Deploy Gen AI
When It Matters

Move from experimentation
to impact

Financial institutions are experimenting with agentic AI, but few are running production workloads at scale. The barriers are specific to regulated environments: agents that don't produce the same output every time given the same inputs, making decisions hard to explain and audit; the need for real-time observability and cost controls; governance requirements for autonomous decision-making; and the complexity of integrating agents with legacy systems.

Move from experimentation to impact

Why AWS

Amazon Bedrock AgentCore provides enterprise-grade runtime, governance, observability, and safety controls purpose-built for regulated industries
Flexibility to use different frameworks (Strands, LangGraph) and models – not locked into a
single vendor
Real-time policy controls that govern how agents interact with tools and data, with full audit trails
AgentCore Payments enables autonomous agent transactions with deterministic spending limits enforced at the infrastructure layer
Bridges the POC-to-production gap: same infrastructure for prototyping and production, reducing the proportion of AI projects scrapped before going live

"When you think about agentic AI use cases in financial services, 99% accuracy is not good enough. They have to be as close to 100% accuracy as possible. And you need the transparency, the lineage, the auditability built out so you're able to trace back exactly how an agent gave the answer it did and where it got the source data from."

Alvin Huang, Head of Agentic and Generative AI,
Market Development, AWS

Making AI work at scale

Why fragmentation is the biggest barrier to agentic AI
Good data, wrong structure – and what to do about it

Where agentic AI enters production and what's next
The sequencing decisions that turn early pilots into enterprise programs

The AWS partnerships powering today's AI leadership
Ten years in the making and
still accelerating

Industry perspectives on what it takes to move generative and agentic AI from experimentation to scale production. Uncover the detail below.

Innovation across the trade lifecycle with generative AI
Generative AI is accelerating research, trading, and operations – cutting analysis time by
up to 80%

Banking on the cloud
How agentic AI is letting banks automate and serve customer segments that were never economically viable before

Insurance reimagined in
the cloud
Agentic AI is reshaping insurance across claims, underwriting, and customer engagement

A primer on AWS Agentic AI
in financial services
How institutions are putting agentic AI to work across CX, fraud, compliance, and ops

Highlights from the 2026 AWS Financial Services Symposium
How the Symposium put agentic AI center stage

How agentic commerce is driving the next era of financial services
Stablecoins, tokenization, and a new ecosystem – at convergence

How agents are now completing transactions end-to-end
From reasoning to checkout, with minimal human intervention

How AI is transforming the capital markets trade lifecycle
Every stage of the trade lifecycle getting smarter

Making AI work at scale

Industry perspectives on what it takes to move generative and agentic AI from experimentation to scale production. Uncover the detail below.

How agentic commerce is driving the next era of financial services
Stablecoins, tokenization, and a new ecosystem – at convergence

How agents are now completing transactions end-to-end
From reasoning to checkout, with minimal human intervention

How AI is transforming the capital markets trade lifecycle
Every stage of the trade lifecycle getting smarter

Making AI work at scale

Industry perspectives on what it takes to move generative and agentic AI from experimentation to scale production. Uncover the detail below.

"Rethink Everything," the AWS Financial Services Symposium, put generative and agentic AI center stage as a production reality reshaping banking, capital markets, insurance, and payments right now. From autonomous fraud detection to agentic commerce, the sessions mapped what it takes to move AI from experimentation to enterprise scale – and what stands in the way.

Rethink everything: Highlights from the 2026 AWS Financial Services Symposium

"Implementing AI across the enterprise isn't a purely technical effort. It includes people and process changes that are fundamental to how companies operate. You can't build tomorrow's applications and products on yesterday's dogma and legacy technology."
Scott Mullins, Managing Director, Worldwide Financial Services, AWS

Key takeaways

Governance and safety are prerequisites – scaling agents requires purpose-built controls from day one
Modernization unlocks AI – legacy migration creates the foundation for capabilities yesterday's architecture can’t support
A shared data foundation across the business powers everything
AgentCore Payments is the first managed payment capability purpose-built for autonomous agents

Why it matters

Moving from proof of concept to production remains the industry's defining challenge. The top blockers are integration complexity, AI-ready data, cross-functional alignment, and centralized governance. The institutions closing that gap fastest are those treating AI as a business transformation program – with governance, infrastructure, and culture to match.

Read more

Traditional banking automation worked for structured, repetitive tasks but struggled with anything requiring interpretation, nuance, or judgement. Agentic AI changes that. This report explores how banks are using AI agents to automate complex human reasoning in loan origination and servicing, deliver memory-driven personalization at scale, make thin-file borrowers economically viable to underwrite, and orchestrate payments autonomously.

