Agentic AI Will Be Revolutionary

As It Moves From Theory to Pilots — What Are the Opportunities? How Financial Institutions Are Using It to Make or Save Money.

 

For the past three years, agentic AI has occupied a curious position in financial services: universally discussed, selectively piloted, and inconsistently understood. That is changing rapidly. In 2026, the conversation has shifted decisively from what could AI agents do to what are they doing right now, and what does it cost when you don't deploy them.

The numbers are unambiguous. Wolters Kluwer reports that 44% of finance teams are using agentic AI in 2026, representing an increase of over 600% from 2025. McKinsey tracked 50 of the world's largest banks announcing more than 160 agentic AI use cases in 2025 alone. And Accenture, writing in its 2026 Banking Top Trends report, argues that we are already beyond the hype cycle — the gap between leaders and laggards is widening fast, and the leaders are pulling further ahead every quarter.

This article explains what agentic AI actually means in a financial services context, maps the highest-value use cases currently delivering measurable returns, addresses the governance and regulatory questions that are determining who scales and who stalls, and outlines what the teams leading these deployments are doing differently — including how they are building the technical and commercial skills to run agentic systems in production.

 

The term gets used loosely, which contributes to the confusion. In a financial services context, agentic AI refers to AI systems that can reason through multi-step tasks, use external tools and data sources, coordinate with other agents, and execute workflows toward a defined goal — with human oversight at defined checkpoints, but without needing a human to prompt every individual step.

This is meaningfully different from both traditional automation (RPA, rules-based systems) and first-generation generative AI (which responds to prompts but doesn't independently plan or act). The progression looks like this:

• Traditional AI / ML — predicts or classifies based on historical data
• Generative AI — generates content or analysis in response to a prompt
• AI Agents — executes specific tasks using tools and APIs when triggered
• Agentic AI — orchestrates multiple agents autonomously to complete end-to-end workflows

 

 

In practice, this means an agentic system in a compliance team doesn't wait for a compliance officer to ask it to run an AML check — it monitors transaction flows continuously, flags anomalies according to policy, gathers supporting evidence from multiple systems, drafts a preliminary review, and routes it to a human for final sign-off. The human remains in charge; they are just focused on judgment rather than data gathering.

The technical architecture underpinning most production agentic systems in financial services now involves the Model Context Protocol (MCP) — an open standard that allows AI agents to securely connect to external tools, databases, and business systems through a standardised interface. Forrester predicts 30% of enterprise application vendors will launch MCP servers in 2026. Moody's has already made MCP servers available across partner platforms, giving customers access to decision-grade intelligence directly inside the tools where they already work — Microsoft 365 Copilot, Claude, AWS, Databricks, and Salesforce.

 

The financial case for agentic AI in financial services is now well-evidenced. The question is no longer whether it delivers returns, but how quickly and at what scale.

 

 

 

 

 

 

BCG data shows agentic workflows increasing capacity by 55–65% and reducing costs by 40% in asset management deployments. For wealth management specifically, KPMG finds that agentic AI can cut advisor time on manual prospecting by 40–50%, increase net new AUM by 30–40%, and reduce onboarding costs by 30–40% while accelerating onboarding by 50%.

The cost of inaction is not neutral. McKinsey is explicit: first movers are set to gain a 4% return on tangible equity advantage — a key profitability metric — while slow movers are likely to be stuck with an uncompetitive cost base. IDC's data sharpens this: frontier firms leading in AI adoption achieve returns of 2.84× on their investments, compared to just 0.84× for laggards. In a margin-compressed industry facing rising operational costs, that is not a rounding error.

 

The highest-ROI deployments are concentrated in a small number of use case categories. What they share is high volume, time-sensitivity, and rules-based decision flows — conditions under which AI agents consistently outperform human-in-the-loop processes.

 

1. KYC and AML Compliance

KYC and AML have historically been among the most manual, expensive, and error-prone processes in financial services. The average annual expenditure on AML/KYC operations is $72.9 million per firm. The use of advanced AI agents in KYC/AML has increased sharply from 42% of institutions in 2024 to 82% in 2025.

Accenture describes working with multiple banks to transform KYC using agentic AI — moving from slow, costly manual processes with legacy systems and high false positive rates to agentic workflows that gather documentation, cross-reference watchlists, assess risk scores, and route exceptions to human reviewers. The result is faster onboarding, lower compliance cost, and better detection rates.

