Read time: 8 minutes
Executive Insights
- Inconsistent lending is not a frontline behavior problem. It is a portfolio-construction problem, and it compounds quietly across credit cycles until it surfaces as examiner findings, reserve surprises, or capital planning failures.
- Decision noise, the variance that builds when officers, branches, and regions apply policy differently, raises default rates, distorts risk ratings, obscures concentrations, and erodes the reliability of CECL inputs.
- A study of more than 242,000 retail loan applications found that a one-standard-deviation increase in discretionary score manipulation raised default rates 12–16% relative to baseline and cost institutions an estimated 1.5 percentage points of ROE.¹
- The Federal Reserve may treat an internal rating system as unreliable when examiner downgrades exceed 10% of loans reviewed or 5% of dollars reviewed. That threshold is closer than most institutions realize.²
- More manual review layers have not closed the consistency gap. The evidence points to structured decisioning frameworks that make judgment governable, not frameworks that eliminate it.
- Financial institutions embedding structured credit-decisioning models have seen 20–40% lower credit losses, 5–15% revenue uplift, and 20–40% efficiency gains, according to McKinsey benchmarks.³
- Only 13% of community-based financial institutions reported high readiness to adapt credit models to changing conditions in a 2024 Cornerstone survey.⁴ That gap is a leading indicator of future stress, not a problem to defer.
The Real Diagnostic: Three Gaps That Drive Most Portfolio Problems
When an institution is seeing rating inconsistency, exception clustering, or examiner criticism of underwriting standards, the root cause almost always comes down to one of three things: weak decision governance, degraded data infrastructure, or an exception management framework that counts deviations without ever analyzing them.
These gaps tend to travel together. Weak governance lets officers apply policy selectively. Poor data means override rationale cannot be verified after the fact. An exception system that tallies up deviations but never breaks them down by officer, branch, or product creates a false sense of control. When all three gaps exist at once, portfolio drift moves faster than any annual review will catch.
The Hidden Cost of Discretion: How Decision Noise Erodes Portfolio Quality
Discretion is not the problem. Unguarded discretion is.
There is a persistent belief in lending culture that experienced officers exercising judgment produce better outcomes than structured rules. That belief is not wrong in principle. The Bank for International Settlements (BIS) identifies “sound, well-defined credit-granting criteria” as a foundational pillar of sound credit management, not the elimination of human judgment.⁵ The OCC similarly acknowledges that exception loans are often acceptable risks and should not be criticized just because they are exceptions.⁶
What the research shows, though, is that discretion without structure produces systematic drift, not random noise. In a study covering 242,011 retail loan applications across more than 1,000 branches, discretionary scoring manipulation was not confined to a handful of outliers. It was widespread. A one-standard-deviation increase in those manipulative scoring trials raised default rates by 0.3 to 0.4 percentage points relative to baseline, a 12–16% relative increase, with a profitability hit of 1.5 percentage points of ROE.¹ The individual officers were not all making reckless calls. The system just had no way to see or govern the aggregate pattern.
Gartner’s banking risk research has found that decision variability strongly correlates with portfolio instability.⁷ The mechanism is straightforward: inconsistent originations contaminate every downstream signal. When credit grades reflect the originator more than the borrower, the migration picture stops being reliable. When exception rates cluster in certain branches without triggering any alert, reserve models get built on inputs that no longer reflect actual risk.
Exception Clustering Is a Systems Problem, Not a People Problem
The OCC puts it plainly: “when aggregated, even well-mitigated underwriting exceptions can significantly increase portfolio risk,” and financial institutions should track exception trends by department, loan officer, and over time, comparing the performance of exception loans against loans made within policy.⁶
Most institutions have exception tracking. Far fewer have exception analysis. Tracking tells you how many exceptions occurred. Analysis tells you which officers are generating them at rates well above the branch average, which product lines cluster during high-volume periods, and whether exceptions in specific geographies share common repayment risk factors.
The OCC’s 2023 consent order against United Fidelity Bank shows what happens when that analysis is missing. The order cited unsafe or unsound practices not just in underwriting and credit administration, but also in capital planning, stress testing, concentration risk management, ACL methodology, data management, and internal controls.⁸ That is not a coincidence. It is what happens when origination-level inconsistency propagates upward through every function that depends on rating reliability. The 2020 OCC action against Gateway Bank told a similar story: weak board and management supervision, inadequate internal audit, and failure to adhere to prudent loan administration.⁹ By the time those institutions were defending their practices in an exam, the origination-level problem had long since become an enterprise-level one.
The leading indicators are available well before losses show up: exception rates by officer, branch, and product; overrides clustering near score cutoffs; documentation exception trends; watch-list inflow velocity. FDIC research found that current underwriting-risk assessments improved prediction of asset-quality and CAMELS deterioration over the following calendar year.¹⁰ Acting on those signals requires infrastructure that surfaces them.
Rating Inconsistency, Reserve Surprises, and What Examiners Actually See
The Federal Reserve’s reliability threshold for internal rating systems is both specific and consequential. When examiner or internal loan-review downgrades reach 10% of loans reviewed or 5% of dollars reviewed, supervisors may treat the rating system as unreliable and direct the institution to reassess ACL and capital adequacy.² Under CECL, where reserve models are built on historical loss experience and current origination quality, grading drift in recent vintages flows directly into forecast error. At that point, a rating problem has become a capital planning problem built on a false baseline.