Banking on the cloud

"We're building a banking model that brings sophistication to everyone, not just the privileged few. One that handles nuanced edge cases as efficiently as routine transactions around what customers want to achieve rather than products. The future is here." Charith Mendis, Head, Banking Market Development, AWS

Key takeaways

Rocket Mortgage deployed eight specialized service agents, cutting customer care transfers by 85%, servicing specialist transfers by 45%, and increasing lead conversion by 33%
72% of consumers say personalization influences where they choose to bank – memory-driven agentic ecosystems are how banks are closing that gap
Agentic AI is making thin-file borrowers viable to serve, cutting decisions from days to minutes

Why it matters

Agentic AI lets banks automate what rules-based systems never could: work that requires interpretation, nuance, and adaptation. The result is that banks can serve more customers, more profitably, with greater specialization – including segments like thin-file borrowers and gig workers that were previously too costly to underwrite.

Agentic AI is giving insurers the ability to automate work that previously required specialized human judgment – from orchestrating income verification for disability claims to analyzing medical records and detecting fraud in weather-related submissions. This playbook explores five priorities shaping the future of insurance, with generative and agentic AI playing a central role in claims, underwriting, distribution, and policyholder services.

Insurance re-imagined in the cloud

"AWS offers us a safe and open environment to continue innovating and has also opened doors to the exploration of emerging technologies as promising as generative AI." Manuel Rodriguez, CIO, Agroseguro

Key takeaways

Agentic AI is automating claims processing, underwriting, and fraud detection – work that previously required experienced professionals to review manually
Generative AI is cutting manual document processing effort by 30%+ and shortening call center operator training time by a similar margin
Multimodal AI can now process documents, images, and historical data simultaneously, enabling next-generation quoting and risk assessment
Insurers are deploying generative AI at the point of customer interaction

Why it matters

The "Great Retirement" is hitting insurance hard, and the industry needs to do more with fewer people. Agentic AI lets insurers automate specialized work in claims, underwriting, and servicing that previously required manual review by experienced professionals. The result: faster processing, lower costs, and better customer experience without growing headcount.

Human-only processes in financial services can no longer keep pace with customers demanding instant decisions, fraud evolving faster than detection systems can adapt, and operational complexity lowering margins. Agentic AI – systems that reason, plan, and act across multi-step processes with limited human involvement – is how leading institutions are responding. This primer covers use cases, real customer results, and the AWS capabilities behind them across banking, insurance, capital markets, and payments.

A primer on AWS agentic AI in financial services

"Financial institutions have moved their data to AWS and set up processes and guard rails that enable them to securely and confidently take advantage of machine learning and generative AI." Scott Mullins, Managing Director of Financial Services, AWS

Key takeaways

The agentic AI market is projected to grow from $7.6bn in 2025 to $50bn by 2030
By 2028, 33% of enterprise software applications will include agentic AI, up from less than 1% in 2024
AI agents are already in production across claims processing, fraud investigation, underwriting, merchant onboarding, trading operations, and compliance workflows
Amazon Bedrock AgentCore provides the runtime, gateway, memory, identity, and observability primitives to move agents from pilot to production securely

Why it matters

The conversation has moved from whether to adopt agentic AI to how quickly to deploy it. McKinsey research shows AI agent workflows in fraud detection alone could generate 200 to 2,000% productivity gains, with a single team member able to supervise more than 20 AI agents. In compliance, where banks assign 10-15% of full-time equivalents to KYC/AML alone, autonomous agents offer a scalable path forward.

Capital markets firms are moving beyond experimentation and deploying generative and agentic AI across production workflows. From pre-trade research and insight generation, through execution and decision-making, to post-trade settlement, surveillance, and reporting – AI is being embedded at every stage of the trade lifecycle. This ebook explores how AWS customers are using generative AI to enhance efficiency, augment human decision-making, and accelerate innovation across capital markets.

Innovation across the trade lifecycle with generative AI

"Tasks that previously took days or weeks are now completed in hours."
Balyasny Asset Management, on their agentic deep research solution built on AWS

Key takeaways

Generative AI is in production across all three phases of the trade lifecycle
Research timelines compressed from weeks to hours; routine trade analysis cut by 80%; post-trade reconciliation reduced by 90%
Agentic systems are handling multi-step processes with minimal human intervention
Firms are scaling post-trade ops without proportional headcount growth

Why it matters

Generative AI is viable across more production workloads as models improve in accuracy and cost per token declines. Agentic AI can now reason over time and act across systems. Meanwhile, rising trade volumes, 24/7 markets, and T+1 settlement are squeezing operations in ways manual processes can't absorb. Firms deploying generative AI across the full trade lifecycle are building a durable edge in speed, intelligence, and operational scale.