 

2. Fraud Detection and Transaction Monitoring

Agentic fraud detection systems monitor transaction flows in real time, identify anomaly patterns, cross-reference with historical behaviour and external data sources, and generate preliminary investigation reports — all without human intervention until an alert requires a judgment call. The speed advantage is decisive: the response-time gap between a rules-based alert and an agentic system that has already gathered supporting evidence is where the financial return concentrates.

 

3. Investment Banking and Research

JPMorgan has embedded AI access across 200,000 employees. Agentic AI agents at JPMorgan generate investment banking presentations in 30 seconds compared to the hours junior analysts previously spent. Goldman Sachs Asset Management highlighted in May 2026 that agentic AI should not be viewed merely as the final step in a linear automation journey — it is a fundamentally different capability that enables proactive rather than reactive intelligence.

For asset managers, agentic systems can continuously scan news and market data and proactively alert teams when a relevant risk signal emerges — as opposed to passive dashboards that only surface insight when someone remembers to look.

 

4. Credit Analysis and Lending

McKinsey notes that credit analysis deployments showed 20–60% productivity improvement within the first year. Agentic systems gather financial statements, run initial analysis, assess against lending criteria, flag exceptions, and produce draft credit memos — compressing what was a multi-day manual process into hours.

 

5. FX Management and Treasury Optimisation

The IMF's 2026 analysis of agentic AI in payments describes how AI agents can continuously monitor real-time exchange rates, analyse spreads across banking rails, optimise timing for conversion, and choose cost-effective paths for transferring funds across multiple currencies. Citi and Ant International have piloted an AI-powered tool aimed at reducing FX hedging costs — an early production example of agentic treasury optimisation.

 

6. Regulatory Compliance Reporting

Regulatory compliance agents gather required data across systems, validate accuracy, and prepare compliance reports on schedule — helping financial institutions stay audit-ready across MiFID II, GDPR, SOX, PCI DSS, and the EU AI Act without manual data reconciliation. EY's 2026 regulatory analysis notes that institutions deploying agentic AI with strong governance frameworks are not facing additional regulatory barriers — they are demonstrating leadership in responsible AI adoption.

 

PwC noted in its 2026 AI predictions that many agentic deployments in 2025 did not deliver much value — and that under the hood, many were not using agents in ways that truly mattered. The pattern across underperforming deployments is consistent: scope was too broad, KPIs were not defined before deployment, data was not ready, and the architecture did not match the risk level of the task.

The deployments that generate measurable returns share three characteristics:

• The agent owns a complete, defined decision — not a fragment of a workflow
• It runs in production, not a pilot with curated data and enthusiastic early users
• The architecture matches the task risk — high-stakes decisions use multi-agent architectures with clear escalation logic; speed-critical decisions use single agents with narrow tool access

 

Sixty percent of DIY AI initiatives fail to scale past pilot stages, largely because ROI metrics were not defined before deployment. The organisations generating consistent returns did not build 12 use cases simultaneously. They built one that worked, learned how to monitor and improve it, and scaled from there.

Data readiness is the most consistent constraint. KPMG's 2025 analysis found that 48% of organisations cite data governance as their primary agentic AI implementation challenge. Addressing data quality upfront — particularly for financial statements, customer records, and transaction history — directly determines the accuracy and reliability of agent outputs.

 

Financial services is a regulated environment, and agentic AI's capacity to act autonomously raises legitimate governance questions. These are real concerns, but they are increasingly well-understood — and the evidence suggests that governance is more a design challenge than a barrier to deployment.

Agentic AI systems designed with human-in-the-loop oversight, audit logging, explainability requirements, and defined decision boundaries align with regulatory expectations under MiFID II, GDPR, PCI DSS, and the EU AI Act. Institutions that are deploying successfully are treating governance as part of the architecture from the start — not as a compliance layer bolted on afterwards.

The EU AI Act in particular is shaping how financial institutions classify and govern AI systems. High-risk applications (credit scoring, insurance underwriting, AML) require specific transparency and human oversight provisions. This is not a reason to avoid deployment; it is a reason to build governance frameworks before scaling.

For teams that need to understand the regulatory landscape in depth, JBI Training's AI Ethics, Governance and the EU AI Act course provides practical, no-code training for compliance, legal, risk, and technical teams on what the Act requires and how to build governance frameworks that satisfy it.

