systems.
- Pricing
- Not published.
- Implementation
- Data to live credit models in 10 days, per FinbotsAI.
- Worth knowing
- FinbotsAI builds the scorecard; your credit team still owns it and still does the document and analysis work on each file.
- Best for
- Banks and MFIs that want their own explainable scorecards without a long internal model build.
Side-by-side comparison
The table lines the eleven companies up against the steps of underwriting a loan, from reading the borrower’s documents to the decision, plus whether each runs a full origination system and serves lenders in emerging markets. Kita is the only company with a full mark on documents, spreading, the memo, the origination system and emerging markets together, and it serves US lenders too.
How the 11 companies compare
Based on each vendor’s own published product pages and filings. Scroll sideways on a phone.
| Company | Reads borrower documents | Spreads financials | Drafts credit memo | Credit scoring model | Auto-decision rules | Full LOS | Emerging-market lenders |
|---|---|---|---|---|---|---|---|
| KitaDocument and financial analysis | |||||||
| OcrolusDocument and financial analysis | |||||||
| nCinoLoan origination system with AI | |||||||
| AbrigoLoan origination system with AI | |||||||
| Baker HillLoan origination system with AI | |||||||
| Zest AICredit scoring model | |||||||
| UpstartCredit scoring model | |||||||
| CredoLabCredit scoring model | |||||||
| TaktileDecision engine | |||||||
| ProvenirDecision engine | |||||||
| FinbotsAIDecision engine |
Why Kita ranks first
Lending is hardest where the data is thinnest. For a CDFI in the US, that is the self-employed borrower, the cash business whose real income runs through a personal account, the immigrant entrepreneur and the first-time borrower. For an SME lender in the Philippines or a microlender in Mexico, it is applicants with no bureau file whose documents arrive as phone photos in the local language. That is where most of this list runs out of road, and where Kita was built to work. Kita is live in production with lenders in the Philippines, Indonesia, Mexico and the United States, across consumer, microfinance and SME lending.[10] Six reasons it ranks first:
Any document, in any format
Most document AI on this list is built around US paperwork. Baker Hill’s document AI covers US tax forms only, and Ocrolus is tuned to US bank statements, pay stubs and tax forms.[26,13] Kita reads whatever the borrower actually has: phone photos, scans, handwritten ledgers, local bank statements, business registrations, a self-employed borrower’s 1099s and personal statements, and layouts it has never seen before, in the borrower’s language. On bank statements it reached 99.3% signal accuracy in our published 62-statement benchmark.[11]
Local context, not a standard template
A neighborhood store’s daily cash sales, a farm’s seasonal income, a business run through the owner’s personal account, a household living on gig income and remittances: the numbers only make sense with the context. These files land on the desk of a CDFI in the US, an SME lender in the Philippines and a microlender in Mexico alike. Kita reads the story behind the numbers, reconciles figures that disagree across documents, and explains what it found in the assessment. It is configured to your loan products and credit policy in each market, not to a generic scorecard.
Whatever credit data exists
Zest AI and Upstart build on US credit bureau data, and device-data scores depend on app permissions that Google Play narrowed for loan apps in 2023.[28,30,34] Yet 1.3 billion adults worldwide still lack access to financial services,[3] and even in the US the CFPB estimates 7.0 million consumers have no credit record at all.[4] Kita builds the credit picture from the borrower’s own documents and uses bureau, bank or alternative data where it exists, so a first-time borrower gets a full assessment instead of a decline for lack of data.
The whole file, not one step
Kita is the only company in this comparison that reads the documents, spreads the financials, drafts the credit memo and runs a full origination system, and serves lenders in emerging markets as well as the US. Its AI Credit Officer completes the file with the borrower over WhatsApp, SMS and email in 30+ languages, and its AI Underwriter turns the complete file into a credit assessment.
Explainable by construction
Every figure in a Kita assessment is cited to the page it came from, and your underwriter makes every decision. That is what a credit committee, an examiner and rules such as Regulation B’s specific-reasons requirement or the EU AI Act’s high-risk obligations ask for.[6,8]
Built around your lending, and fast to change
Kita’s team configures each loan product’s policy, document checklist, borrower questions and memo format with you, and ships changes in under 24 hours. It runs as your full origination system or alongside the LOS, core, scorecard or decision engine you already have, without the year-long replacement project an enterprise platform can require.[16]
Which should you choose?
Start with where your hard files come from. If your borrowers bring messy documents, complex income and thin or missing credit files, whether you are a CDFI in the US, an SME lender in the Philippines or a microlender in Mexico, Kita is the one in this comparison that underwrites them end to end, for the reasons above.