HCLTech and Fiserv: Agentic AI-powered mainframe modernization – without the multi-year timeline

Infosys: AI-first autonomous banking. Redefining the operating model for finance

NVIDIA: Accelerate trading with AI factories for smarter decisions

Pegasystems: Driving real results in financial services through predictable AI and automation

Vertical Relevance: How Numerix is leveraging AWS to build a scalable AI platform and accelerate product development

Deloitte: How specialized AI tools are transforming insurance by streamlining complex workflows

TCS: A guide to transitioning AI agents from initial concepts to full-scale enterprise deployments

Partner insights

AWS partners are at the forefront of financial services innovation, building solutions that help institutions modernize faster, deploy AI at scale, and stay ahead of an industry that never stands still. Explore what they're working on here.

Rocket: AI agents that help everyone home
Rocket AI agent users are 3x more likely to close a loan, with fewer customer care transfers

MUFG: Accelerating digital transformation
What used to take days now takes minutes across one of the largest financial institutions

Hazel by Altruist: Purpose-built AI for wealth management
Purpose-built AI that gives wealth advisors back the hours that matter most

Nasdaq: Operationalizing AI across the enterprise and trade lifecycle
Intelligence at scale – built on two decades of data infrastructure investment

Coinbase: Powering stablecoins, wallets, and agentic commerce on AWS
The open protocol making machine-to-machine payments a commercial reality

Results that speak

Stripe: Reimagining
compliance workflows
Autonomous research agents that find the answer before the analyst even opens the case

Robinhood: Defending against financial crime
Specialized agents synthesize customer and transactional data into investigative summaries

From fraud detection to agentic commerce to personalized homeownership, deployments that delivered measurable impact at scale. Read our compelling customer stories here.

AI agents are the fastest growing class of customer. With 150m X402 transactions in the last 6 months and over 10,000 enabled services, agentic commerce is already happening at scale. Coinbase's infrastructure comprising X402 open protocol, programmable wallets, and a compliance-ready facilitator is now embedded directly into Bedrock AgentCore.

Coinbase: Powering stablecoins, wallets, and agentic commerce on AWS

"Instead of going and spinning up 15 API keys for your agent, give your agent one wallet. Now they can pay for over 6,000 services – integrate once, use anywhere." Erik Reppel, Creator of x402, Coinbase

Key takeaways

X402 is blockchain and stablecoin agnostic – USDC, USDT, Solana, Base and non-USD stablecoins all supported
Enterprise-ready from day one: spend controls, KYT, OFAC screening, and full transaction logs built in
X402 governance sits with the Linux Foundation – AWS and Coinbase are both on the oversight committee

Why it matters

The internet was built on open standards – HTTP, HTML – that nobody owns and everyone can build on. X402 is an attempt to do the same for payments. As AI agents proliferate, the question of how they transact becomes foundational infrastructure. Coinbase and AWS are betting that an open, neutral standard settles faster – commercially and technically – than proprietary alternatives.

Nasdaq spent two decades building the data foundation that makes agentic AI possible at scale. Across a portfolio spanning anti-financial crime, regulatory reporting, market surveillance, and capital markets technology, the intelligence layer unifies fragmented data across products into a single, semantically rich platform – already serving around 500 clients.

How Nasdaq operationalizes AI across the enterprise and trade lifecycle

"Doing nothing is going to expose you to being defensive or being disrupted. It's important that we make moves with our clients and for our clients to prepare for that." Marc Murphy, Senior Vice President, Nasdaq

Key takeaways

500 clients already use Nasdaq agentic workflows – this is production, not pilot
Semantic layers and data models are the foundation – without them, scale
is impossible
Explainability is non-negotiable in regulated markets – every agentic decision must be auditable by a human
Client choice is built in – build your own, bring a partner, or procure from Nasdaq

Why it matters

Financial market infrastructure runs on accuracy and trust. Nasdaq's intelligence layer shows what it takes to deploy agentic AI in an environment where errors carry regulatory and financial consequences in the hundreds of millions. The lesson for any organization operating in regulated markets: data quality, governance and explainability aren't constraints on AI ambition – they're the price of entry.

Financial advisors spend 45% of their time moving data across different software tools. In the 10 weeks since launching Hazel Tax Planning, Hazel – Altruist's vertical AI for wealth management – helps the typical advisor save $5,900 in taxes per client, complete a tax plan in 10 minutes, and save 5 hours a week. Built entirely on AWS, Hazel is vertical AI that earns its place.

