Artificial intelligence in investment banking has moved past the pilot stage at the largest firms. Supervisory speeches, vendor case studies and a run of surveys published across 2025 and 2026 now describe where the technology sits inside a bank, not just where it might go.
The picture that emerges is mixed. Some banks report real client engagement gains and workflow tools with hundreds of thousands of users. Others are still fighting the same data problems that stalled their first automation programs a decade ago.
The gap between the two groups is not appetite for the technology. It is governance, data quality, and what leadership does once the pilot ends.
Where the technology is deployed inside banks
More than 85% of large banks under European Central Bank supervision already use some form of artificial intelligence in investment banking, according to a speech by supervisor Pedro Machado on 24 February 2026, Technology is neutral, governance is not. The tools sit inside daily workflows now, not tucked away in innovation labs.
Machado grouped deployment into three areas: IT operations such as incident management and coding, legal and document analysis including contract review and regulatory interpretation, and front line work such as customer support and relationship management.
The scale is visible at individual firms too. JPMorganChase released its LLM Suite to employees in summer 2024, and the platform reached 200,000 onboarded users within eight months, later winning American Banker’s 2025 Innovation of the Year grand prize in the generative AI category.
Goldman Sachs has reportedly tested an internal assistant called GS AI Assistant with more than 10,000 employees, about a quarter of its global workforce, according to Fortune’s reporting on an internal memo from CIO Marco Argenti. Goldman has not confirmed the tool through any statement of its own. Much of what banks disclose about artificial intelligence in investment banking still comes from leaks and surveys, not press releases.
What the return looks like across the industry
Only 4 of 50 banks analyzed by Evident in 2025 reported realized ROI from artificial intelligence in investment banking use cases, a finding Deloitte cites in its 2026 banking and capital markets outlook, published 30 October 2025. That is a thin harvest for so much investment.
Deloitte’s separate State of AI in the Enterprise 2026 survey, fielded across 3,235 leaders in 24 countries in August and September 2025, found two thirds of organizations reporting productivity and efficiency gains. But only 20% of respondents said their AI initiatives are already growing revenue, against 74% who hope to grow revenue through AI in the future.
Achieved and hoped for are different numbers. For artificial intelligence in investment banking, that gap is the story so far.
Bank profitability has not caught up with the investment. McKinsey’s Global Banking Annual Review 2026 preview reports return on tangible equity falling from 12.4% in 2024 to 11.8% in 2025, still short of the 20% the industry touched between 2003 and 2007. The same preview finds leading banks lifting client engagement by 20 to 30 percentage points and client value by 10 to 25%. Gains concentrate in a few firms, not across the industry.
Spending has not solved the gap either. Banks have spent more on technology than the next four sectors combined, yet retail banking and brokerages sit in the bottom five of 22 US sectors on total factor productivity growth from 2010 to 2022, below the private nonfarm average, according to the same McKinsey preview.
That is a productivity comparison across US sectors only, not a global or banking specific efficiency score, and the preview does not pair it with a matching spending exhibit.
The data problem behind the numbers
Most of the shortfall in artificial intelligence in investment banking traces back to data, not algorithms. More than 90% of data users at banks reported that the data they need is often unavailable or takes too long to retrieve, according to Deloitte’s own 2024 Banking and Capital Markets Data and Analytics Market Survey, cited in its 2026 outlook.
That is self-reported survey data, not a measured system statistic, but it lines up with what Deloitte found when it reviewed the top 40 US banks directly.
That direct review, done by the Deloitte Center for Financial Services from public filings and earnings transcripts, describes “predominantly reactive, siloed efforts that yield inconsistent value.” Where a bank lacks one trustworthy view of client and market data, artificial intelligence in investment banking sits on top of the same fragmented systems that slowed automation projects for two decades. New tools, same old plumbing.
Risk and model governance
The Federal Reserve, the OCC and the FDIC moved first on model risk tied to artificial intelligence in investment banking, issuing SR 26-2 on 17 April 2026, revised guidance on model risk management that supersedes two earlier letters, SR 11-7 and SR 21-8. It applies most to banking organizations with over $30 billion in total assets.
The guidance has a gap worth knowing. A footnote states that generative and agentic AI models are novel and rapidly evolving and therefore fall outside its scope, though its principles still apply to traditional statistical models and to non-generative, non-agentic AI. Banks running artificial intelligence in investment banking for trading, pricing or credit decisions cannot yet point to one settled US rulebook for the newest model types.
The European Central Bank’s Pedro Machado described the same gap from a supervisory angle: banks need clear accountability for AI driven decisions, active senior management oversight, and challenge mechanisms involving risk, compliance and internal audit. As he put it, “AI does not dilute responsibility. If anything, it raises the bar.”
