Grant Thornton’s 2026 AI Impact Survey identifies workforce enablement, data readiness and process redesign as material constraints on AI value. Training was the most underfunded AI investment area among finance leaders in the survey, while many technology leaders reported that fewer than half of their core applications were AI-ready. For lending operations, this points to a gap between automating routine tasks and giving people the context, controls and tools required to resolve non-standard cases.
Quick Answer
The intelligence gap in lending operations is the difference between executing routine, rule-based tasks and supporting the people who resolve exceptions, restructurings and cross-product cases. Automation can route or summarize these cases, but accountable staff still need reliable context, explanations and approval controls. Bottlenecks persist when an exception leaves the automated path and its history must be rebuilt manually.
What “Automation” Has Actually Solved
Many lenders use straight-through or highly automated processing for selected standard cases. Document extraction, bureau pulls, eligibility checks and disbursement steps can require less manual handling when data and policy conditions are clear. Coverage varies by institution, product and risk policy, so it should be measured rather than assumed.
This can shorten processing time for eligible cases. It can also concentrate operations work in exceptions, where staff need more context and discretion than standard workflows provide. Institutions should track the actual straight-through rate, exception rate and time to resolution by product rather than rely on a general industry percentage.
Where the Intelligence Gap Shows Up
Grant Thornton’s 2026 survey found that training is underfunded for many finance leaders and that data readiness limits AI performance. Frontline employees and middle managers were identified as the groups needing the most implementation support. These findings reinforce a practical point for lenders: AI value depends on redesigned workflows, role-specific guidance and reliable data, not deployment alone.
That pattern maps directly onto lending operations:
- Exception handling still routes to humans, slowly. A borrower requesting a tenure change, a hardship modification, or a manual override on a flagged application typically exits the automated path entirely and lands in a queue, often without the context that would let a person resolve it quickly.
- Cross-product and cross-system cases fall through automation boundaries. Automation built for one loan type or one system rarely extends cleanly to a borrower whose situation spans products or legacy platforms.
- Workforce readiness lags the technology rollout. Institutions report deploying AI and automation tools faster than they build the training and workflow redesign needed for staff to use them effectively on non-standard cases.
- Governance and explainability requirements slow expansion into judgment-heavy areas. It’s one thing to automate a document check; it’s another to automate, and defend to a regulator, a decision that required interpretation.
Automated Operations vs. Genuinely Intelligent Operations
| Dimension | Automated (rule-based) | Intelligent (judgment-capable) |
| Standard applications | Automated for eligible cases, with exceptions routed for review | Automation plus monitored recommendations informed by outcomes |
| Exceptions and edge cases | Routed to a manual queue, often with limited context | Routed with relevant history, explanation and optional recommendation |
| Cross-product situations | Frequently breaks or requires manual reconciliation | Supported by a consolidated, permissioned borrower view |
| Workforce role | Staff process what automation cannot handle | Staff review flagged cases with contextual decision support |
| Governance | Compliance checks applied after the fact | Explanation, approval and audit evidence captured in the workflow |
What to Look for When Assessing Lending Operations Maturity
A few questions separate operations that are automated from operations that are genuinely closing the intelligence gap:
- What share of exception cases still requires a person to reconstruct context manually before they can act on it?
- Does the automation stack extend across products and systems, or does it stop at the boundary of a single platform?
- Has staff training kept pace with the automation and AI tools already deployed, or is adoption outrunning readiness?
- Can the system explain why a case was flagged or how a recommendation was reached, in terms a regulator or auditor would accept?
- Is exception volume tracked and analyzed over time, or treated as background noise once it’s off the main dashboard?
Bottom Line
Automation can handle predictable tasks efficiently, but exceptions and cross-product cases still require context, explanation and accountable judgement. Closing the intelligence gap means redesigning exception workflows, improving data access, training staff for specific decisions and monitoring where AI recommendations help or fail. More automation is useful only when it improves an observable operational outcome.
Frequently Asked Questions (FAQs)
It is both. Grant Thornton’s research highlights underinvestment in training and change enablement, while also reporting weak AI readiness across many core applications. Lenders therefore need to address workflow design, skills, data and technology together.
Not automatically. Automating additional steps tends to reduce volume in the routine category without necessarily improving how exceptions and judgment-heavy cases are handled. The latter require a different kind of system design, not just broader coverage of the same rule-based approach.
Exception handling, restructurings, and any case that spans multiple products or systems tend to be hardest. These require context and judgment that most automation stacks were not originally built to carry.