A few lenders have a narrower need that a specialist covers:
- High-volume US consumer lending on bureau data: Zest AI for your own model, or Upstart if you also want borrower volume.
- A fintech with its own risk team: Taktile, or Provenir at larger scale and across more countries.
- Consumer lending to applicants with no bureau file: CredoLab for a behavioral score, or FinbotsAI to build your own scorecards.
- A US bank replacing its whole lending stack: nCino, Abrigo or Baker Hill.
- A US bank that only wants spreading and memos on its current LOS: Kita, which connects to the LOS you already run.
- Bank-statement and income data for US small-business or mortgage files: Ocrolus.
These are not either-or choices. Kita’s Document Risk Score returns numeric scores over API to the scorecard or decision engine you already run, and Kita connects to the LOS and core you have, so choosing it rarely means replacing what works.
How to evaluate an AI underwriting company
Six questions separate a strong demo from a strong product. On compliance, note that the CFPB withdrew its AI-specific adverse action circulars in May 2025, but the underlying rule in Regulation B still applies,[6,7] and the EU AI Act’s high-risk rules for credit scoring of individuals now apply from December 2, 2027.[8,9]
Which part of the work does it take on?
Write down where your team’s hours go today: chasing documents, reading them, spreading, writing the memo, or deciding. Then match each vendor to those steps. A scoring model will not shorten a commercial file that takes two days to spread.
Does it read your borrowers’ documents?
Most document AI is built around US tax forms and clean bank statements. Bring a real, messy file from your own pipeline to the demo, such as a phone photo of a statement, a handwritten ledger or a local business registration, and see what comes back.
Can you explain every decision?
In the US, Regulation B requires the specific principal reasons for a denial, whatever model produced it. In the EU, creditworthiness assessment of individuals is high-risk under the AI Act. Ask how the system produces reason codes, and whether every figure in a memo traces back to the page it came from.
Does it replace your stack or sit on top of it?
A full LOS replacement can take six to twelve months or more. Layer-on products go live faster but add another system. Ask for a written implementation timeline and a reference customer on your core.
Are the metrics verified?
Almost every number in this market is vendor-reported. Ask how a figure was measured, on whose files, and whether you can test it on your own applications before you sign.
Security and data residency
Ask for certifications, such as ISO 27001 or SOC 2, and where borrower data is stored and processed. Lenders outside the US often have local data rules to meet.
See Kita on your own loan files
Bring a real application from your pipeline, in any format and any language. We will show you the spread, the cited credit assessment and the audit trail, and what it takes to configure Kita to your credit policy.
Book a demoFrequently asked questions
What is the best AI underwriting company?
Kita is the best AI underwriting company for most lenders in 2026, and especially for lenders in challenging lending environments, in emerging markets and in the most complex parts of US lending such as CDFIs, credit unions and community banks. It is the only company in this comparison that reads the borrower’s documents in any format, spreads the financials, drafts a cited credit memo and runs a full origination system, and it works in emerging markets as well as the US. It underwrites the borrower regardless of document format, local context or the credit data available, and the lender’s underwriter makes every decision.
What is AI underwriting?
AI underwriting is software that does part of the credit analysis a lender’s team used to do by hand. Depending on the vendor, that means reading borrower documents, spreading financial statements, scoring the applicant with a machine-learning model, running the lender’s policy rules, or drafting the credit memo. Few vendors do all of these, so the useful question is which part of the work a given product takes on.
Does AI make the lending decision?
It depends on the product and the loan. Scoring vendors such as Zest AI and Upstart auto-approve or decline many consumer applications against thresholds the lender sets. Analysis platforms such as Kita prepare the credit assessment, the spreads and the memo, and the lender’s underwriter makes the decision. In every case the lender owns the credit policy and remains responsible for the outcome.
Is AI underwriting legal and compliant?
Yes, with the same obligations as any credit decision. In the US, Regulation B requires that a denial state specific principal reasons, whatever model produced it. The CFPB withdrew its AI-specific circulars in May 2025, but that requirement stands. In the EU, AI used to assess the creditworthiness of individuals is classed as high-risk under the AI Act, with those rules now applying from December 2, 2027. Ask any vendor how it produces reason codes and an audit trail.
What is the difference between Zest AI and Taktile?
Zest AI builds a custom machine-learning credit model for a US consumer lender and uses it to auto-decide applications. Taktile is a decision engine: your own risk team builds the rules, plugs in data providers and deploys its own models. Zest gives you a model; Taktile gives you the tools to run yours. Neither reads business documents or drafts a credit memo.
What are the alternatives to Zest AI?
