The future of finance in 2027: what 140 banks' AI records, the rule calendar and the payment rails show

What will finance look like in 2027?

The short answer from our records: the AI that will define 2027 is already running. The 140 bank profiles on this site (the 120 largest US banks and 20 global groups) record 1,837 named AI use cases, each with a status and a source. 1,266 of them, 69%, are in production; 228 are rolling out, 110 are pilots and 233 are announcements. A forecast for 2027 is mostly a question of how fast the last three columns move into the first.

Generative and agentic AI is now the largest category on the record: 436 entries, 268 of them in production. 99 of the 120 banks with a generative or agentic use case have at least one live. The narrower set is the one to watch for 2027: 190 entries describe AI agents or agentic AI by name, across 82 banks. 96 of those are in production at 56 banks; 94 are rolling out, piloting or announced. That pipeline is less mature than bank AI as a whole (51% live against 69%), which is what you would expect of a technology that most banks first described in 2025 and 2026.

The table counts distinct banks, each once at its strongest status, for every use case the tracker follows. It is computed from the profiles on every page load, so it moves as the profiles are updated; each bank page lists the entries with their sources. The AI in banking pillar reads the same table for 2026; here the question is which columns will grow.

Three changes stand out in the records. First, the shift from assistants to agents: in 2024 the typical entry was an employee assistant; in 2026 the new entries are agents that take a multi-step task (a mortgage file, a settlement break, a fraud claim) and stop for a human before anything leaves the bank. Second, AI governance is now a recorded activity at almost every large bank, because the April 2026 US model risk guidance put generative and agentic AI outside formal model risk and left it to broader governance (SR 26-2). Third, payments: the networks and the payment fintechs have moved further on agents than the banks have, which the payments section below shows.

What the forecasters say (forecasts, not records)

The external forecasts below are scenarios published by named firms and authorities, quoted as written with their date. None is a measurement, and they disagree on timing. McKinsey's Global Banking Annual Review (October 2025) expects AI to bring gross reductions of as much as 70% in certain cost categories but, net of rising technology costs, a 15–20% decrease in banks' aggregate cost base; it warns that bank profit pools could fall by $170 billion, or 9%, over the next decade if banks do not reposition, and that it expects "a breakout agentic business model to emerge in the next three to five years". BCG (June 2026) says agentic commerce is "agent-assisted, not agent-led" for now. Deloitte (May 2026) projects up to $75 billion of incremental revenue for top US banks by 2030 from AI-native institutional products.

What is not on the record: no bank in our profiles publishes a target number of AI agents for 2027, and none publishes an AI-attributed headcount forecast. The closest things are the dated targets in the next section, which banks set for their own programmes.

Which banks are best positioned for finance in 2027?

On the public record, the banks at the top of this site's AI disclosure index are the ones whose 2027 is easiest to read, because they have said the most. The index ranks the 120 largest US banks by percentile on independent press coverage (45% of the weight), use cases in production, a named platform, primary disclosures, dated moves in the last twelve months, disclosed numbers, named leaders and use cases described. It measures disclosure, not capability: a bank can run a large programme and say little. 103 of the 120 US banks have disclosed an AI programme and 80 name an AI platform.

What the top tier has in common is structural, and it is what to look for in any bank's 2027 plan. 10 of the top ten name a platform: a bank-owned layer that routes prompts and data to rented models, such as Bank of America's Erica, Citi's Stylus Workspaces, JPMorgan's LLM Suite or BNY's Eliza. The top ten run 12.7 use cases in production on average, against 8.3 for the other 110 US banks. A platform matters for 2027 because agents need one: an agent that reads a file, calls a system and drafts a decision needs the permissions, logging and human checkpoints a platform already enforces. The leader rankings cut the same records by question (generative AI, agents, fraud, coding).

The index covers US banks only. Among the 20 global groups profiled here, the most AI use cases in production on the record are at Commonwealth Bank (16), Barclays (13), BNP Paribas (12), BBVA (12), Standard Chartered (12). Their profiles carry the same sourced tables, and several of them set the dated 2027–2028 targets listed below, which makes them the clearest outside reference points for what a large bank's AI programme is expected to deliver by then.