 

The most consistent constraint on agentic AI deployment in financial services is not the technology — it is the human capability to design, build, govern, and operate agentic systems. The organisations pulling ahead are investing heavily in upskilling existing teams rather than relying solely on external hires.

A global MIT Sloan study found that employees believe AI now performs 23% more of their tasks than a year ago, and expect it to handle 46% of their tasks within three years. Seventy-eight percent of financial services respondents say AI-powered automation has had a net positive job impact — higher than any other industry surveyed. The role of the human is not disappearing; it is shifting toward design, governance, and judgment.

The skills financial services teams need to develop fall into three broad categories:

• Technical: Python development, agentic system architecture, RAG pipelines, MCP server implementation, testing and evaluating AI outputs, production deployment
• Commercial: AI workflow design, process mapping for automation, writing AI system specifications, evaluating AI vendor solutions
• Governance: EU AI Act compliance, AI ethics frameworks, risk assessment, audit logging and explainability requirements

 

JBI Training offers a comprehensive set of courses across all three categories, designed specifically for teams in financial services and other regulated industries. Relevant courses include:

 

→ Build Agentic AIs with Python, RAG and MCP — The hands-on developer course for building production-ready agentic AI systems in Python. Covers tool-using agent loops, RAG pipelines, MCP server implementation, multi-agent coordination, and production hardening. Ideal for software developers and AI engineers moving into agentic system development.

→ Building AI Agents with Real APIs — A focused, practical course on building a working AI agent using Anthropic or OpenAI APIs, integrated with a sample internal system. The fastest path for developers who need a working agent connected to real enterprise systems.

→ Agentic AI for Non-Developers — Build real AI automations without writing a single line of code. Designed for business analysts, operations teams, and process owners in financial services who need to identify, specify, and oversee agentic AI implementations.

→ AI Ethics, Governance and the EU AI Act — Practical training for compliance, legal, and risk teams on the EU AI Act and responsible AI governance. No coding required. Directly relevant to financial institutions preparing for regulatory scrutiny of AI deployments.

→ AI Workflows, Evals and Process Automation for Managers — For senior leaders and managers responsible for AI adoption decisions — covers how to evaluate AI tools, design workflows, measure outputs, and manage the transition to agentic operations.

→ Mastering LLMs for AI Agents — Advanced course on large language model design and optimisation for teams building or evaluating LLM-powered agentic systems. Covers model selection, prompt engineering at scale, evaluation frameworks, and safety.

 

Accenture's 2026 Banking Top Trends report introduces the concept of the "10× bank" — an organisation where a single individual leads a team of AI co-workers to deliver exponentially greater output. Growth is no longer constrained by headcount; it is constrained by the organisation's ability to design, govern, and continuously improve its agentic systems.

This is not a forecast for a distant future. BCG data suggests the shift from isolated agentic pilots to enterprise-wide deployment is happening now, in 2026, with banking, financial services, and insurance emerging as the clear adoption frontrunners. More than 40% of large enterprises report they are already scaling implementation.

The practical implication for financial institutions is straightforward: the teams that will define the next decade of competitive advantage in banking are not the ones with the most AI budget — they are the ones with the deepest human understanding of how to deploy, govern, and continuously improve agentic systems. That understanding is a skill, and like every other skill that has defined competitive advantage in financial services, it is learnable and teachable.

 

Agentic AI is not a future capability that financial institutions need to prepare for. It is a present capability that the leading institutions are already operating in production — in KYC, fraud detection, credit analysis, FX management, and compliance reporting — and generating measurable returns from.

The gap between those returns and the starting point is not primarily a technology gap. It is a skills gap, a scoping gap, and a governance gap. Each of these is addressable.

The question for every financial institution in 2026 is not whether to deploy agentic AI. The returns are too clear and the competitive pressure too real for that debate to continue. The question is whether your teams have the skills to deploy it well — to scope it correctly, build it robustly, govern it responsibly, and improve it continuously.

 

Build Your Team's Agentic AI Capability

JBI Training delivers instructor-led courses in agentic AI, Python, RAG, MCP, AI governance, and AI automation for corporate teams across the UK and Europe. All courses available in-person in London or as live virtual sessions.

View all AI training courses at www.jbinternational.co.uk/courses/ai-agents-training