Where AI Actually Works in Finance (and Where It Doesn’t)

AI is everywhere, but in finance and treasury, leaders are still asking the same question: what’s actually usable today? In this episode of the AI Driven Enterprise Podcast, we sit down with Jim Barry, Global Director for Finance & Treasury at Stifel Financial Corporation, to explore where AI can deliver real value inside a financial services enterprise.

Jim shares what Stifel is seeing across the market, including AI-powered cash forecasting pilots from platforms like Kyriba, as well as forecasting and modeling tools offered by major banks. But the real takeaway is clear: data integrity and tagging matter more than the AI model itself. “Junk in, junk out” isn’t just a saying; it’s the core blocker to enterprise AI adoption.

We also dive into high-confidence use cases like reconciliations, accounts payable automation, and recurring cash movement, along with the more complex question: what parts of finance should remain human-led? Jim offers a strong perspective on why the advisor-client relationship and trust-heavy financial decisions won't be replaced by AI anytime soon.

Top 10 Highlights / Takeaways

• Cash forecasting is one of the most “AI-ready” treasury use cases, especially across 13-week and 13-month horizons.
• Tools like Kyriba and banks like JPMorgan are already offering forecasting models worth piloting.
• The biggest limiter isn’t the AI; it’s data tagging, cleanliness, and governance.
• “Junk in, junk out” is still the #1 truth in AI-driven finance automation.
• Stifel’s approach reflects a pattern of being an early adopter (Oracle Cloud since ~2017/2018, Microsoft ecosystem, Copilot).
• Finance leaders are tired of AI “what-if” hype; they want real working solutions, not slides.
• Visualizing real use cases is key: leaders need to see AI working to trust it.
• Strong automation candidates: reconciliations, AP, daily processes, recurring cash movement.
• AI agents + payments rails (ACH, wires, RTP) could become a powerful operational accelerator.
• Some areas shouldn’t be automated: subjective valuations and trust-based client relationships (especially in wealth advisory).

AI in Finance: The Use Cases That Are Real, and the Ones That Still Feel Risky

Featuring Jim Barry, Global Director for Finance & Treasury at Stifel Financial Corporation

If you work in finance or treasury today, you’ve probably heard a version of the same message a hundred times:

“AI is going to change everything.”

But the bigger question isn’t whether AI will matter. It’s this:
Where will AI actually work in a real enterprise environment, and where is it still more hype than impact?

That’s what made this conversation with Jim Barry, Global Director for Finance & Treasury at Stifel Financial Corporation, so valuable. Jim didn’t come in with buzzwords. He came in with something better: practical insight from someone sitting inside a global financial services organization that’s already exploring real AI pathways.

And in a world full of “AI evangelists,” that’s refreshing.

The Most Realistic Starting Point: Cash Forecasting

One of the first areas Jim called out was cash forecasting, particularly from a treasury perspective.

Cash forecasting has always been a major part of treasury operations, but it’s also one of those functions where AI has a clear advantage, because it’s built on patterns, historical behavior, and structured data.

Jim shared that Stifel has looked at multiple paths here, including:

• AI capabilities rolled out by Kyriba (their treasury management platform)
• Modeling tools offered by large banks such as JPMorgan

The promise is clear: AI could help treasury teams forecast cash needs across:

• 13-week horizons (short-term liquidity planning)
• 13-month horizons (longer-term planning)

This is exactly the kind of finance problem AI is well-suited for: repetitive, data-heavy, and operationally meaningful.

But Jim also pointed out a crucial caveat.

The AI Truth Nobody Can Escape: “Junk In, Junk Out”

Every AI conversation eventually lands here, and Jim hit it early.
AI can only interpret what it can understand.

And if the data isn’t tagged properly, structured properly, or governed properly, AI won’t magically “figure it out.”

It will simply generate:

• Unreliable outputs
• Incomplete models
• Misleading forecasts
• False confidence

In other words, it won’t just fail quietly; it can fail in a way that looks convincing.

That’s why the real foundation for enterprise AI isn’t the AI tool. It’s:

• Clean data
• Consistent data models
• Strong governance
• Clear security guardrails
• Contextual metadata (so AI knows what it’s looking at)

This is especially true in finance, where “almost right” is still wrong.

Why Enterprise AI Adoption Needs Working Demos, Not More Presentations

One of the strongest themes in the conversation was the difference between:

• AI theory
• AI execution

Jim mentioned something many finance leaders are quietly thinking:

There’s been a lot of “talk” and not enough “practical.”

He even drew a parallel to blockchain, another technology that generated massive excitement, huge hypotheticals, and then struggled to produce widespread real-world business adoption.

AI is definitely further ahead than blockchain, but in finance, the same skepticism still exists:

“What are people actually using, not just what might be possible?”

This matters because finance teams don’t get rewarded for experimenting.

They get rewarded for:

• Accuracy
• Risk reduction
• Auditability
• Compliance
• Operational reliability

So when vendors come in with AI slides, the reaction is often:
Show me something real.

That point was reinforced by a story the host shared about a CIO who said (paraphrasing):

“If you’re going to give me another one-hour presentation, it’s not worth my time.”

That’s where enterprise AI is today.

Not “Tell me about the future.”

But: Show me what works now.

The Finance Use Cases That “Naturally Fit” AI

When asked where AI is most likely to succeed inside finance and treasury, Jim identified a few strong candidates:

1) Reconciliations

Daily reconciliations and repetitive matching processes are perfect AI territory.

These tasks tend to be:

• Structured
• Repeatable
• Rules-driven
• Time-consuming
• Low-value for humans

AI can help by identifying mismatches faster, flagging anomalies, and reducing manual review time.

2) Accounts Payable (AP)

AP keeps coming up as one of the highest-confidence AI use cases, and Jim’s view aligns with what many industry analysts (including Gartner) are seeing.

Why?

Because AP involves:

• Vendor invoices
• Approvals
• Exception handling
• Compliance rules
• Recurring payments
• High volume and low strategic value

It’s not glamorous. But it’s impactful.

3) Recurring Cash Movements and Payments

Jim also mentioned recurring wires and cash movements, and this is where the conversation got especially interesting.

In banking and treasury, payments move across “rails” such as:

• ACH rails
• Wire rails
• Real-time payments rails (RTP)

The idea of using AI agents to help manage these movements, especially for recurring workflows, could be a major operational unlock.

Not in a “replace humans” way.

But in a “reduce friction and error” way.

Where AI Should NOT Be Used (Yet): Trust-Heavy Decisions

One of the most important parts of the episode was Jim’s answer to this question:
Where would you say no to AI?

His response was direct:

AI struggles in areas that are subjective, trust-based, or deeply human.

He specifically called out:

• Subjective valuations
• The financial advisor–client relationship

And this is a big deal, because Stifel’s business, like many financial firms, is built on relationship-based advisory.

Jim shared a reality many people overlook:

Even if AI can optimize a portfolio…

Even if AI can forecast outcomes…

Even if AI can generate a recommendation…
People still want to talk to someone.

Especially when the decisions involve:

• Life savings
• Retirement
• Family needs
• Market uncertainty
• Emotional stress

In those moments, AI isn’t just limited by intelligence.

It’s limited by trust.

The Better Future: AI as an Advisor’s Co-Pilot

Jim and the host aligned on a smart middle ground:

AI probably won’t replace the advisor.

But it can absolutely support the advisor.

For example, AI can help with:

• Preparing reports
• Summarizing performance
• Generating client-ready visuals
• Spotting trends
• Pulling insights faster

That makes the human conversation better, not obsolete.

This is the real enterprise AI win:
augmentation, not replacement.

Final Thought: AI Won’t Shrink the World Unless We Build the Foundation

The conversation began with a fun reminder that the world is small, a shared connection through the University of Missouri.

But it ended with something more meaningful:

AI is powerful, but it won’t transform finance through hype.

It will transform finance through:

• Clean data
• Governed systems
• Practical workflows
• Proven use cases
• And careful boundaries around trust

And for finance leaders, that’s a reassuring message.

Because it means the future isn’t chaos.

It’s a step-by-step evolution, led by people who understand both the technology and the business.