Author: Charitarth Sindhu, Fractional Business & AI Workflow Consultant
Agentic AI is no longer a pitch deck buzzword in financial services. It is in production. Banks, fintechs, and lending platforms are deploying AI systems that do not just flag problems or spit out predictions. They investigate, decide, and act.
We asked five industry leaders a simple question: what is one practical use case where agentic AI will deliver real value in financial services this year? Their answers covered everything from treasury cash management to cross-border freelancer payments. But a clear thread ran through every response. The biggest gains are not coming from replacing humans. They are coming from removing the grunt work that stops humans from doing their jobs properly.
The numbers back this up. The agentic AI market in financial services is on track to grow from roughly $5.5 billion in 2025 to $33 billion by 2030. Seventy percent of banking executives say their firms now use agentic AI in some form. JPMorgan, HSBC, Bank of America, and dozens of fintechs have moved well past pilot stage.
Here is what each leader had to say.
Smarter cash, fewer surprises
Treasury management might not grab headlines, but it is where agentic AI is delivering some of its fastest returns. Bank of America’s CashPro Forecasting tool helped over 3,000 companies save more than 250,000 hours in 2025 alone. HighRadius reports up to 95% cash forecast accuracy and a 50% reduction in idle cash across its client base. When S&P 1500 companies are sitting on roughly $707 billion in trapped working capital, even small improvements in cash positioning translate to millions in recovered value.
“An AI agent can monitor cash positions across accounts and forecast short-term needs based on real-time inflows. It can then create a daily funding plan and suggest transfers according to rules set by the treasury team. Humans remain in control by approving these suggestions. This helps avoid surprises at the end of the day and reduces manual spreadsheet work.
The value of this system appears quickly because it eliminates the need for constant updates. The agent can explain each recommendation and highlight the risks if no action is taken. It can also simulate different stress scenarios, like delayed settlements. This helps banks and fintechs avoid overdrafts and reduce idle cash, which is especially useful in a volatile rate environment.”
- Sahil Kakkar, CEO / Founder, RankWatch
Cutting through the alert noise
Fraud teams are drowning. Global scam losses hit $1.03 trillion in 2024. Traditional rule-based detection systems generate false positive rates between 30% and 70%, and alert volumes have surged 800% according to Sardine CEO Soups Ranjan. The result? Burned out analysts spending most of their time chasing dead ends instead of catching real criminals. JPMorgan achieved a 95% reduction in false fraud alerts, while HSBC cut false positives by 60% and detected two to four times more financial crimes.
“Agentic AI will provide value this year in fraud operations by acting as a case triage and action agent. Analysts are often overwhelmed with alerts that require the same checks to be repeated across multiple systems. This process slows down response times and increases false positives. By deploying an AI agent, we can pull the relevant evidence for each alert and create a decision packet.
The agent will verify recent device patterns and transaction context before recommending the next action based on policy. If confidence is high, it can lock the session or request step-up authentication. If confidence is low, it will close the case with notes and supporting evidence. This approach saves time per alert, reduces queue time and improves consistency, allowing for faster containment and better focus for analysts.”
- Christopher Pappas, Founder, eLearning Industry Inc
Following the money across silos
One of the oldest problems in banking is that data lives everywhere and talks to no one. Marketing runs campaigns in one system. Sales tracks leads in another. Underwriting and servicing operate on completely different platforms. The result is that nobody can tell you which efforts are driving revenue and which are burning budget. Organizations with fragmented data regularly miss conversion touchpoints or double-count others, leading to skewed ROI calculations. Companies that crack this problem see up to 30% better campaign performance.
“Autonomous Revenue Attribution and Deal Intelligence across complex financial institutions is one of the first orders. Countless banks, fintech companies and lending organizations fail not because they don’t have the data, but because their data exists in marketing systems, CRM platforms, underwriting tools and servicing software that do not talk to one another. An agentic AI system can go beyond passively reporting but actually actively track a deal’s motion, resolve competing signals, find gaps in the funnel and suggest next moves for revenue organizations. The value is mirage-free, not merely a theoretical placeholder. Capital allocation improves, almost overnight, if leadership can see which campaigns, partnerships or sales motions are actually leading to funded loans or new accounts. In an industry where margins are so dependent on cost of acquisition as well as risk quality, the focus to understand what actually drives revenue is a competitive edge. Agentic AI creates value for leaders this year not by replacing human decision makers, but by perpetually integrating and interpreting fractured news signals in ways that can enable leaders to act with confidence rather than intuition.”