What banks have committed to for 2027 and 2028

The profiles hold the dated targets banks have set themselves. The table below is filtered from every disclosed number on the 140 profiles that names 2027 or 2028 and an AI, technology or operating-model goal (deal synergies and earnings guidance are left out); each row links to the bank page that cites the source.

Read these as commitments with owners, not forecasts. Bank of America's AI expense budget for 2027 is expected to double versus 2026, RBC targets C$700 million to C$1 billion of AI enterprise value by 2027, Santander more than EUR 1 billion a year of AI value by 2028, and OCBC set 2027 targets for AI-assisted customer service and staff productivity. The pattern across them: the largest groups state a value target, regional banks state a modernisation milestone (a core or app conversion, a cloud share), and few state either for agents specifically.

Which AI and finance rules take effect by the end of 2027?

The calendar below is computed from the regulation tracker (245 documents from 42 authorities): every effective date and milestone from today to December 31, 2027, each linked to the tracker page that cites the official text. Comment deadlines are left out. 7 tracked documents have an effective date in 2027. Only dates the tracker documents appear here.

Four dates matter most for banks. December 2, 2027 is when the EU AI Act's high-risk obligations for stand-alone Annex III systems, which include creditworthiness scoring of natural persons, start to apply: the Digital Omnibus on AI, in force since July 27, 2026, moved that date from August 2, 2026 and moved the date for AI embedded in regulated products to August 2, 2028. January 1, 2027 is when Colorado's SB 26-189 takes effect: it repealed the 2024 Colorado AI Act and replaced its duty of care and impact assessments with a transparency regime, including a plain-language explanation within 30 days of an adverse outcome and no safe harbour for federally supervised banks, though an ECOA adverse-action notice satisfies the notice duty for the same decision. The same day, California's AB 1609 bars large businesses from presenting a customer-service chatbot as human and requires a route to a human agent.

Two EU dates arrive before 2027 and shape it. The recast Consumer Credit Directive applies from November 20, 2026: where a creditworthiness assessment uses automated processing, the consumer may request human intervention, a clear explanation of the assessment and its logic, and a review (CCD2). The AI Act's Article 50 transparency duties, which were not deferred, have applied since August 2, 2026, with a grace period to December 2, 2026 for marking content from systems already on the market. A lender whose credit models go through both regimes in 2027 meets the explanation duty twice, once under consumer credit law and once under the AI Act's high-risk rules from December 2027.

Outside the US and EU, Canada's OSFI Guideline E-23 applies to all federally regulated institutions from May 1, 2027, with a model definition that expressly includes AI and machine learning and an inventory requirement for vendor models. Singapore's MAS Guidelines on AI Risk Management, final on October 7, 2026, expect board oversight and an AI inventory from October 7, 2027 and lifecycle controls by October 7, 2028; they cover generative AI and AI agents. The EU Anti-Money Laundering Regulation, a directly applicable single rulebook, applies from July 10, 2027.

In the US, the model risk baseline already moved in 2026: SR 26-2, issued April 17, 2026 by the Federal Reserve, OCC and FDIC, replaced SR 11-7 with a risk-based framework most relevant to banks above $30 billion of assets, and states that generative and agentic AI models "are not within the scope of this guidance". The agencies said a request for information on AI and model risk would follow; whether it becomes guidance in 2027 is the open question for US banks. Internationally, the Financial Stability Board finalised twelve sound practices for responsible AI adoption on October 15, 2026, after a June 2026 consultation.

What will payments look like in 2027?

Payments in 2027 rest on four changes already in motion: instant rails with mandates behind them, network tokens replacing card numbers, regulated stablecoins and tokenised deposits, and AI agents that initiate payments. Each has a dated milestone in 2027, set out below with its source.