BIS is clear on this: internal ratings should be consistent, independently confirmed, and integrated into credit-risk analysis and capital adequacy assessment.⁵ When they are not, the portfolio on paper stops matching the portfolio that was actually built. Portfolio decay rarely starts with charge-offs. It starts with noisier grades, hidden concentrations, and delayed problem-loan recognition. By the time an institution is defending rating integrity in an exam, the governance failure happened long before.
What Structured Decisioning Actually Looks Like
High-performing financial institutions are not replacing credit officers with automation. They are building a governance layer that enforces policy consistently at every origination point, surfaces exception patterns in near real time, and routes non-standard cases to experienced officers with better information, not fewer guardrails.
McKinsey’s documented case study of a leading European financial institution is instructive. After redesigning its commercial lending workflow with automation and analytics, time-to-yes fell from 24–48 hours to four minutes for standard applications and origination costs dropped 30–40%. Roughly 40% of applications moved end-to-end without manual intervention. More complex cases were routed to credit officers with richer analytical support, improving portfolio-level oversight rather than cutting experienced judgment out of the loop.³
That architecture does something additional manual review layers simply cannot: it catches the exception at the moment it occurs, records the justification in a standardized format, and aggregates patterns by officer and branch in real time. It is worth noting that the OCC warns a lack of exceptions may signal a loan policy that is too general to set clear underwriting limits.⁶ The goal is not zero exceptions. It is controlled, visible, and analyzable exceptions. Structured decisioning delivers that.
Five Actions That Make a Measurable Difference
- Quantify your decision variance before your examiners do. Pull exception rates by officer, branch, product, and quarter for the past 24 months. Compare default and delinquency performance of exception loans against within-policy originations in the same vintage. If your team cannot produce this report within 24 hours, the data infrastructure gap is already a control weakness. Make this a standing monthly report.
- Test your rating reliability against the Federal Reserve threshold. Pull internal loan-review downgrade data from the past 12 months and calculate your rate against the Fed’s 10%/5% benchmark.² If you are approaching that threshold, address grading calibration now, before the examiner does it during an exam.
- Move exception management from counting to analysis. Require written justification for every policy deviation in a standardized, aggregable format. Review exception patterns at the board level by officer, branch, and product, not just total volume. This directly addresses BIS Principle 14 on timely exception reporting⁵ and FDIC expectations for board-level visibility into aggregate exception trends.¹⁰
- Integrate origination, grading, and performance data into a single view. Siloed platforms are the structural reason exception patterns go undetected for so long. When credit grades, override records, and performance outcomes live in separate systems, you cannot reliably own your portfolio risk picture. Closing this gap is the prerequisite for reliable CECL inputs, trustworthy concentration reporting, and exam-ready documentation.
- Enforce policy at origination, not after the fact. Policy enforcement built into the origination workflow, applied consistently across every officer and branch, eliminates the variance that downstream review can only partially catch. Route standard applications through structured decisioning. Route complex cases to experienced officers with better data. That is how the benchmarked institutions cut credit losses 20–40% while improving efficiency: not by reducing human judgment, but by making it governable.³
Make Credit Decisions Governable
Portfolio performance does not drift by accident — it drifts when decisioning lacks structure, visibility, and accountability. The institutions that lead are not eliminating judgment; they are making it consistent, measurable, and aligned to policy at every origination point. That is what builds examiner confidence, strengthens reserve reliability, and creates portfolios that perform the way they are expected to. Baker Hill helps financial institutions bring discipline to credit decisioning — so outcomes are driven by design, not left to chance.
Because in credit, consistency isn’t control — it’s performance.
Sources
- Conditional on a dataset of 242,011 retail loan applications across 1,000+ branches; officer incentive and scoring-manipulation study cited in Baker Hill research brief, Why Inconsistent Lending Decisions Create Long-Term Portfolio Decay (2025).
- Federal Reserve, SR Letter guidance on internal loan-review reliability thresholds; referenced in Baker Hill research brief (2025).
- McKinsey & Company, The Value in Digitally Transforming Credit Risk Management (2023).
- Cornerstone Advisors, What’s Going On in Banking (2024).
- Bank for International Settlements (BIS), Principles for the Management of Credit Risk, BCBS Publication d595. Principles 4, 10, and 14.
- Office of the Comptroller of the Currency (OCC), Comptroller’s Handbook: Credit Risk.
- Gartner, Banking Risk Management Insights (2023).
- OCC Consent Order, United Fidelity Bank (2023).
- OCC Formal Agreement, Gateway Bank (2020).
- FDIC, supervisory research on underwriting-risk assessments as predictors of asset-quality and CAMELS deterioration; referenced in Baker Hill research brief (2025).
Baker Hill is the leading provider of lending technology for banks and credit unions across the United States. Each month, financial institutions use Baker Hill’s platform to process over $7 billion in loan originations. With trusted fintech innovation, AI-enabled automation, and deep banking expertise, we help institutions Lend Better, Lend Faster, and Lend More™.
We’re more than a platform. We’re your partner in building stronger communities through smarter banking.
Ready to build a faster, smarter SMB lending process?
Explore Baker Hill’s Digital Small Business solution.