- Mada Seghete, Co-founder, CEO and Marketing, Upside.tech
Making non-QM underwriting less painful
Non-QM lending is booming. Its share of total mortgage originations has jumped from about 3% in 2020 to a projected 10-15% in 2025, with DSCR loans making up half of all non-QM mortgage-backed securities. But underwriting these loans is complex. Borrowers bring bank statements instead of W-2s. Rental income needs specialised appraisals. Multi-property portfolios demand separate DSCR calculations for every asset. The average cost to originate a mortgage hit $11,800 in mid-2025, and the average underwriter touches each application 4.2 times because of incomplete information.
AI platforms like Ocrolus now classify over 1,600 financial document types with 99%+ accuracy. Figure launched a DSCR platform that cuts processing from 21-30 days down to five. Prudent AI built the first automated underwriting system specifically for non-QM, partnering with Angel Oak and Newfi Lending.
“An extremely useful use case here is intelligent underwriting collaboration in non-QM mortgage and investor lending. In most loan origination surroundings, the bottleneck isn’t document availability it’s reconciling income streams, rental proforma scenarios, bank statements, entity set-ups and appraisal data with a cogent narrative of creditworthiness. An AI which is able to be proactive with the borrower, but also identify missing documentation, cross reference cash flow assumptions and create an underwriting summary for review can save huge amount of cycle time. It’s not in replacing loan officers or underwriters at all; it is about relieving them of administrative assembly to concentrate on risk judgment and structuring. DSCR and multi-property financing in particular, where income analysis can be stacked and driven by entity, this level of intelligent coordination increases consistency and throughput into the same fiscal period. Standardization and responsiveness with underwriting This in turn leads to more predictable capital deployment to support revenue growth as well as borrower experience.”
- Christopher Ledwidge, Co-Founder & Executive Vice President of Retail Lending, theLender.com
Cleaning up cross-border payments
The gig economy has gone global, but the payment infrastructure has not kept up. Freelancers working across borders deal with a tangle of VAT rules, withholding obligations, and anti-money laundering checks that change depending on where they live, where their client is based, and what currency they are using. Get any of it wrong and payments get stuck, penalties pile up, or freelancers get pushed toward working unregistered.
“Cross-border freelancer payments. Right now, compliance checks for international payouts are a mess. Every transaction touches multiple regulatory frameworks depending on where the freelancer lives, where the client is based, and what currency they’re paying in. A platform like ours processes thousands of payments across dozens of countries every month, and the compliance burden grows with every new market.
An AI agent can handle the heavy lifting here. It can verify a freelancer’s identity documents against local requirements, flag sanctions risks in real time, match invoices to the right tax treatment based on jurisdiction, and route payments through the most cost-effective rails. All of this before a human even needs to look at it. The agent learns which document combinations satisfy regulators in each country and proactively asks for what’s missing before a payment gets stuck.
This matters because the current process is slow, expensive, and full of friction that pushes freelancers toward unregistered work. When someone in Indonesia invoices a client in Germany, there are VAT rules, withholding obligations, and AML checks that all need to line up. Getting any of those wrong means delays, penalties, or both. An agentic system that handles this end to end, with a human reviewing only the edge cases, cuts processing time dramatically and keeps compliance teams focused on real risks instead of routine paperwork.”
- Hasan Can Soygök, Founder, Remotify.co
The common thread
Every one of these leaders pointed to the same shift. Agentic AI is not about building a smarter chatbot or a better dashboard. It is about systems that can investigate, reason through options, and take action within defined boundaries.
The technology is ready. The ROI is proven. KPMG found companies earn $3.50 for every $1 invested in agentic AI. But the risks are real too. Gartner predicts over 40% of agentic AI projects will be cancelled by end of 2027 due to unclear business value or poor risk controls. No jurisdiction has enacted regulations specifically targeting agentic AI systems yet, and questions around liability when AI agents make errors remain unresolved.
The institutions that get this right will not be the ones chasing the flashiest AI demos. They will be the ones picking the right boring problems, like cash forecasting and document reconciliation, and letting AI agents handle the repetitive complexity that has been choking their teams for years.