Agentic commerce is where the record moves fastest. 24 companies in our bank and fintech profiles have an agentic payment product on the record, 8 of them with at least one in production. Visa Intelligent Commerce is recorded at status "Pilot": APIs, tokenized credentials and spend controls that let AI agents discover, initiate and complete purchases. Mastercard Agent Pay is recorded at "Rolling out": Agentic Tokens and agent registration let verified AI agents pay within consumer-set limits. Visa said in December 2025 that "millions of consumers will use AI agents to complete purchases by the 2026 holiday season"; BCG said in June 2026 that commerce "remains agent-assisted, not agent-led" and that scaled agentic commerce will start in low-ticket, repeatable categories. Both can be true in 2027: agents that shop on a consumer's behalf with a token, a spending limit and a final human approval.

For banks the 2027 question is whose agent the customer uses. McKinsey's October 2025 review notes that only 4% of new US credit card applicants choose their existing bank without first exploring alternatives, down from 10% in 2018, and models a scenario in which agents move deposits to the best rate. The networks' answer is to put the bank's credential inside the agent's payment as a token the issuer can see and score. The payments hub has the rails, networks, open finance and tokenisation explainers, including real-time payments and SME cards and tokenisation.

Will fintechs and banks converge by 2027?

Yes, through charters. The largest fintechs are becoming banks, and 2027 is when several of those conversions land. Nubank received conditional OCC approval for a US national bank on January 29, 2026, with 12 months to capitalise it and 18 months to open it, which puts the opening deadline around July 2027. Revolut received conditional OCC approval on September 3, 2026 and says it remains on track for a 2027 launch, subject to FDIC, Federal Reserve and final OCC approvals. Chime agreed on September 8, 2026 to buy Stride Bank, its partner bank, with closing expected in the first half of 2027 and assets kept below $10 billion. The OCC said on August 11, 2026 that it had received 40 de novo applications in 18 months. Not every application succeeds: the OCC denied bunq's US bank charter on August 4, 2026.

The economics point the same way. BCG (June 2026) puts global fintech revenues above half a trillion dollars in 2025, up 22% year on year, says fintech is now roughly 4% of global banking and insurance revenues, up from 3% a year earlier, and finds 74% of the 85 largest public fintechs profitable, against 68% in 2024.

In our own records the convergence shows up in AI use. The 52 fintech profiles (23 neobank, 5 bnpl & lending, 10 payments, 2 card network, 5 brokerage, 7 business finance & infrastructure) record 375 AI use cases, 78% of them in production, against 69% at the banks. 24 fintechs record AI in credit underwriting, 22 in production: the move from debit to credit that the neobank explainer from debit to credit documents is an AI underwriting story as much as a licensing one. The 23 neobank profiles are summarised on the neobanks hub.

What converges less: supervision. A chartered fintech inherits the bank rulebook, including the interagency model risk guidance (SR 26-2 / OCC Bulletin 2026-13) for its models, third-party risk guidance for its vendors and the BSA/AML programme expectations that, in our enforcement data below, remain the most common source of formal actions.

What will the AI workforce in banks look like in 2027?

The hiring snapshot on this site reads the AI-titled postings on each bank's own careers site. On the latest snapshots (Sep 24, 2026 to Oct 14, 2026), 58 of the 140 banks were surveyed, 54 had AI roles open and the profiles list 324 postings (at most 12 per bank, so the count is a floor, not headcount). By seniority, 97 are leadership roles, 156 senior individual contributors and 71 other individual contributors. Postings are evidence of hiring only, not of deployment. The full data is at AI hiring.

The titles say where 2027 is going. Head of AI, chief AI officer or equivalent: 8 postings at 4 banks; Agentic AI engineers and architects: 21 postings at 12 banks; Responsible AI, AI governance and AI risk leads: 18 postings at 14 banks; Forward-deployed AI engineers: 3 postings at 3 banks; AI security and agentic-identity engineers: 26 postings at 16 banks; AI internships (2027 intake): 5 postings at 4 banks. The agentic-engineer and AI-security counts are the new ones; the internship postings recruit the 2027 intake directly.

What banks disclose about their workforce is mostly adoption, sometimes headcount. The table picks those numbers from the profiles, each as reported with its date.

