There’s no shortage of promises in the fast-paced world of AI. Over the last three years, founders and leaders in the industry have bellowed assurances of faster insights, smarter agents, autonomous workflows and even so-called super-intelligent AI systems that will surpass humans at all levels of intelligence.
But for Gerard Francis, chief product officer of AI and data at JPMorganChase, all of that hype falls flat without a disciplined, enterprise-wide approach to data. Speaking during a customer spotlight session at Snowflake Summit 2025, where Snowflake’s clients discussed how they’re building and scaling AI systems, Francis emphasized that “in the absence of a great data, AI and governance platform, every AI experiment is non-repeatable.”
As enterprises rush to harness AI, few have cracked the code on moving from proof-of-concept to large-scale deployment. One big factor behind this, according to Francis, is that many organizations lack the foundational data infrastructure needed to scale beyond early pilots.
From AI Experiments To Enterprise Impact
The story of AI adoption across Fortune 500 companies has followed a familiar arc: First a proof of concept, then a press release and then comes the plateau. While most firms have dabbled in GenAI, few have moved beyond contained trials into full-scale deployments that deliver repeatable, enterprise-grade value.
“It’s all about the data,” Francis began when I asked what enterprise AI truly means at JPMorganChase. He noted that success isn’t about flashy models, but about solving real problems across banking, asset management, fraud detection and more — and doing it at scale. “How do you identify the highest-value use cases and scale them to have maximum business impact?”
That scaling is precisely where most AI efforts falter. Research from analyst firm Gartner suggests that at least 30% of generative AI businesses currently testing will be abandoned after proof of concept by the end of 2025. The bottleneck, Francis suggested during our conversation, lies in infrastructure and governance, not in how large or powerful an AI model is. “Without clear vision and readiness at the infrastructure level,” he told me, “no amount of AI investment will deliver sustained value.”
That understanding helped JPMorganChase pilot generative AI within the organization by building a unified platform that connects data, AI and governance into real-time workflows and repeatable insights. The company’s in-house GenAI chat application “LLM Suite” allows employees to safely interact with large language models, protected by access controls and data usage policies. Initial deployments focused on use cases like document drafting, workflow simplification and internal communication — where value is clear and risk is manageable.
“We had the right governance and controls, ensuring the data is protected,” Francis explained in the interview. “The concept was simple: Deploy AI where it can immediately deliver value and stay safe doing it.”
The Data Discipline Behind AI Readiness
What does AI readiness actually look like inside one of the world’s largest financial institutions?
For JPMorganChase, it begins with data discoverability and access. “Is your data in a place where it can be discovered?” Francis asked. “Does it have the right level of entitlement so people can only get the data they should be getting?” These aren’t just technical concerns but real compliance imperatives for a firm regulated across multiple jurisdictions and client categories.
From there, the focus shifts to unstructured data: Documents, notes, excel sheets, contracts and more. Historically hard to parse, these sources are now becoming valuable thanks to retrieval-augmented generation (RAG) and other GenAI techniques. But even with AI, data quality still matters. “Avoid duplicate documents,” Francis said. “Make sure you’ve got the right version control so people can get accurate answers.”
Structured data — scattered across countless internal systems — tends to come last, yet often proves the most powerful when integrated. That’s why JPMorganChase built Fusion — an internal data platform for customers that acts as a “data factory” for orchestrating pipelines, normalizing formats and making datasets AI-ready.
JPMorganChase’s infrastructure spans multiple vendors and platforms, including Snowflake, which supports its broader efforts to unify enterprise data for AI readiness. “Think of us as the data factory that operates at scale,” Francis said.
Governance As A Backbone
Talk to any enterprise AI leader, and governance will eventually come up. But at JPMorgan, said Francis, it’s not an afterthought. Rather, it’s embedded in the strategy from day one.
“When you are part of a regulated entity,” Francis explained, “you’ve got to always make sure that the data you’re using for a particular purpose is an approved use of that data.” That means aligning use cases not only with internal policies, but also with country-specific laws, contractual obligations and client privacy terms.
Managing those controls manually would be a nightmare. JPMorganChase’s goal is to transition “from a heavy human process aided by technology to an entirely technology-enabled process.” Until then, scalability and compliance depend on how well governance is operationalized into the AI development lifecycle.
AI Agents Are The Next Frontier
Imagine an AI system that doesn’t just summarize a document but reconciles data, files reports, books appointments and updates compliance records. These are the early signals of what’s next in enterprise AI: Autonomous agents that can act on behalf of users with limited supervision.
While many AI deployments today still focus on summarizing text or generating content, the industry-wide shift to agentic AI is already underway and JPMorganChase is paying close attention.
These autonomous systems, capable of reasoning and decision-making, promise immense value, particularly in complex workflows like software development, research, or operations. But Francis isn’t rushing. “Agentic solutions offer phenomenal value,” he said, “but also bring a lot of risks. That’s an area we’ve got to educate ourselves on.”
He’s cautious and strategic. Rather than replacing jobs outright, the goal is augmentation: helping teams move faster, armed with better data and smarter suggestions. As Francis noted, “It’s less about agents or not, but more about whether you can really solve a use case well.” If you can, you either lower your cost or improve your revenue opportunity.”
For an organization with JPMorganChase’s scale and complexity, AI makes sense only when it drives value. “We determine the values we want to pursue and that determines how we prioritize AI,” Francis said.
That value-driven mindset also extends to ROI. While Francis declined to cite exact numbers, he noted that the firm reports publicly on AI value generation — with much of that value still coming from traditional machine learning — though GenAI is advancing quickly.
“If we can dramatically drop the price of any use case, then the return on investment becomes easier to justify and easier to scale.”
The Long Game Of AI Leadership
Looking ahead, Francis hopes JPMorganChase will be remembered for solving one of enterprise AI’s hardest problems: Building a platform that integrates data, AI and governance across multiple technology stacks.
“It’s very often you can do this for one vendor,” he said. “But doing it across vendors is incredibly difficult.”
Still, the payoff is clear. If Fusion and platforms like it can eliminate the friction between pilots and production, AI will stop being an experiment and start becoming the enterprise default.
And in that world, the winners won’t be those with the flashiest models, but those with the strongest data discipline.