There’s no shortage of conversation around artificial intelligence in banking. Every conference agenda, boardroom discussion, and vendor pitch seems to promise transformation. But as the recent Baker Hill webinar “AI in Lending: What’s Real, What’s Hype, and What Actually Works Today” made clear, progress with AI isn’t about bold promises — it’s about disciplined execution.
Hosted by Rob Foreman, SVP of Client Strategy and Growth at Baker Hill, the webinar brought together a practical panel to cut through the noise and focus on evidence. Joining Rob were Josh Spisak, Director of product Management at Baker Hill, and Todd Cooper, CEO and Co-Founder of NuArca, each offering grounded perspectives on where AI is delivering results today — and where caution is still warranted.
The Reality Check: Ambition Outpaces Execution
A key statistic set the tone for the discussion: although most bank executives view AI as essential to their strategy, only a small fraction have achieved AI maturity at scale. The gap isn’t caused by a lack of interest or investment, but by the challenge of moving from pilots to production — where measurable ROI depends on focus, strong governance, and clarity around AI’s role.
The institutions pulling ahead aren’t chasing every new model or headline. They’re starting with proven use cases, measuring outcomes rigorously, and scaling what works.
Where AI Is Working Today
Some of the strongest, most proven applications of AI in lending are happening in document processing and data intelligence. Lenders deal with endless unstructured documents, from tax returns to credit memos, and AI shines here by automating the routine work — classifying files, pulling out key data points, and dramatically cutting down manual effort.
These gains aren’t theoretical. In production environments, document intelligence has delivered dramatic cost reductions, shortened cycle times, and freed analysts to focus on higher-value work. The pattern is consistent: high-volume, repetitive tasks with clear success metrics are where AI delivers the fastest and most durable returns.
AI as an Augmenter, Not a Replacement
Success in AI-enabled lending doesn’t mean removing people from the process. Fully autonomous lending — especially for complex commercial or CRE scenarios — remains far from practical. Today, the real value comes from using AI to enhance human expertise, not replace it.
In underwriting, that means using AI to surface insights, incorporate alternative data, and accelerate analysis, while keeping experienced lenders in the loop. Banks applying AI this way are seeing faster decisions, improved risk differentiation, and better borrower experiences — without sacrificing sound credit discipline.
Governance Isn’t Optional — It’s the Enabler
Governance and compliance emerged as a critical theme, especially given how many institutions still hesitate in this area. The message was clear: existing regulatory frameworks already apply to AI, and there are no shortcuts or “black box” exceptions.
Explainability, data lineage, privacy, and third-party risk management must be built in from day one. When they are, AI adoption becomes far less intimidating. In fact, AI can strengthen compliance functions by reducing noise, accelerating regulatory change management, and allowing compliance professionals to focus on strategic oversight instead of manual review.
The institutions succeeding with AI aren’t avoiding governance — they’re using it as a foundation.
Why So Many Pilots Stall
Another recurring theme was why AI initiatives so often fail to reach production. Data integration challenges, inconsistent data quality, underestimated change management, and talent gaps all play a role. Many pilots work in isolation but struggle when connected to real systems and real workflows.
The takeaway: technology alone isn’t enough. AI adoption requires operational readiness, clean and governed data, and teams that understand how to work alongside intelligent systems.
A Practical Path Forward
Instead of wrapping up with high-level theory, the discussion shifted to a practical 90-day roadmap. The approach starts with understanding what’s already in place, taking an honest look at data readiness, and establishing cross-functional governance early. From there, the focus is on selecting a few high-impact, low-complexity use cases — often in document processing or workflow automation — and defining clear success criteria from the start.
This crawl-walk-run approach doesn’t just reduce risk. It builds trust internally, with regulators, and with boards.
The Bottom Line
AI in lending is no longer theoretical, but its real value shows up when institutions prioritize discipline over flashy ambitions. The most successful teams treat AI as an extension of established banking practices, applying it in ways that enhance — not replace — human judgment. By focusing on measurable outcomes, establishing governance early, and scaling what works, institutions move beyond experimentation and toward sustainable competitive advantage.
The question now isn’t whether to adopt AI, but where to start — and how intentionally you’re willing to execute.
