AI Lending That Feels Human for Credit Unions

Most credit unions are leaving high-quality loans on the table.

Not because members aren’t creditworthy, but because legacy lending models can’t see them.

Here’s the thing about AI in lending: done well, it doesn’t replace your lending team or your member-first philosophy. It extends both. That’s the core message from Jeff Keltner, SVP of Business Development at Upstart, in his conversation on The CUInsight Network—and it fits squarely into our AI for Credit Unions: Member-Centric Banking series.

This matters because your members’ expectations have changed faster than your lending tech stack. They want fast, fair, mobile-first decisions and still expect a human who knows their story when things get complicated. AI lending is one of the few tools that can serve both expectations at scale.

In this post, we’ll unpack what “AI lending with a human touch” actually looks like for credit unions, how it changes underwriting, and what you can do this quarter to get started responsibly.

---

AI lending for credit unions: what actually changes?

AI lending for credit unions changes three things at once: who you approve, how you price risk, and how your members experience the process.

Traditional scorecard underwriting is built around a handful of variables—credit score, DTI, income, loan amount. AI-driven underwriting, like the models Jeff Keltner describes at Upstart, can use hundreds of variables (often 800+), both traditional and non-traditional, to form a more accurate picture of risk.

That means you can:

• Approve more loans without increasing losses
• Identify “hidden prime” borrowers inside your subprime pool
• Offer instant decisions to qualified members instead of multi-day waits

“More people are creditworthy than we think.” – Jeff Keltner

For a member-centric institution, that’s a big deal. AI lending isn’t just about efficiency—it’s about inclusive lending and living your mission to serve more members, more fairly.

---

From FICO-only to 800+ variables: inclusive lending in practice

AI-based loan decisioning expands beyond FICO to understand a member’s full financial reality.

What goes into an AI lending model?

While every provider is different, modern AI lending platforms for credit unions often consider:

• Traditional variables: credit score, credit history length, utilization, DTI, inquiries
• Income & employment: level, stability, occupation, field of work
• Behavioral patterns: payment histories, trends over time, not just snapshots
• Application context: loan purpose, amount vs. income, existing relationships

When you combine 800+ variables, you get much better risk segmentation than a simple cutoff like “FICO < 660 = decline.”

The hidden-prime opportunity

Many credit unions carry a mental model like this:

• Prime: obvious approvals
• Subprime: mostly high-risk borrowers with a few exceptions

AI lending flips that script. You start to see a more nuanced picture:

• A 640 FICO member with stable income in a resilient field and strong recent payment history may actually behave like a 720 borrower.
• Another member with a 680 score but volatile income and rising utilization may be riskier than their score suggests.

The result? You can:

• Approve more of the good risks in your subprime pool
• Avoid overpricing or underpricing loans based on blunt cutoffs
• Expand access to affordable credit without mission drift

For a mid-sized credit union, even a 2–4 percentage point increase in approval rate at a stable loss rate can mean millions in additional annual loan volume, plus deeper member relationships.

---

Digital lending: stop digitizing a broken process

Most institutions don’t have a lending problem—they have a lending experience problem.

Jeff Keltner makes a crucial point: if you simply put your legacy loan process online, you’ve digitized friction, not removed it. AI lending for credit unions pays off when it’s paired with a reimagined end-to-end digital experience.

What a member-centric AI lending journey looks like

A modern, AI-enabled lending experience for your members should:

1. Start online and mobile-first
   Members can apply from their phone in minutes, with a clean interface and clear progress indicators.

2. Provide real-time decisions where appropriate
   For straightforward applications, AI underwriting returns an instant decision and rate with no branch visit.

3. Reduce document fatigue
   Instead of “upload your last three paystubs,” the model uses verified income data, payroll connections, or other trusted signals, asking for documents only when genuinely needed.

4. Escalate to humans when the situation is nuanced
   Complex cases, edge scenarios, or members who show risk but strong relationship history get routed to a loan officer who can apply judgment.

5. Close quickly with flexible options
   E-signature, digital closing, and transparent terms. For many members, the whole journey takes under 10 minutes.

The reality? Members now compare your lending process not to other credit unions, but to the experience of getting a ride, booking a flight, or ordering groceries from their phone. Friction feels like disrespect.

---

Keeping the human touch: where people still matter most

AI lending doesn’t replace your lending team; it refocuses them where they create the most value.

