Community
Picture this: Your legal team spends 360,000 hours annually reviewing loan agreements. That’s 180 people working full-time, every single day, just reading contracts. Now imagine reducing that to seconds. JPMorgan Chase didn’t just imagine it. They built it, deployed it, and saved millions.
This isn’t another breathless AI prophecy about the future of finance. This is about what’s working right now, what’s failing spectacularly, and what senior finance executives actually need to know before their next board meeting asks why they’re not “doing more with AI.”
Let’s start with the number that should make you uncomfortable: median AI ROI in finance sits at just 10%, according to BCG’s 2025 survey of 280 finance executives. That’s half of what most CFOs target. One third of leaders report limited or no gains at all.
Yet spending continues to surge. Financial institutions are expected to invest $97 billion in AI by 2027. JPMorgan Chase plans to spend $18 billion on technology in 2025, with AI accounting for a significant portion. Why the disconnect between investment and returns?
Simple: Most organizations are building solutions in search of problems. They’re chasing AI because the board demands it, not because they’ve identified specific, measurable business challenges that AI can solve better than existing approaches.
The winners? They start with problems, not technologies. They measure everything. And they accept that 30% of their AI projects will fail after proof of concept, as Gartner predicts. The difference is they fail fast, learn faster, and scale only what works.
JPMorgan’s Contract Intelligence (COiN) platform remains the gold standard for AI implementation done right. Before COiN, reviewing commercial credit agreements consumed those 360,000 hours annually I mentioned. Human lawyers pored over 12,000 contracts, extracting 150 key attributes from each document. Errors were common. Costs were astronomical.
COiN processes those same 12,000 agreements in seconds. Not minutes. Seconds.
The platform identifies default terms, renewal conditions, and regulatory requirements with near-zero error rates. More importantly, it freed JPMorgan’s legal teams to focus on negotiation strategies and complex advisory work instead of document grunt work. Real lawyers are doing real legal work, while machines handle the repetitive extraction tasks.
This wasn’t magic. It was methodical. JPMorgan trained the system on thousands of documents, built robust validation processes, and integrated it seamlessly into existing workflows. They didn’t try to replace lawyers. They augmented them.
Every millisecond matters in fraud prevention. Traditional rule-based systems catch obvious patterns but miss sophisticated attacks. Modern AI fraud detection analyzes transaction patterns, device fingerprints, geolocation data, and behavioral signals in real time.
Banks using advanced fraud detection AI report 50% reductions in false positives while catching 20% more actual fraud. That translates to millions saved annually and happier customers who don’t get their cards blocked for buying coffee in a new neighborhood.
The key? These systems learn continuously. Each flagged transaction, whether fraud or false positive, improves the model. The AI adapts to new fraud patterns faster than human analysts can update rule sets.
Traditional credit scoring excludes millions of potential borrowers with thin credit files. AI-powered credit models incorporate alternative data like utility payments, mobile phone usage patterns, and cash flow analysis to assess creditworthiness.
Upstart, using machine learning for credit decisions, approves 27% more loans than traditional models while maintaining the same loss rates. They’re not being reckless. They’re being smarter about risk assessment.
For established banks, this means accessing entirely new customer segments profitably. For customers, it means fair access to credit based on actual financial behavior, not just credit history.
85% of finance leaders cite data quality as their primary AI challenge. Your AI is only as good as your data, and most financial institutions’ data is a mess. Customer information is scattered across dozens of systems. Transaction data in incompatible formats. Historical records with missing fields.
Before you spend a dollar on AI, audit your data. Can your teams access the data they need within a week? If not, fix your data infrastructure first. AI amplifies data problems; it doesn’t solve them.
Financial institutions run on legacy systems. Some banks still process transactions on COBOL mainframes from the 1970s. Layering AI on top of this technical debt is like putting a Ferrari engine in a horse-drawn carriage.
Successful implementations take an incremental approach. Start with contained use cases that don’t require deep integration. Build APIs and data pipelines gradually. Accept that modernization and AI adoption must happen in parallel, not sequentially.
Unified APIs help streamline these integrations and provide modern SDKs on top of SOAP and REST APIs, offering the speed and robustness we need.
Your biggest obstacle isn’t technology. It’s people. Employees fear AI will replace them. Managers don’t understand how to measure AI success. Executives want immediate returns from long-term investments.
JPMorgan addressed this by focusing on augmentation messaging. Their LLM Suite, now used by 200,000 employees, explicitly positions AI as a productivity tool, not a replacement. Employees using their coding assistant report 10% to 20% productivity gains. They’re not unemployed; they’re more effective.
Financial services face unique AI challenges. Every algorithm must be explainable to regulators. Every decision must be auditable. Every model must avoid discriminatory bias.
The EU AI Act and similar regulations aren’t suggestions. They’re requirements with teeth. Financial institutions using AI for credit decisions, risk assessment, or customer interactions must demonstrate fairness, transparency, and accountability.
