Need a Loan? The (AI) Agent Will See You Now
Where AI Agents Meet Financial Services
Words: Andrea You; Edits: Jaclyn Hartnett
The finance industry is undergoing a profound transformation driven by automation, with AI agents at the forefront. Morgan Stanley just rolled out AskResearchGPT. Klarna’s AI assistant now handles two-thirds of its customer service inquiries (equivalent to 700 full-time agents). Bank of America has doubled its number of AI patents in two years. These intelligent programs are not merely owning routine tasks, but ushering in a new era of financial automation that has the potential to drive unprecedented cost and labor efficiencies — completely upending traditional business models.
Despite limited AI adoption in financial institutions today (~30% utilize basic applications), the advent of LLM-based AI agents is poised to dramatically expand AI’s role across financial sectors. This agent-driven shift comes as the industry faces mounting challenges from widespread talent shortages and tedious workflows with manual, error-prone labor. AI agents offer a promising solution to build expansive yet specialized knowledge repositories and emulate human ability to learn from and execute on new information.
The magic happens when these self-driven systems work together to solve nuanced problems with minimal human involvement, enabling people to do 10x more with less. While the agent revolution is completely redefining automation, it also necessitates cautious development and governance to ensure safe, compliant and ethical operations.
THE BASICS — WHAT EXACTLY IS AN AI AGENT?
AI agents are software programs that autonomously interact with environments, make decisions and perform tasks to achieve specific goals. Utilizing machine learning and natural language processing, AI agents improve performance over time to tackle complex, dynamic functions. Agents vary in sophistication based on model architecture and it’s important to understand the nuances of each so they can be purpose-built and optimized for time and cost efficiencies.
- Simple Reflex Agents: based on predefined rules and immediate data (not past experiences). Example: automated trading system that buys or sells stocks when they reach a specific price.
- Model-Based Reflex Agents: maintains internal models of environments and considers past experiences in decision-making. Example: fraud detection system that uses historical transaction data to identify suspicious activities.
- Goal-Based Agents: creates strategies to achieve goals. Example: investment robo-advisor that creates and adjusts portfolios to meet clients’ financial goals.
- Utility-Based Agents: makes decisions by evaluating the desirability of different outcomes with utility functions. Example: risk assessment tool that evaluates investment options based on potential returns and risk tolerance.
- Learning Agents: improves performance over time through experience and feedback. Example: chatbot that improves its responses based on user interactions and feedback.
- Hierarchical Agents: organizes decision-making processes into hierarchical structures for complex tasks. Example: AI-powered financial management system that coordinates various aspects of a bank’s operations, from customer service to risk management.
- Multi-Agent Systems: multiple agents working together to solve complex problems. Example: network of AI agents managing different aspects of a large-scale trading operation, including market analysis, risk assessment and execution.
Specifically, Learning Agents, Hierarchical Agents and Multi-Agent Systems are incredibly exciting — models advanced enough to not only 10x human productivity, but enable individuals to focus on tasks where human judgment and relationships are most valuable.
AGENTIC AI x FINTECH — THE OPPORTUNITY
What started as a simple Excel macro has gone through many evolutions to get to the static, binary rules-based system that RPA is today. Now, agentic AI is up-leveling the stack by connecting LLMs with code, new data sources and simple UX to execute workflows from end-to-end.
The real value of automation is unlocked when AI agents can wholly own some of the most pressing (and expensive) challenges that financial institutions face. For example, manual risk management processes often comprise north of 30% of a firm’s overall cost structure and human-driven errors are leading to fines in the billions. In lending, inefficient underwriting and onboarding can lead to losses of up to 30% of potential revenue.
V1 of generative AI acted on static predefined rules for single tasks and has now evolved into end-to-end automation — from intaking customer information and managing the relationship, analyzing disparate data points, to making the loan decision and leveraging learnings to inform the next decision. Now no longer as brittle as before, agents are thinking, acting and reasoning, integrating new learnings to execute increasingly complicated tasks.
Financial institutions are notoriously risk-adverse and impose substantial regulatory and compliance hurdles that can cause decision-makers to take a more wary approach — grappling with concerns around data accuracy, privacy and apprehension around “black box AI”.
Pro Tip: Startups that prioritize strong security frameworks, build in proactive compliance monitoring and features like workflow monitoring, audit logs and human-in-the-loop capabilities are well-positioned to remedy buyer concerns. Delivering ROI is a requisite, but startups must also provide reliability, transparency/traceability and a staunch commitment to data privacy and security to build the trust that is critical for adoption.
