Artificial intelligence commercial banking work now sits inside credit files, treasury dashboards and fraud alerts, not just innovation slide decks. Relationship managers use it to draft credit memos. Treasury teams use it to forecast cash positions. Fraud teams use it to catch a deepfake voice on a payment authorisation call before a wire goes out.
The figures behind that work are published, dated and attributed to source. It also covers a legal wrinkle that matters if your bank writes business loans: the EU AI Act’s creditworthiness rule was written around natural persons, not companies, and the edge of that line is sharper than most lenders assume.
None of this replaces legal advice. Where the AI Act comes up below, confirm the analysis against your own lending products with counsel before you rely on it.
You can read more on the client facing side of this in AI chatbots in banking. For how AI spots fraudulent transactions, see this overview of AI in fraud detection.
Where AI is deployed in commercial banking today
The European Central Bank’s supervisory arm put a number on this instead of guessing. In a speech dated 24 February 2026, ECB Banking Supervision’s Pedro Machado said more than 85% of the large banks it supervises already use AI in some form.
He grouped that use into three areas: IT operations such as incident management and coding, legal and document analysis such as contract review and regulatory interpretation, and front line applications such as client support, relationship management and internal knowledge tools.
That third category is where artificial intelligence commercial banking teams feel the change first. It touches a relationship manager’s daily workload, not something sitting behind the scenes in IT. Machado’s own framing was blunt: AI does not dilute responsibility. If anything, it raises the bar for the people running it.
JPMorganChase’s LLM Suite shows how fast that adoption can move inside one bank. The platform went from zero to 200,000 onboarded users within eight months of its release to eligible employees in summer 2024. It won the American Banker 2025 Innovation of the Year Grand Prize in the Generative AI category.
The bank describes its next step as agentic: generative AI paired with workflows, so agents carry out a series of actions toward a goal instead of answering one query at a time. Executives increasingly treat artificial intelligence commercial banking as a core budget line, not a passing pilot.
Credit underwriting and monitoring
Artificial intelligence commercial banking credit teams see their clearest published gain in memo drafting, not full underwriting automation. McKinsey’s research on generative AI in services describes a North American bank applying gen AI to credit risk memos. The result: credit decisions are now 30% faster, relationship manager productivity has more than doubled, and revenue per relationship manager has risen by 20%.
That is a gain in drafting the document that supports a lending decision, not evidence that AI underwrites the loan end to end. The distinction matters if a client, a regulator or your own board asks what the model decides versus what it only drafts for a person to decide.
For artificial intelligence commercial banking programmes built around credit, the memo drafting result is the case that has survived contact with a real portfolio. Monitoring, the ongoing side of underwriting, extends the same idea. A model that has structured the data behind a credit file can flag covenant drift or portfolio deterioration between formal review dates. It doesn’t have to wait for the next one on the calendar.
Treasury and cash management
Bank of America’s CashPro platform, used by more than 35,000 companies worldwide, is where artificial intelligence commercial banking treasury infrastructure shows up as a working product, not a pilot. In an 8 April 2026 release, Bank of America reported CashPro app sign-ins growing 20% year over year from full year 2024 to full year 2025. Clients approved a record $1.2 trillion in payment value through the app, an average of $38,000 every second.
CashPro’s AI features reach beyond the login screen. CashPro Capital Markets Insights uses an AI-driven algorithm that produces an AI-driven Trade Evaluation Driver score. That score gives debt issuers a read on market conditions ahead of an investment grade bond issuance.
CashPro Forecasting integrates account data and applies machine learning to analyse global cash positions and generate forecasts. CashPro Chat is a virtual assistant built on Bank of America’s AI-driven Erica technology. These features put artificial intelligence commercial banking to work for treasury teams. The job is managing cash across accounts and currencies, not offering consumer conveniences.
A separate, earlier release, dated 27 October 2025, reported a different 20% figure. Phone and email inquiries fell 20% among an unspecified group of early adopters of CashPro Chat, a cohort Bank of America does not define by size, criteria or measurement window. That figure is a distinct measure from the April sign-in figure above and not directly comparable to it.
The same October release said client use of CashPro Chat rose 21% year over year. Close to 70% of corporate clients used it for account information, transaction tracking and service resolution. Bank of America reports that figure as its own usage data, not a surveyed percentage.
It also recorded a Q3 2025 containment rate of 43%, a term the release does not define further. CashPro Search, launched in February 2023, passed 18 million total searches, including nearly 2.4 million in Q3 2025 alone, a quarterly record.
Relationship manager productivity
The same McKinsey research ties credit memo drafting directly to relationship manager output. At the bank studied, relationship manager productivity more than doubled and revenue per relationship manager rose by 20% after gen AI took over the drafting work behind each credit decision. Freeing relationship managers from memo assembly does not on its own explain the revenue increase, but McKinsey ties the two changes to the same rollout.
