AI in corporate and commercial banking: treasury, payments, lending and onboarding.

What does AI in corporate and commercial banking mean?

On the record of the banks profiled here it means four jobs: treasury and cash management (client cash forecasting, receivables matching, knowledge assistants for the staff who serve corporate clients), payments (screening, repair and the new agent-initiated rails), commercial lending (reading loan packages, spreading statements, extracting covenants and drafting credit memos, with the decision left to a person) and client onboarding (document review, signer records, trade-finance paperwork and KYC). 81 of the 140 banks on this site record one of those, 51 in production.

Corporate banking is the part of a bank that sells to companies rather than to people: deposits and cash management, payments, lending, trade finance and the onboarding that precedes all of them. Its economics differ from retail in a way that shapes where AI lands. A corporate client file is long (a loan package of 200 or more pages at Texas Capital, a beneficial-ownership chain, a trade-finance set of invoices and bills of lading), the number of clients is small and the revenue per client is large, so the saving from a model that reads the file is measured in analyst hours per client rather than in cents per interaction. That is why the records below cluster in document reading, drafting and matching, and why the chatbots that dominate the retail page appear here only as knowledge assistants for the bank's own staff (Bank of America's AskGPS, trained on 3,200 internal documents in 29 languages for the staff who serve 40,000-plus business clients) or as client-facing chat inside a treasury portal (CashPro Chat, with a 43% containment rate).

The second difference is the decision. In retail, automatic credit decisions are routine and regulated in detail; in commercial lending, every record on this site keeps a person on the approval and uses AI to prepare it. The exception is small-business lending, where the ticket is small enough for a decision engine: Eastern Bank's Express Business Loan takes a five-minute application and decides “often instantly” up to $200,000, and Commonwealth Bank's BizExpress more than doubled auto-decisioned small-business loans between FY23 and FY26. The sibling AI credit underwriting page covers decisioning across both.

Which banks use AI in treasury and cash management?

27 records across 21 banks: Bank of America's CashPro Forecasting serves 3,000 companies and saved 250,000 hours in 2025, U.S. Bank's Liquidity Manager forecasts with Kyriba, Wells Fargo turns balance, receivable, payable, FX and liquidity data into cash-flow forecasts, and Wells Fargo, Truist, M&T and KeyBank match receivables to remittance data with AI. Huntington automates treasury connectivity across 500-plus interfaces.

The table lists every bank-profile record that describes AI in treasury or cash management for corporate clients, strongest status first and largest bank first within a status; the date is the earliest sourced event on the bank's timeline for the record. The AI in treasury management page covers the forecasting, liquidity and reconciliation models in detail, including the banks' own treasuries.

Which banks use AI in corporate payments?

37 records across 35 banks. JPMorgan's machine learning monitors roughly $10 trillion of daily payment flows and its Payments AI cut manual exceptions by more than 50%; Wells Fargo's AI sits on more than $1 trillion a day and 26 of 335 payments experiments are live; BNY's payments digital employees repair cross-border data in under 30 seconds; Fifth Third, Customers Bank, U.S. Bank, Citi and American Express have opened agent-initiated payment interfaces.

Corporate payments is where machine learning has run longest (screening) and where the newest layer is forming (agents that initiate payments under limits a client sets). The table lists both. Person-to-person and card payments for consumers are on the fraud and retail pages; the control questions that agent-initiated payments raise (audit trails, dispute processes, who is liable) are on the agentic-banking control plane.

Which banks use AI in commercial lending?

51 records across 37 banks, almost all reading and drafting rather than deciding: Texas Capital extracts covenants from 200-page loan packages, First Horizon spreads statements and starts the credit memo, DBS chains 70 to 80 agents to assemble corporate credit memos, KeyBank cut analyst research time in approvals from 60% to 25% of the process, Customers Bank closed selected loans in seven days instead of 30 to 60, Old National's reviewers are about 31% more efficient, BMO targets commercial underwriting time roughly 50% lower by fiscal 2028.

The commercial-lending records split into three generations: automated origination systems with no disclosed AI component (several regional banks name nCino projects without an AI claim; they are listed because the bank's profile records them), document intelligence that reads and spreads (Texas Capital, Axos, First Horizon, Beacon, Live Oak, Zions), and agentic credit-memo drafting (DBS's chain of 70 to 80 agents, WaFd's nCino Analyst Digital Partner pilot, RBC's commercial auto-adjudication that McKay says is “just starting to deploy”). Loan review after origination is its own niche: Old National and Seacoast both use Abrigo's Loan Review Assistant, and Seacoast says a clean line sheet that took 8 to 10 hours now takes 4 to 6.

Which banks use AI for corporate onboarding and KYC?

25 records across 19 banks: Citi cut onboarding document review by 80% in Services and US client-onboarding review from about an hour to 15 minutes; NatWest onboarded 24,000 start-ups in Q1 2026 with agentic support and had four agents live across onboarding and mandates; BNY supports 25% of onboarding with AI; Customers Bank's pilot takes commercial deposit onboarding from hours to minutes; HSBC's Smart Checking structures trade documents in three markets.