Hazel by Altruist: Purpose-built AI for wealth management

"AI is going to optimize anything you can put a number to. What's really important now is everything that's left unsaid – the goals, the legacy, the trade-offs. Advisors bring that clarity."
Gokul Ramanathan, Product Lead, Altruist

Key takeaways

New foundation models are put through Hazel's evaluations before deployment
Hazel connects to 30 to 40 tools per advisor and routes each query to the best model combination for that task
Zero data retention with all AI model providers – SOC2 certified, with online scoring and alerting running in real time
The same foundational harness powers the next agent launching in July: holistic financial planning

Why it matters

Vertical AI earns trust differently from general-purpose tools – it must be accurate enough for a regulated, fiduciary environment where a wrong answer costs a client real money. Altruist's evals-first, domain-specific approach shows what it takes to build AI that advisors and compliance teams will actually trust.

Fraud investigators at Robinhood are racing against increasingly sophisticated threats – synthetic identities, coordinated cyberattacks, and case volumes that grow faster than headcount. To help analysts stay ahead, Robinhood built the FinCrimes Agent: a system that draws on data from across the organization to deliver clear investigative summaries, so analysts can act on evidence, not just instinct.

Robinhood: Defending against financial crime

"The FinCrimes Agent has put all relevant information at our investigators’ fingertips, helping them to quickly identify illicit activity with increased focus and accuracy."
Keith Kelley, Head of Financial Crimes, Robinhood

Key takeaways

Achieved a 20% cumulative efficiency gain in fraud
investigative workflows
Scaled from 500m to 5bn tokens daily in just 6 months
Reduced AI costs by 80% and cut development time in half using Amazon Bedrock
Automated hallucination checks ensure every response meets regulatory explainability standards

Why it matters

Financial services breach costs average $6.08m – second only to healthcare. Keeping pace with that risk means giving analysts better tools. AI that reasons across evidence and puts the right information in front of investigators helps them protect customers faster and with greater confidence.

Read more

Why it matters

Homeownership involves dozens of steps across multiple systems, and clients expect help at any hour on any channel. Rocket's approach shows what's possible when you deploy multiple specialized agents behind a single interface – each handling a specific domain, with guardrails and memory built in – rather than one general-purpose chatbot trying to do everything.

Rocket serves nearly 10 million clients and handles 65 million calls a year. With the launch of Rocket AI Agent, built on Amazon Bedrock Agents, 8 specialized agents were deployed across loan origination, servicing, and broker support. The result: clients who interact with Rocket AI Agent are 3x more likely to close a loan. Transfers to customer care dropped 85%, transfers to specialists dropped 45%, and users save ~15 minutes per application.

Rocket: Streamlining the home buying experience with Amazon Bedrock

"Together with AWS, we’re getting started. Our goal is to empower every client to move forward with confidence and, ultimately, to Help Everyone Home."

Key takeaways

Clients who interact with Rocket AI Agent are 3x more likely to close a loan
85% reduction in transfers to customer care, 45% reduction in transfers to servicing specialists
68% high satisfaction ratings across servicing and origination chat interactions
Eight specialized agents behind a unified interface
24/7 multilingual support including Spanish, with seamless handoff to human bankers when needed

Read more

For MUFG's bankers, keeping up with clients used to mean waiting days for the right materials. Now it takes minutes. Through a multiyear partnership with AWS, MUFG is using Amazon Bedrock to put generative AI in the hands of every employee – compressing the gap between insight and action, across the entire organization.

MUFG: Accelerating digital transformation

"By using AWS, we can accelerate the time to production, and in that way, we can build much more time for internal employees to directly contact with our client." MUFG Executive

Key takeaways

Shortened one-week development and preparation processes into just a few minutes
Broad deployment of Amazon Bedrock to ship AI and ML models to every employee
Significant acceleration of time to production for client-facing materials and internal apps

Why it matters

The best bankers spend their time with clients, not waiting on tools. MUFG built their AWS partnership around that belief – and turned weeks of preparation into minutes.

Compliance analysts at Stripe face growing case volumes, tighter timelines, and the weight of getting every decision right. Stripe built AI research agents that investigate cases autonomously before a human analyst ever sees them. The result: faster reviews, sharper insights, and analysts spending more time on the decisions only they are qualified to make.

Stripe: Reimagining compliance workflows

"Analysts focus on the why, whereas the agents focus on the what and how." Hasan Tariq, Principal Solutions Architect, AWS

Key takeaways

26% reduction in average handling time for compliance reviews
Analysts gain 8-12 more hours of strategic work weekly
AI-generated insights found helpful by human reviewers 96% of the time
Over 100 agents deployed across Stripe, with a rigorous QA process and humans completely in the loop during review
Built on Amazon Bedrock with a DAG-based orchestration that turns compliance workflows into parallelized, bite-sized tasks

Why it matters

Regulated industries run on trust. By shifting from manual, sequential reviews to parallel AI-assisted workflows, Stripe keeps pace with global growth without compromising the rigor their customers depend on.