The executives' own words, as recorded on the profiles: Barry Harvey (Trustmark, Jul 29, 2026): "Moving to a supported environment as opposed to a self-supported environment, it's going to allow us over time to reposition a lot of the jobs that supported our previous deposit system." Tony Schoen (First Merchants, Jun 5, 2026): "It's really about doing more with the existing staff, increasing efficiency, not eliminating jobs." Iqbal Khan (UBS, May 27, 2026): "If we can use that capacity to serve our clients better, gain more share of the wallet, grow faster, grow more, then the impact on costs and jobs is going to be less" Georges Elhedery (HSBC, May 20, 2026): "We all know generative AI will destroy certain jobs and will create new jobs" Gunjan Kedia (U.S. Bank, May 5, 2026): "The workforce likes AI about as much as they liked the return to office mandates after Covid." The forecast side is less certain. McKinsey's October 2025 central scenario has consumers approving transactions orchestrated by AI agents rather than fully delegating, and treats a scenario of drastic headcount cuts as hinging on regulatory acceptance of autonomous agents, which it calls unlikely in the medium term. Deloitte (May 2026) projects AI could raise adviser capacity in wealth management by 30% to 100%.

What do enforcement trends say about technology risk in 2027?

This site keeps a dataset of 445 formal enforcement actions against institutions by six US regulators (OCC, FDIC, CFPB, Federal Reserve, NYDFS and FinCEN) from 2016 to Oct 15, 2026, each read from the regulator's own order and classified by the failures it names. 235 actions (53%) cite failures of BSA/AML systems; 279 (63%) cite a technology failure: cybersecurity or data, third-party and vendor oversight, model risk, automated decisions, or consumer harm caused by systems. The peak year was 2024, with 74 actions.

Three readings for 2027. First, technology failures became a larger share of actions over the decade: 21 of 36 actions in 2016 against 54 of 74 in 2024, with third-party and cyber failures driving the rise, which is the risk a bank buying AI from a vendor takes on. Second, the count fell sharply after 2024 (25 actions in 2025, 14 in 2026 to date); the dataset records orders, not the reasons fewer were issued, and the rules the orders enforce did not change. Third, model risk appears in a small but steady number of orders each year, and BSA/AML systems remain the single most cited failure; an AI transaction-monitoring model that fails is a BSA/AML failure in the order, not an "AI" one.

The largest stated penalty in 2026 so far is against Bank of America, N.A. (FinCEN, Oct 13, 2026): $1.20 billion, with the order's failures classified as BSA/AML systems, model risk. That is the shape to expect when a monitoring or screening model fails inside an AML programme: the order cites the programme, and the model appears among its failures. The AI in AML page lists the banks using AI in financial-crime work and the controls they disclose.

Caveats. The classification of failures is machine-extracted from each order's text and quote-checked against it, but not yet reviewed by hand; counts by failure type are indicative. FDIC orders in the dataset start in 2018 and the FDIC source is the most fragile of the six, so recent FDIC years may be undercounted. The dataset covers formal actions against institutions only, not individuals, supervisory findings or matters requiring attention, which is where most AI model issues are raised first.

What are the leading indicators to watch before 2027?

Ten signals, each with where this site tracks it and the next known date. They are chosen because each one moves before the outcome it predicts: an agent pilot turns into production, a proposed rule into a final one, a conditional charter into an open bank.

How this page stays current: every count above is computed from the profiles, the tracker and the datasets when the page is built and on hourly revalidation, so a status change on a bank page or a new effective date in the tracker changes this page without an edit. The external forecasts are fixed quotations with their dates; when a firm publishes a new one, the table gets a new row rather than an overwritten figure. The AI in banking pillar covers 2026 as it is; the agents playbook covers how a bank should run agents.

