SAP

Erin Earl is a Financial Services Sector Executive Advisor at SAP with a focus on maximizing operational efficiency to boost revenue through data-based decisions. Her extensive experience in software operations, marketing and sales, goto-market strategy, product management, treasury operations, AP automation, and working capital management. Working across organizations to build strategies and solutions to maximize revenue growth, she has successfully led organizations through turbulent times to financial stability and growth with a strong understanding of financial reporting, operations management, and compliance.

Midsize banks find themselves sandwiched between two classes of competition, and both have a leg up on the AI-implementation front. The behemoths have deep pockets and their own large language models. Fintechs, born in the cloud, are already AI-ready.

To compete, midsize institutions need a disciplined, ROI-first AI strategy that prioritizes efficiency gains, data readiness, and scalable deployment.

A midsize bank’s AI needs an ROI

Start by insisting that every AI use case has an ROI-positive business case.

Identify business bottlenecks. Identify profit leaks in operations, CX, or underwriting, then apply AI where ROI is measurable. Ask, “What are my biggest drags on profitability, customer experience, and internal efficiency?” Then work backwards from potential ROI wins to AI technology solutions.

Focus on improving efficiency ratios, which PwC estimates AI can boost by 15 percentage points. A three-phased approach might look like this:

1) Quick AI wins with low-risk, high-ROI automation

These include internal process automation in areas such as loan-documentation processing and back-office reconciliation, and also in compliance applications such as Bank Secrecy Act and AML alerts. These are low risk, and they boost efficiency, cut labor costs, and are clearly explainable to a regulator.

2) Revenue expansion via customer intelligence

AI helps predict churn and improve product recommendations through a better grasp of where customers are in their lifecycle. Midsize banks can punch above their weight here. For example, with AI, a community-based bank can assess the fluctuations of a small business’s cash flow and suggest a working capital line of credit to better navigate seasonal dips. AI increasingly enables the sort of consultative selling that brings the right products to customers at the right time.

3) AI in loan underwriting

In credit underwriting, AI can extract data from tax returns, bank statements, and financial statements; read unstructured documents such as PDFs and emails; standardize financials and calculate key ratios; automate risk assessment, looking for inconsistencies across filings or suspicious patterns to cut fraud risk; generate credit memo drafts; and automate workflows and document tracking for loan applications.

“We compete by demanding ROI positive AI focusing on efficiency ratios, unified data, and governance, ensuring innovation scales responsibly while people remain the most vital intelligence.”

Underwriting is a core strategic activity, one that hinges on having your data house in order (more on this below). Assume a 24-to-36-month implementation for most midsize banks.

To exploit AI, a bank must first understand its process flows and data architectures. Technology tools are indispensable here. Two in particular: enterprise architecture management software to determine and visualize what systems the bank is running, and business process management software to do process modeling and mining to show how business processes work and fit together.

Take AI banking demos with a big grain of salt

Working backwards from wins to AI also helps midsize banks avoid the two biggest AI-strategy mistakes I’ve observed. The first is being lured in by vendor pitches and demos. Midsize banks should be looking at AI to improve basic functions only after deliberately and strategically prioritizing fundamental needs.

With demos, you should ask: Does it integrate with your core system(s)? What happens if the model goes astray and a regulator wants to know how it happened? And, vitally, who owns the model’s governance?

AI governance considerations should be core to a midsize bank’s AI strategy. Regulators have been very clear that, if you’re using an AI model, you must be able to explain the decisions it makes, audit it, and demonstrate that consistent input scenarios deliver consistent outcomes.

“Pilot purgatory” also applies to AI in banking

The second biggest AI-strategy mistake I’ve seen is getting stuck in “pilot purgatory.” Why? Because pilots don’t scale, often because the bank hasn’t gotten its data in order. AI needs relevant, reliable, responsible data to deliver real value. The legacy systems of most midsize banks don’t lend themselves to unified data models.

Unified data models are a precondition for effective AI

Fortunately, enterprise data analytics platforms, a.k.a. data fabric platforms, can provide a virtual unified data model atop existing systems. These platforms have grown increasingly attractive to midsize banks. And there are other approaches, including data warehouses and data lakes. The key is, unified data is increasingly a precondition for competitive advantage.

People will make or break AI in banking

Finally, no system will pay off if the team doesn’t use it. Change management remains vital, and that starts with positioning AI as a workday ally rather than another system imposed. Synovus is one example of a midsize bank that’s getting AIfocused change management right.

Midsize banks may not have the resources of the giants or the clean-slate architectures of the fintechs. But they can harness AI for real competitive advantage. Insist on an ROI for all AI. Look for wins with an eye on efficiency ratios. Prioritize AI governance early. Get the data house in order. And ensure employee buy-in, because the most vital intelligence at any bank is still human.