AGENTIC AI x FINTECH — THE VENTURE PERSPECTIVE
Venture capital continues to pour funding into AI-related startups, including agentic AI, accounting for 35% of US startup investments this year. When it comes to fintech, here are the top areas where we see the most promise for agentic AI.
Banking & Insurance
Most financial institutions leverage decision-tree type algorithms to “automate” the decisioning process, anchoring on the same few, manually-collected data points (e.g. credit score, payment history, number of accounts, etc.) to build a holistic view of the customer/business. AI agents are becoming critical to determine optimal loan amounts based on detailed risk profile analyses (with an unprecedented number of data points) — supporting more fair and accurate credit risk decisions within seconds vs. days.
IRL: Startups like Casca are automating all pieces of the flow, from collecting documentation to determining eligibility and approving loans based on changing regulatory standards.
In insurance, claims costs account for ~80% of policyholder premiums, with manual processing often taking 2x longer and costing 3x more. The industry estimates up to $45B in annual savings potential from real-time adjudication. We’re seeing an influx of startups building automated solutions across the stack, from pricing to managing the entire claims process from initiation to payout.
IRL: Startups like Sixfold enable insurers to encode unique risk appetites into AI models, wholly automating and driving more transparency across the decisioning flow. Agents like Multimodal and Assured Claims automate claims handling to speed up operations and reduce costs while improving customer satisfaction by minimizing denials and delays with augmented data.
KYC, Compliance & Risk Management
Financial institutions face numerous challenges across compliance and risk management. This includes high error rates (estimated at 2–5% for manual KYC checks), amounting to substantial compliance costs and potentially hefty fines for non-compliance (global AML fines totaled $3.2B in 2020). AI agents’ ability to automate data capture, validation and support always-on compliance monitoring represent a step-function change — enhancing efficiency and accuracy as well as reducing the likelihood of regulatory penalties and reputational damage.
IRL: Manual KYC checks for banks can take 60+ days, while AI-powered solutions like Inscribe can verify customers under 24 hours. With automation, financial institutions can potentially reduce KYC-related costs by up to 70%. Agents like Parcha and Greenboard continuously monitor submissions for accuracy, flagging potential compliance risks in real-time and adapting to evolving regulatory requirements.
Finance Back-Office
The CPA labor shortage continues to squeeze firms with ~17% of professionals resigning in the past 2 years.
IRL: Companies like Black Ore and Basis are building customized agents for firms and specific use cases (e.g. tax prep). For in-house bookkeepers, platforms like Sema4 and Nominal enable users to transform complex business logic into automated workflows, putting tedious processes like multi-entity accounting reconciliations on autopilot.
Lastly, perhaps the most glaring (and ironic) gap in fintech automation has been the movement of money. We’re beginning to see the emergence of agents capable of facilitating payments in a safe and secure way, representing a significant breakthrough that pushes the boundaries of autonomous decision making as agents move beyond information processing to real-world actions.
IRL: Platforms like Skyfire are creating payment networks that enable AI agents to execute transactions within set parameters. Examples include autonomously sourcing materials for manufacturers, hiring and paying contract workers and managing personal finances and purchases on behalf of individuals.
PARTING THOUGHTS — ENTERING THE AI-POWERED FINANCIAL FUTURE
AI agents are not just the future of finance — they’re rapidly becoming its present and we’re seeing new applications emerge across the fintech stack every day (we’ll have an exciting update to share here very soon!). The firms that successfully integrate AI agents stand to gain a real competitive advantage, armed with the tools to operate more seamlessly, make better-informed decisions and uphold high safety and security standards.
At Cathay Innovation, we’re well-versed on the intersection of financial services and AI — being backed by Fortune 500 corporations like BNP Paribas and leading asset managers like Bpifrance. From an investor perspective, this presents a wealth of opportunity as newfound productivity flips entire business models, transforming how the world approaches labor as we know it. As we navigate this exciting frontier, we remain keenly aware of the regulatory challenges and risks. Algorithmic biases, data security and unforeseen market impacts are significant considerations, and it will be crucial to strike a balance between innovation and building AI responsibly to earn the trust of a notoriously complex industry.
If you’re building here, we’d love to hear from you!
For more insights on the world of AI, fintech and startups across our key sectors, check out our previous posts:
