How Credit Unions Build the Board Business Case for AI in Lending

How do credit unions build a business case for AI in lending? They scope one workflow, price it from their own file volume, and set a baseline and a success metric the board can check. Funding is then released in stages against measured results, so the CFO approves a number with an exit.

The money is moving. In Jack Henry’s April 2026 survey, 48% of financial institutions named AI as their top planned technology investment and 88% expected to raise technology budgets within two years. Yet in Bank Director’s 2025 technology survey, just 18% of respondents said their bank measures return on investment for technology projects, and 68% said it does not.

‍

That gap is where a lending request stalls. Most credit unions have more AI ideas than budget, and for credit unions entering the 2027 budget process, the way you scope, price, and package a request decides which AI-powered automation gets funded.

How Do Credit Unions Build a Business Case for AI in Lending?

A lending business case answers five questions: which workflow, what it costs today, what it costs with AI at your volume, which metric and date prove it worked, and who owns it. AI for credit unions earns funding when each answer fits on one page.

‍

What is AI Use Case Scoping?

AI use case scoping is the step between a list of AI ideas and a budget line your board will approve. It sets where the workflow starts and stops, what it costs today and with AI, and how you will know it worked. It names the AI technologies involved, such as machine learning or generative tools, and what the AI systems will not do, because a fixed boundary makes two prices comparable. Before implementing AI, start with identifying AI use cases, then scope, then compare vendors and pricing models.

Why Do Long AI Idea Lists Stall at Budget Time?

Cross-functional teams produce a long list: lending wants document collection, document processing, and manual work cleared, and member service wants to streamline time-consuming tasks in member communications. Risk aims to reduce fraud losses through transaction monitoring and fraud detection.

The Federal Reserve’s payments team describes checking application data for inconsistencies as a way to spot synthetic identities. None has been priced against the others. Spreading money across all of it creates pilot sprawl. Deloitte’s 2026 State of AI in the Enterprise survey of 3,235 leaders found that only 25% of organizations have moved 40% or more of their AI pilots into production.

‍

AI adoption is accelerating. Cornerstone Advisors reports that 59% of credit unions have already deployed generative AI, and America’s Credit Unions notes that younger members expect digital-first experiences. Boards will ask for a number.

How Do You Cut 50 AI Ideas Down to Three or Four?

Run four cuts in order. Each removes ideas for a stated reason and parks them for the next cycle. The strategy cut ties the list to business objectives. The baseline cut checks data readiness; our look at whether you need clean data for AI explains what “ready” means. The dependency cut exposes loan origination system and core connections before they reach a statement of work. The risk cut covers risk assessment, regulatory compliance, fair lending compliance, data privacy, and sensitive member data. Federal regulators, including NCUA, have said alternative data has the potential to expand access to credit, which puts fair lending review within scope.

‍

NIST’s AI Risk Management Framework (Govern, Map, Measure, Manage), CISA’s guidance on verifying the integrity of AI training data, and COSO’s February 2026 guidance on internal control over generative AI give the risk cut its checklist. Then balance the survivors into three or four: one that shows a measured result in its first quarter, one or two in adjacent workflows, and at most one longer bet. Our guide to picking your first AI workflow covers that choice.

What Goes on a Use Case Scope Card?

Write one page for every surviving idea, so every AI vendor prices the same boundary, such as what counts as a finished loan file. The handling time and exception rows reflect the current cost: loan processing steps such as data entry from pay stubs, bank statements, and tax returns; document chasing via phone calls; and rework due to human error.

Say which tasks document AI or AI agents take on and where a person reviews any member-facing step. Run the card past your vendor due diligence questions. When a credit union relies on third-party models, a cross-functional AI governance committee should keep the model inventory and ongoing monitoring, and document its decisions as internal controls.

‍

How Do You Price an AI Use Case From the Bottom Up?

Scope drives price: monthly volume, document types, systems the AI must read and write, and exception handling. Build the cost component by component: platform fees, usage at your volume, implementation, staff time, training and governance. Plan for more than one refinement cycle, because a model that flags too many false positives adds review work. Then write the value side as arithmetic:

• Hours returned: minutes saved per unit, multiplied by annual units, divided by 60.
• Labor value: hours returned multiplied by loaded hourly cost.
• Everything else: rework avoided, faster turnaround, and capacity for more volume without new hires.

‍

AI tools that boost productivity and streamline operations belong here, with operational efficiency. Because credit unions serve members, frame the return as member impact: turnaround time on loan approvals, member satisfaction, and staff time freed from data processing for member conversations and member interactions. Those protect member relationships and the member experience, and measurable returns rarely stop at cost reduction.

How Should AI Budget Planning Stage Three or Four Use Cases Across 2027?

• Fund the first use case in full. Approve its complete cost at the budget meeting.
• Approve the rest in principle. Release funding when the first hits its success metric, and carry its valuable insights into the next scope cards.
• Write kill criteria in advance. Name the result and the date that stops funding.

A staged AI implementation budgets for production from the start.

What Does the Champion Forward to the CFO and the Board?

The employee who champions a new technology, such as AI-driven tools, is rarely the one who signs off on it. The CFO, the board, and the risk lead each read the request independently, so the pack has to answer their respective first questions on their own pages and support the board’s decision-making. On our calls, champions often tell us they lack a script for taking a request upward and want a one-pager the CFO can read without them.

‍

What Should the Board Ask Include?

The NCUA Examiner’s Guide states that “the board of directors reviews and approves the business plan, including a budget, in the context of its consistency with the credit union’s strategic plan.” That sentence, part of the existing supervisory framework, is the test for every AI line.

Smaller institutions and larger ones meet it the same way. NCUA’s 2026 supervisory priorities letter, 26-CU-01, does not mention AI or artificial intelligence and itself sets no AI-specific regulations, so the board approves an AI line the way it approves any other budget line. Boards are already asking, as Jason Stverak of the Defense Credit Union Council describes:

“I found even at the smallest credit union, there's always somebody who … they know that their local credit union needs to make it a part of their discussion, and they're bringing it up at board meetings. And they're like, I want this on the agenda for discussion. Hey, CEO, what is your team doing to make sure that we are on the front of this wave and not the back of it, and we're not getting the benefits?” — Jason Stverak, Chief Advocacy Officer, Defense Credit Union Council

Give the board one page per use case with the key information in seven items: the baseline, the scope, year-one and ongoing cost, the value arithmetic, the controls (human review point, vendor due diligence status, fair lending review, applicable laws, regulatory requirements and handling of sensitive member data), the funding release condition with kill criteria, and a named owner. Add one portfolio page with the total ask, the sequence, and what you parked.

What Does This Look Like in Practice?

With AgentFlow, scoping starts with your volume: we fill in the scope card with your team, run a sample of your documents, and price the use case per unit. At FORUM Credit Union, AgentFlow reached 99% document classification accuracy across 62 loan packages and 99% data extraction precision.

What makes AI worth funding is a number the board can check. Bring us the three or four AI use cases on your list. We will scope each one with you and hand back a per-use-case cost. If you want the framework first, take the board-ready business case for credit union lending automation.

‍

Frequently Asked Questions