Smart organizations are building compliance into their AI development process from day one. They’re creating model cards documenting training data, performance metrics, and known limitations. They’re establishing AI governance committees with real authority. They’re investing in explainable AI techniques that can justify decisions to regulators and customers alike.
You don’t need another proof of concept. JPMorgan has 450 AI use cases in various stages. Most will never scale. The ones that do follow a clear pattern: specific problem, measurable success metrics, gradual rollout with continuous measurement.
Pick three high-impact use cases. Set clear KPIs. Give them six months. Kill what doesn’t work. Scale what does.
65% of JPMorgan’s workloads now run on cloud infrastructure, up from 50% a year ago. This isn’t a coincidence. AI requires computational flexibility, data accessibility, and rapid deployment capabilities that legacy infrastructure can’t provide.
Your AI strategy is really a cloud and data strategy. Invest accordingly.
Unless you’re JPMorgan with 55,000 technologists and $14 billion to spend, you’re not building foundational AI models. Unless AI is your core differentiating competency, like it is for Renaissance Technologies or Two Sigma, you’re wasting resources trying to compete with OpenAI or Anthropic.
Your competitive advantage isn’t in building better models. It’s about applying existing models more effectively than your competitors. Partner with established AI vendors. Use pre-trained models. Focus your internal resources on integration, customization, and domain-specific applications where your industry expertise actually matters.
Traditional ROI metrics often fail to capture the full value of AI. Productivity improvements, error reductions, and employee satisfaction matter as much as direct cost savings.
Track time saved, decisions improved, and risks avoided. These leading indicators predict long-term value better than quarterly profit impact.
Agentic AI is where the real disruption begins. Forget chatbots responding to prompts. These AI agents will proactively execute complex, multi-step tasks without hand-holding. Imagine an AI that doesn’t just analyze loan applications but negotiates terms, prepares documentation, and manages the entire origination process while you sleep.
Only 12% of organizations have deployed agentic AI, but over half are exploring it. Early adopters will gain significant competitive advantages. Late adopters will struggle to catch up.
Generative AI will move from back-office productivity to customer-facing applications. AI will draft personalized financial advice, create custom investment strategies, and provide real-time financial coaching. The banks that nail the user experience while maintaining compliance will emerge as winners.
AI in finance isn’t about the technology. It’s about solving real problems for real businesses with real constraints. The organizations succeeding with AI share three characteristics:
First, they start with problems, not solutions. They identify specific, painful, measurable challenges that AI can address better than current approaches.
Second, they invest in foundations. Data quality, cloud infrastructure, and employee training matter more than algorithm sophistication.
Third, they maintain realistic expectations. They accept that most experiments will fail, returns take time to materialize, and AI augments rather than replaces human judgment.
The $97 billion question isn’t whether AI will transform finance. It’s whether your organization will be among the few that capture real value or the many that subsidize expensive experiments.
Stop talking about AI strategy. Start solving actual problems. The technology works when you do.
The choice is stark: strategically solve for your own ‘360,000-hour’ problems, or join the 90% burning cash on buzzword implementations that never deliver meaningful returns. JPMorgan chose to solve. What will you choose?
ROI and Implementation Data: - BCG Study (June 2025): How to Get ROI from AI in the Finance Function - The CFO (January 2025): The ROI puzzle of AI investments in 2025 - McKinsey (March 2025): The state of AI: How organizations are rewiring to capture value - Guidehouse (June 2025): Closing the ROI gap when scaling AI
JPMorgan COiN Case Studies: - DigitalDefynd (August 2025): 10 ways JP Morgan is using AI - In Depth Case Study - Medium/Ahmed Raza (May 2025): How JPMorgan Uses AI to Save 360,000 Legal Hours a Year - Constellation Research (June 2025): JPMorgan Chase’s IT, AI bets: Where the returns are - Tearsheet (May 2025): JPMorgan Chase’s Gen AI implementation: 450 use cases and lessons learned
Industry Analysis: - World Economic Forum: Here’s how AI is transforming finance, according to CFOs - RGP (July 2025): AI in Financial Services 2025 - RTS Labs (August 2025): Top 7 AI Use Cases in Finance - Workday: How AI Is Changing Corporate Finance in 2025
This content is provided by an external author without editing by Finextra. It expresses the views and opinions of the author.
Anurag Mohapatra Director of Fraud Strategy and Marketing at NICE Actimize
26 September
Anil Kollipara Vice President, Product Management at Spirent
Nkahiseng Ralepeli VP of Product: Digital Assets at Absa Bank, CIB.
25 September
Konstantin Rabin Head of Marketing at Kontomatik
24 September
Welcome to Finextra. We use cookies to help us to deliver our services. You may change your preferences at our Cookie Centre.
Please read our Privacy Policy.