The stakes show up in the risk data too. Cybersecurity remains banks’ top near term risk at 86%, per the EY and IIF 15th Global Bank Risk Management Survey, published 24 February 2026 from a base of 101 banks in 31 countries.
Among chief risk officers, 72% say AI adoption in the risk function remains in early stages and 80% name data quality as the primary barrier, while 61% report active AI deployment in fraud and financial crime detection, the most mature use case in the survey.
On the security side, 63% of breached organizations either had no AI governance policy or were still building one, per IBM’s 2025 Cost of a Data Breach report, and 97% of those compromised had no AI access controls in place. Our guide to AI risk management in banking covers how banks are structuring these controls.
Where EU regulation stands
The EU AI Act’s toughest provisions for artificial intelligence in investment banking are already active, not pending. Prohibited practices and AI literacy requirements applied from 2 February 2025. General purpose AI obligations, governance rules and the general penalty regime applied from 2 August 2025.
Two more milestones remain. From 2 August 2026, general applicability broadens further, Article 50 transparency duties take hold, and the European Commission gains its Article 101 power to fine general purpose AI model providers directly. Annex III high risk obligations, including creditworthiness and credit scoring, were deferred to 2 December 2027 by the Digital Omnibus, Regulation (EU) 2026/1744, which entered into force on 27 July 2026.
None of this is optional for a bank operating across the EU. Credit models fall under Annex III, and the deferral buys time, not exemption, for banks building artificial intelligence in investment banking into credit decisions. Annex III does not wait for anyone.
Agentic AI arrives ahead of the rulebook
Agentic AI is the next stage of artificial intelligence in investment banking, and clients are adopting it faster than most banks can build guardrails for it. Traditional banking technologies took 5 to 10 years to reach mainstream customer use; agentic AI is on pace for 2 to 3 years, per McKinsey’s Global Banking Annual Review 2026 preview, based on a July 2025 survey of 3,945 people.
Trust in gen AI outputs ran at 77% for simple research and 69% for complex advice in that survey, with usage highest among Gen Z at 61% and Millennials at 63%.
Internal governance has not caught up with that pace. Only one in five organizations has a mature governance model for autonomous AI agents, according to Deloitte’s State of AI in the Enterprise 2026 survey.
JPMorganChase has already described an agentic roadmap for artificial intelligence in investment banking, combining generative AI with workflows so agents can carry out a series of actions toward a goal. That ambition is ahead of most regulatory frameworks, including SR 26-2, which explicitly excludes generative and agentic models from its current scope.
The workforce question
Wall Street’s own technology leaders expect the cuts to be gradual, not sudden. A Bloomberg Intelligence survey of 93 chief information and technology officers at global banks found an average expected net workforce reduction of 3%, with nearly a quarter of respondents expecting cuts of 5 to 10%, over a three to five year horizon. The cuts concentrate in back office, middle office and operations roles, the areas most exposed to automation from artificial intelligence in investment banking.
Access to the tools is spreading faster than headcount is shrinking. The same survey found worker access to AI rose by 50% in 2025. Deloitte’s State of AI in the Enterprise 2026 survey also found that insufficient worker skills are the biggest barrier to integrating AI into existing workflows, a gap that outlasts the access problem. Tools without training just sit there.
The EY and IIF survey found a similar gap inside the risk function itself: only 30% of chief risk officers expect smaller risk teams over the next three years, almost double last year’s figure of 16%, and 79% call data and AI skills essential to closing it. Reskilling, not headcount, looks like the harder problem banks face.
What separates banks that get value
Speed differs sharply from return. AI adoption inside banks is running seven times faster than digital banking adoption did, McKinsey’s Global Banking Annual Review 2026 preview found, but speed alone does not explain who wins.
Where banks put their innovation effort matters more. About 70% of banks’ AI ideas concentrate in a handful of high popularity areas, the same preview found, and McKinsey argues banks must rethink their innovation posture rather than chase the crowd. Business as usual projects in familiar areas succeed more than 90% of the time. Incubator and accelerator work succeeds about half the time, and seed and research work succeeds about 20% of the time.
That same tension between quick wins and long term platform investment shows up in commercial banking, and across the wider industry the trend lines point the same way. The banks pulling ahead treat artificial intelligence in investment banking as a portfolio, not a single bet, funding proven use cases while running smaller, higher risk experiments in parallel.
Talk to us about your AI strategy
Every bank named above started the same way: one proven use case, measured, then expanded. Start with client relationship management or regulatory reporting, wherever your data is already in reasonable shape, and measure results before the next rollout. Bring in a governance review early rather than after the first incident, since that is the step most banks in the survey data above skipped.
For a broader look at generative AI in banking, see our companion guide.
Contact Amperly if you want help turning any of this into a working plan for your bank.