McKinsey’s own counterweight belongs here too. In a February 2024 survey of 150 executives at large North American and European companies, only 3% said their organisation had scaled a gen AI use case in an operations related domain. A productivity gain at one bank is not evidence of an industry wide rollout.
Artificial intelligence commercial banking productivity gains, where they are documented at all, cluster around relationship managers and credit staff, not back office processing. That matches the front line category ECB Banking Supervision flagged in its own review of supervised banks.
That concentration is worth remembering before anyone promises broad efficiency gains from artificial intelligence commercial banking rollouts elsewhere in the bank.
Payments fraud and deepfake authorisation risk
Deepfakes turn payment authorisation into the point fraud teams now watch hardest. Deloitte’s Center for Financial Services describes a case from January 2024: an employee at a Hong Kong based firm sent US$25 million to fraudsters after a video call told her to do it.
She believed she was speaking with her chief financial officer and other colleagues, but every participant on the call was a deepfake. Deloitte does not name the company involved and gives no headcount for who was on the call.
Deloitte’s Center for Financial Services also projects that gen AI enabled fraud could push US fraud losses to $40 billion by 2027, up from $12.3 billion in 2023, a compound annual growth rate of 32%. That figure is a US projection, not a global or already measured loss.
Signicat’s own platform detections show the same shift in volume from a different angle. Deepfake attacks were 0.1% of the fraud Signicat detected three years before its release, rising to about 6.5%, or 1 in 15 cases, a 2137% increase over that period. Signicat also reports that 42.5% of the fraud attempts it detects in the financial sector are now AI related. Both figures describe Signicat’s own detections, not the industry as a whole.
For artificial intelligence commercial banking payment teams, the practical response is procedural, not purely technical: a second verified channel before a large wire moves. No single detection tool catches every deepfake on the first try. That habit costs little and closes a gap no detection tool closes alone.
What the AI Act does and does not cover for business lending
Annex III of the EU AI Act lists the AI use cases classed as high risk. Point 5(b) covers systems used to evaluate the creditworthiness of natural persons or establish their credit score, with an exception for systems used to detect financial fraud. No other point in Annex III mentions credit or lending, and none of the four sub points under point 5 uses the term legal person.
On a strict textual reading, an AI system used purely to assess the creditworthiness of a legal entity, a company rather than a person, sits outside point 5(b). That is a real distinction, and it is part of why business lending is not automatically high risk under Annex III the way consumer credit scoring is.
But the dividing line is legal personhood, not business purpose. Point 5(b) has no purpose test, unlike the Consumer Credit Directive’s definition of a consumer. A sole trader borrowing in their own name is still a natural person, so that assessment stays inside 5(b) regardless of what the loan is for.
Any natural person scored as part of the same lending decision pulls the assessment back into scope: a guarantor, a director underwritten personally, an ultimate beneficial owner. Mixed use systems have to be classified per use case, not per product, so labelling a tool “business lending” does not exempt the retail slice of what it does.
Sitting outside 5(b) is not the same as sitting outside the Act. Article 5’s prohibited practices, Article 50 transparency obligations and any general purpose AI model obligations still apply regardless of Annex III classification. The Commission can also amend Annex III under Article 7, so this scope reading is not fixed for good.
Timing matters here too. Article 50 transparency applies from 2 August 2026. Annex III’s high risk obligations, including the creditworthiness point, were deferred to 2 December 2027 by the Digital Omnibus, Regulation (EU) 2026/1744, which entered into force on 27 July 2026. Prohibited practices, AI literacy duties, governance rules and the general penalty regime were already in force well before either of those dates, not newly starting on them.
The European Banking Authority’s factsheet, “AI Act: implications for the EU banking and payments sector”, published 21 November 2025, is worth reading alongside this analysis. Its three pages are built from images and could not be read as text here, and the EBA is silent on lending to legal entities in every version available. Citing the factsheet is not the same as citing an EBA endorsement of the reading above.
None of this is legal advice, and artificial intelligence commercial banking teams should confirm the scope analysis against their specific lending products with counsel before relying on it. A bank can sit outside 5(b) on one product and inside Annex III on another, for example staff scoring under point 4 or credit insurance pricing under point 5(c).
Data quality, the barrier that decides everything else
Every deployment described above depends on data the bank can retrieve when it needs it. Deloitte’s own 2024 Banking and Capital Markets Data and Analytics Market Survey found that more than 90% of data users in banks reported that the data they need is often unavailable or takes too long to retrieve. That is a self-reported survey finding, not a measured system statistic, and Deloitte gives no breakdown by job title.