Onboarding a company means collecting its documents, identifying its signers and owners, screening them and setting up its mandates and products; AI reads and routes, people sign. The records here are the commercial ones; the identity, screening and perpetual-review steps that apply to all customers are on AI for KYC, and monitoring after onboarding on AI in AML. Trade finance appears here because its paperwork is the oldest onboarding-style reading problem in banking: HSBC's Smart Checking and Deutsche Bank's named-entity extraction for sanctions checks on scanned invoices and transport documents are the two production examples.

What results do banks report?

Citi: onboarding document review down 80%, about an hour to 15 minutes. JPMorgan: KYC unit cost down 40% since 2022. NatWest: 22,000 meeting-preparation hours saved by 1,200 Commercial Mid-Market colleagues; start-ups onboarded up 25%. BMO: 18,000-plus commercial leads in the first 12 months of its AI lead engine. KeyBank: analyst research time 60% to 25% of the approval process. Customers Bank: commercial loans closed in seven days versus 30 to 60. ConnectOne: manual lending process time down over 50% on some processes; document search from about 20 minutes to 30 seconds. SouthState: factoring invoice review 20% more accurate with 75% less employee time. Fifth Third: real-time payments throughput from 35 to 140 transactions a second.

The banks' own figures as reported, with date and source. They measure different things (time per file, hours saved, leads generated, throughput, cycle time) and are not comparable; the strongest have a stated before-state (Citi's hour to 15 minutes, Customers Bank's seven days from 30 to 60, KeyBank's 60% to 25%), the weakest are vendor-reported averages and internal projections, which are labelled as such.

How to read them: the corporate-banking numbers are mostly about analyst and operations time, which is real but is capacity rather than cost until the bank changes its staffing. NatWest's 22,000 hours across 1,200 Commercial Mid-Market colleagues is about 18 hours per person per year; BMO's target of 960 US commercial bankers by fiscal 2028, up from about 800, is a bank planning to add people and have each cover more. The quotes below are the executives describing what they expect the tools to do to the job.

Which rules apply to AI in corporate banking?

US model-risk and third-party guidance (SR 26-2, OCC 2026-13, SR 23-4) reach any machine-learning credit, forecasting or screening model and any bought platform; ECOA still owes a declined business borrower its reasons; FinCEN's deepfake alert covers business-account opening and payment instructions. In the EU the AI Act's high-risk category is credit scoring of natural persons, so commercial credit AI is governed by the EBA's loan-origination guidelines and the ECB's outsourcing guide rather than the Act's conformity regime; the Basel Committee's 2025 third-party principles reach the vendor chain.

Corporate banking has fewer AI-specific rules than retail because the laws written for automated decisions mostly protect natural persons. What remains is the prudential layer that applies to any model or vendor, the anti-money-laundering layer that applies to any account, and two places where the business customer is still protected in the US: ECOA, which reaches business credit, and the UCC and bank-regulatory expectations on payment security that the deepfake alert extends. The documents below are the ones a corporate-banking AI programme is examined against; the tracker holds 238 documents from 41 authorities.

The EU is the clearest case of the retail-corporate split. The AI Act makes credit scoring of natural persons high-risk and leaves corporate credit alone, so an EU bank's commercial credit models answer to the EBA's loan-origination guidelines (explainable, documented, human-reviewed, monitored through the loan's life) and, where the model is bought or hosted, to the ECB's outsourcing guide and DORA. In the US the 2026 model-risk revision (what changed from SR 11-7) narrowed what counts as a model and carved generative and agentic AI out, which is why most of the credit-memo and onboarding agents above sit under a bank's AI governance programme rather than its model inventory; AI governance in banking covers that programme and risks of AI in banking the ten risks it has to address.

Which vendors supply AI for commercial banking?

31 published vendors in this site's directory describe commercial lending, onboarding, treasury or business-banking products; the ones most often named by the banks here are Microsoft Copilot (16 banks), ChatGPT Enterprise (8 banks), Gemini Enterprise (6 banks), GitHub Copilot (5 banks). The largest banks build their own platforms on rented models.

The table lists the vendors in this site's AI tools directory whose products cover commercial lending, onboarding, treasury or business banking, with the number of banks in the US set that name them on the record. A bank naming a vendor is a disclosure, not a market share; the document AI, credit decisioning and banking platforms category guides explain what to check when buying.

The pattern is the one on every page of this site, with one twist. The largest banks build (CashPro, LLM Suite, Wells Fargo's payments platform that enables 85% of use cases in its payments business, BNY's Eliza); regional banks buy a platform and switch on its agents (nCino's Banking Advisor and Digital Partners at ConnectOne and WaFd, Fiserv's agentOS at First Interstate and SouthState, Abrigo at Old National and Seacoast); and the twist is that some of the suppliers are banks' fintech acquisitions, bought for the commercial client base and the software together: Valley's Bluevine, where a majority of new code is AI-generated and about 80% of inbound inquiries are AI-contained, Fifth Third's and Capital One's Brex relationships, Axos's Arc. The fintech section follows those companies on their own terms.