Published
A commercial loan is approved on tuesday and funded two weeks later because of booking delays caused by a missing covenant schedule. Borrower was waiting for the funds.
A payment investigation passes through four teams and takes a week to close. There is no fraud. No processing error. Just fragmented data, disconnected workflows, and handoffs that nobody owns.
A mortgage offer expires before settlement because a flood-risk overlay is identified after credit approval rather than before it.
None of these failures stem from a lack of technology. The banks involved have AI strategies, successful pilots, and transformation roadmaps built around scale. Yet a stubborn gap remains between completing a process step and delivering an outcome.
That gap is where costs accumulate, risks emerge, and customer experience deteriorates. It's also where a significant share of the banking industry's investment in transformation has struggled to deliver its promised value.
In a recent interview with CNBC, Genpact CEO Balkrishan "BK" Kalra describes the challenge simply: ambition is high, but readiness remains low. The era of proofs of concept is ending. The era of experimentation is giving way to a more demanding challenge: operationalizing AI at scale.
In banking operations, this challenge is the last mile, the context.
What the last mile and context really mean
The last mile is the portion of work that AI alone cannot fully resolve. Not because the technology is inadequate, but because the required knowledge is not contained within a foundation model. It exists inside the institution itself.
The context is the product nuances, process complexities, data quality, and legacy platforms.
It's the exception a credit officer instinctively escalates. The investigation that appears routine until a third system contradicts the first two. The local regulatory nuance, process variation, or customer circumstance that never made it into a policy document because experienced teams already understood it.
The first wave of AI has made the initial 80% of work increasingly accessible. Agentic capabilities are becoming increasingly available across the market. The real differentiation lies in everything that follows.
Just as importantly, the last mile and context require accountability.
Agentic operations shift work from human-processed and human-validated to machine-processed and human-validated. The processing changes. The accountability does not.
An expert still signs off on exceptions. A business owner still exercises judgment. A regulator still expects someone to explain the outcome.
The last mile, therefore, is where AI, process intelligence, and human expertise intersect. It's where exceptions are resolved, decisions are operationalized, and enterprise accountability remains intact.
There is little value in artificial intelligence without process intelligence. The two must work together.
What the last mile looks like in practice
Commercial lending
The challenge is that commercial lending is not a single process. It varies by asset type, industry, loan structure, geography, and ticket size. That complexity creates the operational judgment layer that generic AI cannot easily replicate. Reading and interpreting a complex, negotiated credit agreement requires much more than just extraction automation.
Retail banking and mortgages
Banking value is not created when a loan is approved. It's created when an approval becomes a funded loan, serviced correctly, and maintained without downstream risk or rework.
Many AI-led workflows struggle not because the models are ineffective, but because operational exceptions were never designed into the process.
The answer is not a more powerful model. It's a more context-aware operating model.
The same principle applies to localized risks and property-specific considerations. Flood exposure, heritage restrictions, environmental overlays, and other regional factors are often identified late in the process, creating expensive delays and avoidable rework.
Organizations that address those exceptions upstream achieve improvements that extend beyond efficiency. They reduce cycle times, improve customer experience, and increase the predictability of operational outcomes.
Why the last mile is difficult to replicate
Many organizations continue to face challenges operationalizing AI through the last mile because the required process context, governance, and exception handling can be difficult to replicate at scale.
Different participants in the AI ecosystem often bring distinct strengths, whether in model development, technology implementation, platform capabilities, or operational delivery. Operationalizing AI at scale typically requires combining these capabilities with deep process knowledge, governance, and exception management.
The result is that many transformation programs improve decisions without fundamentally improving execution.
One important differentiator is institutional process knowledge at scale. It's accumulated through years of managing exceptions, navigating regulations, resolving edge cases, and understanding how work actually flows through an organization.
Industry recognition can reflect a range of factors, including operational expertise. Market leadership is rarely earned through technology alone. It's earned through the ability to consistently deliver outcomes where operational complexity is highest.
Where banks should start
Three priorities matter most.
1. Stop treating technology as the problem
Most banks do not need to replace their existing technology estates.
Their environments have evolved over decades and contain critical institutional investments. The opportunity lies not in wholesale replacement but in orchestration: creating a layer of intelligence that connects AI with business context, governance, and operational decision-making.
2. Address more than technology debt
Technology debt is only one constraint.
According to Genpact research, data debt, process debt, and talent debt sit alongside it. Agentic AI exposes all three rapidly.
That's why governance cannot be an afterthought. It must be built into the operating model from the beginning. Business leaders, technology leaders, and data leaders need alignment early, because scaling AI is ultimately an enterprise transformation challenge rather than a technology deployment exercise.
3. Start with the friction
The best opportunities are usually easy to find.
They appear in delayed decisions, recurring exceptions, repeated handoffs, unresolved investigations, and processes that require constant intervention.
Map where work slows down. Identify where rework accumulates. Understand where decisions sit waiting for context.
That map is the last mile.
Then evaluate every AI initiative against a simple test:
Does it help the bank grow faster?
Does it help the bank operate leaner?
Does it accelerate the conversion of decisions into outcomes?
If the answer is no, it may still be an interesting experiment. But the industry has reached the point where experiments alone are no longer enough.
The next competitive battleground
The next generation of banking leaders will not be differentiated by the number of AI pilots they launch. They will be differentiated by their ability to operationalize AI where it matters most: in the messy, exception-driven reality of execution.
