The Lending Brief

Welcome to The Lending Brief,

For the next three weeks, we’re running special-focus issues on AI. We start today with agentic AI. Next week I’ll be at FinovateFall in New York - if you’ll be there, drop me a line. I’ll share conference takeaways in the following newsletters.

This week: Two insights. Three minutes. Zero fluff. 

Agents, Not Ads: How Borrowers Will Find Your Loans

AI assistants (“agents”) are already moving money for consumers. They don’t care about brand; they optimize. McKinsey argues this will squeeze deposit spreads and card economics as agents sweep idle balances to higher-yield accounts, or route spend to the best rewards.

You can see the deposit shift happening already: MaxMyInterest moves consumer cash among linked banks to chase the best rate, Wealthfront Autopilot transfers surplus funds automatically when checking exceeds a threshold, and Raisin funnels tens of billions into higher-yield savings across hundreds of FIs - optimization at scale even before full “agents.”

🤔 What does this mean for lending? 
Agents will pre-shop for borrowers: pulling your publicly posted rates and fees, reading lock/turn-time rules, and ranking “cleanest terms + fastest clear-to-close.” If those facts aren’t easy to find and easy for software to read, you’ll be invisible to both humans and assistants. Financial Brand urges FIs to prepare forgenerative engine optimization” (GEO): write clear product pages that AI tools can reliably parse. SouthState Bank goes further - arguing for agent-friendly pages (or sections) with simple markup and consistent data so agents don’t get lost in carousels and pop-ups.

Quick test this week: paste your SBA/term-loan page into an AI tool and see if it can answer eligibility, fees, docs, and timeline in 30 seconds. If it can’t, borrowers’ agents won’t ‘see’ you either.

💡 Will FIs also use agents? 
A new survey shows 93% - 31% of which have $1-$9B in assets - of financial institutions plan to adopt agentic AI within one to two years, focusing first on fraud detection, KYC maintenance, onboarding automation, and transaction monitoring - with humans in the loop. Respondents expect an average of about $3.04M in annual savings in compliance operations from fewer manual steps and faster cycles.

Action Step: 90-Day Plan to Get Agent-Ready (for Lending)

Be agent-findable:
Agents skip vague pages. Publish simple, accurate rates/fees, eligibility, timelines, and a downloadable rate sheet so assistants (and people) can compare you fairly.

Reply GEO for the one-page website checklist.

Test it where the risk is low.
Start small: missing loan documents, or AML alerts. Let agents do the repetitive steps, and keep staff for the final approval.

Keep humans in the loop.
Just as you track staff permissions, keep a log of every agent, what it can do, and who manages it. Build in checkpoints so humans can override big or unusual actions.

Reply ‘CHECKLIST’ for the 90-day agent-ready plan.

Why AI Pilots Fail (This Time It’s People)

Last week was about scope. This week is adoption.

OpenAI just learned this lesson the hard way. After a recent ChatGPT update, they did the logical thing: defaulted everyone to the new model and removed the old options. The new version automatically picks the best model for each query, saving users money and giving better results. Users hated it. Why? They felt like something was being "taken away" from them.

💡 The psychology? 
Behavioral economists call this the endowment effect - we hate losing things we feel we "own," even when the replacement is objectively better. ChatGPT users were paying for GPT-4 access, and suddenly it was gone. The fact that the new version was superior didn’t matter.

🤔 What does this mean for bank AI rollouts?
Your staff exhibits the same bias. When you replace their familiar loan processing workflow with an AI assistant, they don't see "better technology" — they see their expertise being devalued. Even if the AI handles routine tasks so they can focus on relationship building, the immediate reaction is loss, not gain.

🚨 The real problem?
Most FIs announce AI changes rather than preparing teams for them. Change feels imposed rather than chosen.

Action step:
Before rolling out any AI tool, show staff what stays the same. Frame it as "AI handles the paperwork so you can spend more time with borrowers" rather than "AI will improve efficiency." Let teams pilot the tool voluntarily before making it mandatory. A simple heads up like "we're testing new document review software next month" reduces resistance significantly.

Don’t rip away the old path overnight
Run “Classic” and “Faster (assisted)” side-by-side for a few weeks with a clear banner and a sunset date. Let teams try it before it’s mandatory. That’s essentially what OpenAI had to do after backlash.

Show the effort, not just the magic.
Have AI narrate its work: “Checked 12 docs, found 3 date issues, pre-filled 5 forms.” Visible effort raises perceived value and shortens the trust curve. (Users discount invisible “magic.”)

🔍 Want to go deeper?
Check out this 7-minute video that breaks down the endowment effect.

🙌 Help Us Grow

Know someone in lending who’d benefit from this? Forward the brief - or hit reply if there’s a topic you want covered.

📅 Next newsletter drops Tuesday, 09/09.

Warmly,

Sandra Wasicek
Founder & CEO ZorroFi