AI is now firmly on the lending agenda. But before lenders start choosing platforms, running pilots or comparing vendor claims, they should step back and ask a more important question: what are we actually trying to solve?
Too many firms begin with the technology. They look at tools, demonstrations and features before they have properly defined the operational challenge. For lenders, that can lead to activity without impact.
The starting point should be the business problem. Where is work getting stuck? Where are teams spending time on repetitive administration? Where are service levels being affected by delays, incomplete information or avoidable back-and-forth?
In lending, the biggest AI opportunities are often found in the least glamorous parts of the process. Much of the operational drag still sits in the gather, check and chase cycle: gathering documents, checking whether they are complete, fixing errors and chasing missing information.
This is important because incomplete or poorly packaged cases affect underwriting capacity, broker relationships, turnaround times and the consistency of service.
The next question lenders should ask is: are we thinking about AI in the right way?
AI should not be viewed only as traditional software that automates isolated tasks. Modern AI, particularly agentic AI, is better understood as something that can support bounded workflows. It can follow rules, work through information, identify gaps, produce outputs and hand off to humans when needed.
In that sense, adopting AI is closer to bringing a new operational colleague into the business than installing a conventional system.
If a lender hired someone into an operations team, it would explain the outcome required, the policies to follow, the quality standards expected, the escalation points and the controls around the role. AI needs to be framed in a similar way.
That leads to another critical question: how mature is our understanding of AI?
Some lenders are still at the education stage. Some have experimented with generative AI tools. Others are already considering how AI could become part of their future operating model. Each position is valid, but the next step should reflect where the organisation really is.
Do teams understand the difference between a basic chatbot and an outcome-driven AI agent? Has the business considered governance, oversight, security, auditability and deployment in a live lending environment? AI in lending cannot sit outside the standards and controls of the business.
With this in mind, lenders should then ask a deeper question: who is accountable for the results?
AI can increasingly perform work that previously required people. It can analyse, extract, summarise, communicate and progress operational tasks. But it cannot currently take accountability for whether the wider system is delivering the right outcomes, whether risks are being managed properly, or whether priorities are being set in the right order.
A lender may be able to build an impressive AI proof of concept quickly. But building something flexible, adaptable, resilient, scalable and secure enough for live operations is a different undertaking. It needs continual design, implementation, testing, monitoring, optimisation, security management and governance.
At that point, a lender isn’t just adopting a tool. It’s taking on a technology operating capability. For many firms, that could mean needing several specialist roles to run it properly.
So, the choice becomes clearer: does a lender want to recruit and manage those specialisms internally, or work with a partner that has already built the expertise, infrastructure and accountability structures required?
Lenders should also ask: are we starting in the right place?
The best starting point is often a specific administrative workflow where the pain is obvious and measurable: intake, document collection, case packaging or pre-underwriting checks.
Finally, the commercial case must stack up. How much manual work is being done today? Where are the delays? What improves if the process becomes faster, more consistent and less resource-intensive?
The lenders that get the most from AI will not be those making the most noise about it. They will be those asking better questions at the start and applying the technology where it can make a genuine operational difference.
