There's a version of AI adoption happening inside financial institutions right now that looks bold from the outside.
New tooling. New vendors. New budget lines. Press releases about "AI-powered" everything.
And yet the results are underwhelming. Pilots stall. Adoption is low. ROI is hard to measure. The technology gets blamed — and in most cases, the technology isn't the problem.
The problem is the question being asked.
When organizations approach AI, the default instinct is optimization. Take an existing process — a form, a workflow, an approval chain — and make it faster, cheaper, or lighter. The structure stays the same. The AI just removes some of the friction. This is the wrong problem to solve.
Not because efficiency doesn't matter. It does. But because AI's real capability isn't in accelerating old workflows — it's in replacing interactions that were never designed to scale in the first place.
In lending, that interaction is human. It has always been human.
A customer sitting down with a loan officer. Walking through what they can afford, what they qualify for, what their options are. Getting a recommendation they can trust, from someone who understands their situation. That conversation is what moves people from curious to committed — and it has never scaled well. You can't have that conversation with a form. You can't replicate it with a comparison table.
That's where AI creates genuine value. Not in making the form digital. In replacing the form entirely.
There's a version of AI deployment that's become common in large organizations — the copilot model. Give an AI tool to the person doing the job. Let it assist. Keep the human in the loop.
It sounds responsible. In practice, it often fails.
The person using the copilot quickly realizes the copilot is better at the job than they are. So their feedback to management is that the copilot doesn't work. The rollout stalls. The technology gets blamed again.
This isn't a technology failure. It's a deployment failure — the result of avoiding a hard organizational decision by half-implementing a transformational one.
The organizations getting real results from AI aren't mixing it with legacy workflows. They're running AI fully in parallel, measuring it against traditional channels, and making clear-eyed decisions based on what the data shows.
Part of why the wrong questions keep getting asked is that the frame of reference is wrong.
Most AI deployments are still being designed in the logic of the digital era — take something that was paper-based, make it digital, make it faster. Mobile apps replaced branch visits. Online forms replaced paper forms. The underlying assumption stayed the same: the customer comes to the institution, fills something out, waits.
The AI era breaks that assumption.
The interaction model changes completely. Instead of a customer navigating a structured workflow you designed for them, they have a conversation. Contextual. Personalized. Guided by what they actually need in that moment — not by what your process requires them to provide.
This isn't incremental. It's structural. And organizations that treat it as an incremental improvement will consistently underinvest in the right places.
At PILLAR, we build AI infrastructure for financial institutions — not to automate their existing workflows, but to scale the human interaction that existing workflows were never able to replicate.
In practice, that means deploying product-aware agents that can qualify a lead, walk a customer through their options, and drive a loan submission — the same way a skilled advisor would, at the scale a skilled advisor never could.
Not a chatbot reading from an FAQ. Not a copilot assisting a banker. A purpose-built agent that understands the product, understands the customer, and knows what to do at every point in the conversation.
We've deployed this across markets with very different contexts — different languages, different regulatory environments, different customer behaviors. What stays consistent is the underlying problem: human interaction in lending doesn't scale. AI, done right, makes it scale. The technology isn't the hard part. The hard part is identifying the right problem to give it.
Before the next AI initiative kicks off, one question is worth sitting with: Are we using AI to make an existing process faster — or to do something that the existing process was fundamentally incapable of?
The first is useful. The second is transformational.
Financial institutions that get this distinction right will build AI that compounds. Those that don't will keep running pilots that look impressive and deliver little.
This is the conversation our Co-Founder, Ilya Kravtsov, had with Brendan Le Grange on How to Lend Money to Strangers. If you work in financial services and are thinking seriously about where AI creates real value — and where it doesn't — it's worth a listen.
→ Try PILLAR now: pillarlab.ai