The EY and IIF 15th Global Bank Risk Management Survey, fieldwork September to November 2025 across 101 banks in 31 countries, found the same barrier from the risk side. 80% of chief risk officers identify data quality and availability as the primary barrier to AI adoption, and 79% emphasise the importance of data and AI skills specifically.
Data quality is the barrier that decides whether an artificial intelligence commercial banking initiative reaches production or stays a pilot, more than the choice of model or vendor does.
The same EY survey found 72% of chief risk officers say AI adoption in the risk function remains in early stages, while 61% report active AI deployment specifically in fraud and financial crime detection. Artificial intelligence commercial banking progress is uneven even within one function at the same bank.
Governance and accountability for AI decisions
ECB Banking Supervision’s February 2026 review named three governance gaps supervised banks still need to close: clear and unambiguous accountability for AI driven decisions, effective senior management oversight given AI’s strategic importance, and robust challenge mechanisms involving risk management, compliance and internal audit.
IBM’s 2025 Cost of a Data Breach research puts a number on what happens without that accountability. Among organisations that suffered a breach, 63% either had no AI governance policy or were still developing one, and 97% of those compromised had no AI access controls in place. That same exposure runs across artificial intelligence commercial banking, where weak governance turns one breach into a systemic risk.
Deloitte’s Center for Financial Services reviewed the top 40 US banks, using public announcements and earnings transcripts. It found predominantly reactive, siloed efforts that yield inconsistent value. Evident’s 2025 AI Index, cited by Deloitte, found that only 4 of the 50 banks it analysed reported realised ROI from AI use cases in 2025. That figure belongs to Evident, not to Deloitte’s own review.
Governance is the difference between an artificial intelligence commercial banking pilot that produces a press release and one that produces realised return, and the ECB’s own supervised banks have not closed that gap yet.
What the results look like across the industry
Cybersecurity remains the top near term risk for banks generally, not only an AI specific risk: the EY and IIF survey put it at 86% of the 101 banks surveyed. Within that picture, digital fraud rose to 59%, up from 23% a year earlier, and financial crime rose separately to 43%, also up from a 23% baseline. Those are two different metrics that happen to share a prior year figure, not one metric counted twice.
Set against the risk numbers, the adoption numbers describe genuine scale. More than 85% of ECB supervised banks already use AI in some form. JPMorganChase took one internal tool from zero to 200,000 users in eight months. Bank of America’s CashPro reaches more than 35,000 companies. None of that scale, on its own, tells you whether the AI behind it is governed well.
Put these together and artificial intelligence commercial banking has real adoption, real productivity gains in specific workflows like credit memo drafting, and real gaps in governance and realised ROI, all at once. None of those findings cancels out the other two.
Artificial intelligence commercial banking buyers gain more from reading these figures together than from picking the one that flatters a vendor pitch, or from chasing a single headline statistic.
Artificial Intelligence Services
Commercial banks are under pressure to do more with the same headcount, and marketing is no exception. The three services below focus on the client facing and relationship side of the artificial intelligence commercial banking work described above.
AI generated content for corporate banking marketing
Corporate banking marketing still needs a steady stream of content: product explainers, case studies, social posts describing a new treasury feature. Generative AI can draft a first pass quickly. Your team edits for accuracy and tone instead of starting from a blank page. See our approach to AI social media posts and product descriptions.
AI personas for lending and treasury outreach
Prospecting for lending and treasury clients benefits from personas built on top of the segments a relationship team already tracks. Instead of one generic pitch, each persona gets messaging matched to its stage in the pipeline and its product interest. Read more on dynamic product descriptions using AI personas.
AI email replies for relationship manager pipelines
Relationship managers spend hours a week on email that follows a pattern: a lead nurturing sequence, a reply to a routine client question, a follow up after a call. Automating the first draft of that email frees time for the calls and meetings that build the relationship. Our guide to AI email replies and lead nurturing flow covers the mechanics.
These three services are a starting point. For the wider set of tools available, including document review and treasury reporting, see our AI services overview.
For more on how artificial intelligence commercial banking teams are improving client facing service quality specifically, read AI in banking customer service.
Talk to Amperly about AI in your commercial bank
If your bank is weighing where to start, begin with the workflow that already produces a document: a credit memo, a treasury report, a client email. That is where artificial intelligence commercial banking tools show a measurable difference fastest, based on the McKinsey and Bank of America results cited above.
Governance decisions can wait for the second project, not the first, but they should not wait past it. Build the accountability structure the ECB describes, even for a small pilot, so scaling later does not mean retrofitting oversight into several live systems at once. The artificial intelligence commercial banking figures gathered here point to the same conclusion for any lender running a first project.
Amperly helps commercial banking teams turn this into a working plan: which workflow to automate first, which vendor claims to verify before you sign, and how to write the governance layer around it. Get in touch and we will map the fastest path for your team.