1. Global Banking Annual Review 2025: Why precision, not heft, defines the future of banking — McKinsey & Company, Oct 23, 2025 (mckinsey.com)
2. From Recovery to Resurgence in Global Fintech — Boston Consulting Group, Jun 1, 2026 (bcg.com)
3. Deloitte Predicts Financial Services Reshaped by 2030 (2026 FSI Predictions) — Deloitte, May 20, 2026 (deloitte.com)
4. Revised Guidance on Model Risk Management (SR 26-2) — Board of Governors of the Federal Reserve System, Apr 17, 2026 (federalreserve.gov) · our page on SR 26-2
5. Regulation (EU) 2026/1744 (Digital Omnibus on AI) — European Parliament and Council (EUR-Lex), Jul 24, 2026 (eur-lex.europa.eu) · our page on Regulation (EU) 2026/1744 (Digital Omnibus on AI)
6. SB 26-189: Automated Decision-Making Technology Act — Colorado General Assembly, May 14, 2026 (leg.colorado.gov) · our page on SB 26-189
7. Guideline E-23 – Model Risk Management (2027) — Office of the Superintendent of Financial Institutions, Sep 11, 2025 (osfi-bsif.gc.ca) · our page on OSFI Guideline E-23
8. Guidelines on Artificial Intelligence Risk Management for Financial Institutions — Monetary Authority of Singapore, Oct 7, 2026 (mas.gov.sg) · our page on MAS Guidelines on Artificial Intelligence Risk Management
9. FSB finalises sound practices for the responsible adoption of artificial intelligence (AI) — Financial Stability Board, Oct 15, 2026 (fsb.org)
10. GENIUS Act Regulations: Notice of Proposed Rulemaking (OCC Bulletin 2026-3) — Office of the Comptroller of the Currency, Feb 25, 2026 (occ.treas.gov)
11. Eurosystem moving to next phase of digital euro project — European Central Bank, Oct 30, 2025 (ecb.europa.eu)
12. Digital euro pilot (ERPB technical session, agenda item 2) — European Central Bank, Apr 9, 2026 (ecb.europa.eu)
13. Next-generation monetary and financial system takes shape, based on a tokenised unified ledger — Bank for International Settlements, Jun 24, 2025 (bis.org)
14. Regulation (EU) 2024/886 as regards instant credit transfers in euro (Instant Payments Regulation) — European Parliament and Council (EUR-Lex), Mar 19, 2024 (eur-lex.europa.eu)
15. EPC Update on Payment Schemes and VOP (AMI-Pay meeting) — European Payments Council via European Central Bank, May 6, 2026 (ecb.europa.eu)
16. FedNow Service Volume and Value Statistics — Federal Reserve Financial Services, Jul 6, 2026 (frbservices.org)
17. RTP Network Volume and Value Statistics — The Clearing House, Jul 15, 2026 (theclearinghouse.org)
18. Visa Credentials Soar as Payments Hyperscaler Eyes Agentic Commerce — PYMNTS, Jan 30, 2026 (pymnts.com)
19. Mastercard Advances Europe's Checkout Transformation on the Road to 2030 — Mastercard, Jun 1, 2026 (mastercard.com)
20. Visa and Partners Complete Secure AI Transactions, Setting the Stage for Mainstream Adoption in 2026 — Visa, Dec 18, 2025 (usa.visa.com)
21. Nu Secures Approval to Establish US National Bank — Nu, Jan 29, 2026 (nu.com)
22. Revolut Receives Conditional Approval from U.S. Office of the Comptroller of the Currency to Form a National Bank — Revolut, Sep 3, 2026 (revolut.com)
23. Chime Announces Agreement to Acquire Stride Bank — Chime, Sep 8, 2026 (investors.chime.com)
24. Corporate Decision #1384: Denial of application to charter bunq US Bank, National Association — Office of the Comptroller of the Currency, Aug 4, 2026 (occ.gov)
25. OCC Commends FDIC Reform, Advances Priority to Reinvigorate De Novo Chartering (News Release 2026-67) — Office of the Comptroller of the Currency, Aug 11, 2026 (occ.gov)
26. Enforcement actions (search) — Consumer Financial Protection Bureau, Oct 15, 2026 (consumerfinance.gov)