“Don’t underestimate the value of the human touch.” – Jeff Keltner

The credit unions that win won’t be the ones that automate everything. They’ll be the ones that automate the right things and humanize the rest.

Where AI should lead

AI and automation work best where:

• The application fits a well-understood risk profile
• The member clearly meets (or doesn’t meet) your risk appetite
• The request is simple: personal loans, auto refis, small-dollar lending

In these cases, automation:

• Frees staff from repetitive data entry and basic underwriting
• Provides consistent, explainable decisions
• Shortens cycle times and lowers origination costs

Where humans should lead

Your lending team is most valuable in:

• Edge cases: borderline approvals, unusual income streams, thin file borrowers
• Member advocacy: helping members understand decisions, options, and steps to improve
• Complex products: HELOCs, mortgages, business lending

Here’s a healthy operating model I’ve seen work:

• AI makes the first call on risk and suggests a decision
• Rules define thresholds for auto-approve, auto-decline, and “manual review” buckets
• Loan officers focus on that middle bucket—where their relationship skills and local knowledge matter

You end up with a lending operation that feels more human to members because staff finally have time to listen, explain, and advise.

---

Getting started with AI lending: a practical roadmap

If you’re a credit union executive or lending leader, the path to AI underwriting doesn’t need to be overwhelming. Start small, stay aligned with your mission, and keep members at the center.

1. Clarify your goals and constraints

Before you talk to vendors, answer a few hard questions internally:

• Are we aiming to grow loan volume, improve approval rates, reduce losses, or all three? In what order?
• Which products make the most sense to start with—personal loans, auto, refi?
• What’s our risk appetite and how much variance are we comfortable testing?

Clear goals will keep you from chasing shiny AI features that don’t serve your members.

2. Choose a pilot that can prove value fast

Pick a product with:

• High demand (e.g., unsecured personal loans)
• Painful friction today (long turn times, high abandonment)
• Room to expand approvals among members you’re currently declining

Measure hard outcomes:

• Change in approval rate at a given loss rate
• Average time from application to decision
• Member satisfaction or NPS for the lending process

3. Demand transparency and controls

AI lending for credit unions has to meet higher standards of fairness and compliance than fintechs aiming purely at growth.

When you evaluate AI lending platforms, insist on:

• Clear documentation of model inputs and governance
• Fair lending analysis by protected class proxies
• The ability to explain adverse actions to members in plain language
• Strong controls around overrides and policy settings

If a partner can’t support your compliance team’s questions, they’re not a fit for a member-centric institution.

4. Design the human experience with your team

Involve front-line staff early:

• Ask them which parts of the process should never be automated
• Co-design escalation flows, scripts, and guidance for manual reviews
• Train them not just on the tool, but on how to talk about AI with members

Members will ask: “Did a computer deny me?” Your staff need honest, confident answers that reinforce trust, not undermine it.

5. Iterate fast, but communicate faster

AI lending models improve over time. Treat your rollout as a continuous improvement program, not a one-and-done project.

Internally:

• Share outcomes with staff: approvals, losses, member feedback
• Celebrate the time you’ve given back to member-facing work

Externally:

• Reinforce that you’re using better data and smarter tools to say yes more often, not to automate members out of a conversation
• Highlight stories (with permission) of members who gained access to fair credit through the new process

---

AI lending as a pillar of member-centric banking

Within the AI for Credit Unions: Member-Centric Banking series, AI often shows up in conversations about fraud detection, chatbots, and financial wellness tools. Lending is where all of that comes together in a way members feel immediately.

Here’s the reality: the credit unions that thrive over the next decade will be the ones that treat AI as infrastructure for empathy—technology that lets them say yes more often, respond faster, and reserve human time for the moments that truly matter.

AI lending platforms like the one Jeff Keltner describes aren’t about replacing loan officers. They’re about rebuilding the lending experience so that:

• More members are recognized as truly creditworthy
• Fewer good loans are lost to slow, clunky processes
• Staff are free to act as advisors, not data-entry machines

If your credit union is serious about member-centric banking, AI-based loan decisioning shouldn’t be a “nice to have” experiment in 2026—it should be on your roadmap now.

The next step is simple: pick one lending product, one clear goal, and one cross-functional team. Then ask the only question that really matters:

How can we use AI to make this lending experience feel more fair, more human, and more aligned with our mission?