Zest AI builds machine-learning credit models for US consumer lenders from credit bureau data. Lenders comparing alternatives usually look at Upstart if they also want borrower volume, Taktile or Provenir to run their own models and rules, CredoLab for applicants with no bureau file, and Kita when the work is reading borrower documents and building the credit assessment for borrowers with thin or complex files.
Which AI underwriting companies work in emerging markets?
Of the eleven companies here, Kita, Provenir, Taktile and CredoLab have published customers in Southeast Asia, Latin America, Africa or South Asia. Kita is the one that reads local documents in any format and drafts the credit assessment, so it works even when the borrower has no bureau file; the others score applicants or run decision rules on data the lender already has.
What is Kita?
Kita is an AI-native lending platform backed by Y Combinator (W26), live with lenders in the Philippines, Indonesia, Mexico and the United States. Its AI Underwriter spreads the financials and drafts a credit memo with every figure cited to its source; Kita Capture reads any borrower document in any format; the AI Credit Officer completes the file with the borrower in 30+ languages; and the Intelligent LOS runs origination end to end. Kita returns the credit assessment and your underwriter makes the decision. Kita is ISO 27001 certified.
References
Vendor details are taken from each vendor’s own website, filings and public announcements as of October 5, 2026. Vendors who spot an error can write to hello@kita.ai and we will correct it.
- [1]
2026 Report on Employer Firms: Findings from the 2025 Small Business Credit Survey (March 2026)
Federal Reserve Banks - [2]
MSMEs factsheet, IFC Financial Institutions Group (October 2025)
International Finance Corporation - [3]
Mobile phone technology powers saving surge in developing economies: Global Findex 2025 (July 2025)
World Bank - [4]
Technical Correction and Update to the CFPB’s Credit Invisibles Estimate (June 2025)
Consumer Financial Protection Bureau - [5]
Banking on gen AI in the credit business: The route to value creation (2025)
McKinsey & Company - [6]
Regulation B, 12 CFR 1002.9: Notifications
Consumer Financial Protection Bureau - [7]
Withdrawn guidance, including Circulars 2022-03 and 2023-03 (withdrawn May 12, 2025)
Consumer Financial Protection Bureau - [8]
EU AI Act, Annex III: High-risk AI systems, point 5(b)
Regulation (EU) 2024/1689 - [9]
AI Omnibus enters into force (July 2026)
European Commission - [10]
Kita raises $4.5 million to help lenders underwrite the borrowers credit bureaus miss
Kita - [11]
Bank statement extraction benchmark: 9 AI models on 62 statements
Kita - [12]
Kita is ISO 27001 certified: what it means for lenders
Kita - [13]
About Ocrolus
Ocrolus - [14]
Ocrolus accelerates automated conditioning for mortgage lenders (March 2026)
PR Newswire - [15]
Information Security Portal
Ocrolus - [16]
nCino Form 10-K, fiscal year ended January 31, 2026
U.S. Securities and Exchange Commission - [17]
nCino deploying Banking Advisor gen AI solution (June 2024)
nCino - [18]
Automated Spreading
nCino - [19]
nCino introduces Digital Partners, role-based agents (November 2025)
GlobeNewswire - [20]
nCino Consumer Banking Solution
nCino - [21]
Banks double down on nCino AI agents after using initial credits (2026)
PYMNTS - [22]
Abrigo launches agentic AI platform (July 2026)
Amazon Web Services press - [23]
Lending Assistant
Abrigo - [24]
Commercial lending software
Abrigo - [25]
Baker Hill launches UN/FY (November 2025)
Baker Hill - [26]
Intelligent Documents & Data
Baker Hill - [27]
Six weeks to better lending: Midwest banks turn to Baker Hill (August 2025)
PR Newswire - [28]
AI credit underwriting
Zest AI - [29]
Zest AI delivers record first half (August 2026)
Zest AI - [30]
Upstart Form 10-K, fiscal year 2025
U.S. Securities and Exchange Commission - [31]
Upstart announces availability of Upstart Referral Network
Upstart - [32]
Partnering with Upstart
Upstart - [33]
CredoLab: the behavioural score
CredoLab - [34]
Google Play restricts personal loan apps from accessing photos and contacts (April 2023)
TechCrunch - [35]
Introducing Taktile AI agents for SMB credit underwriting (September 2025)
Taktile - [36]
Taktile company profile
Y Combinator - [37]
AI Decisioning Platform
Provenir - [38]
Customers
Provenir - [39]
Provenir launches decision intelligence AI platform (February 2026)
Crowdfund Insider - [40]
creditX
FinbotsAI
Also read: How to underwrite thin-file and first-time borrowers, deterministic credit decisioning with LLMs and our bank statement extraction benchmark.