A mid-sized lender looked at its dashboard and saw what most executives would call “fine.” Losses were on plan, delinquency had ticked up but not alarmingly, and portfolio growth was steady enough to avoid tough conversations.

Then they cut the data by segment.

In one product, a single FICO band in a handful of geographies was suddenly responsible for an outsized share of new losses and early-stage roll rates that had quietly doubled.

That’s the real underwriting challenge in 2026.

The biggest risk isn’t a visible blow-up. It’s slow, uneven erosion that only shows up after it has already eaten into your margin.

The Environment Moved. Your Underwriting Didn’t.

On paper, many portfolios still look acceptable. Growth patterns have shifted, loss profiles look different than they did two years ago, and regulators are focused on a wide range of issues, not just underwriting.

Trade policy shifts, rate uncertainty, and changing regulatory priorities have compressed the window between when conditions change and when that change hits your portfolio. The institutions that feel this most acutely are the ones whose refresh cycles were built for a more predictable environment.

Under the surface, the mix has shifted. Customer behavior, channels, and product strategies have all evolved faster than most institutions’ policy cycles, data pipelines, and analytics routines.

You’re being asked to defend returns and capital deployment in a world where small mispricings matter.

When underwriting stays “good enough” while conditions change around it, you end up taking risk you didn’t intend and leaving safe growth untapped.

Three Symptoms That Show Up Again and Again

1. Portfolio averages look fine, but dispersion is widening.

Your overall loss and delinquency metrics stay within appetite, while certain FICO bands, geos, or products quietly deteriorate and consume more capital than anyone realized. FICO’s 2025 Credit Insights show how performance and delinquency can vary dramatically across FICO brans, even when portfolio metrics look acceptable.

2. Manual overrides are becoming the unofficial policy.

Your underwriters increasingly override scores and rules to “save” deals, but no one is measuring whether those overrides actually perform or just add noise and hidden risk.

3. Policy refresh cycles lag the environment.

Cutoffs, pricing tiers, and exposure limits that worked in 2022–2023 haven’t been recalibrated to current performance cohorts and macro conditions. You’re effectively underwriting to a world that has already moved on.

The executive-level impact is straightforward: diluted risk-adjusted returns, less reliable forecasts, and less confidence that your portfolio is aligned with today’s reality.

Five Practical Ways to Get Underwriting Back on the Front Foot

This isn’t about rebuilding your models from scratch. It’s about using data, performance, and analytics to see what’s really happening and tune underwriting decisions with intent.

1. Build a Single Performance View Across Systems

Most institutions still view underwriting performance through a fragmented lens. LOS, core, collections, and fraud data live in different places, which means no one sees risk and profitability across the full lifecycle.

Consolidating these sources into an enterprise-level view makes it possible to answer simple but critical questions: Which segments are driving outsized losses? Which are reliably outperforming? And how does that map to your current policies and pricing?

Once you can see delinquency, loss, and return by segment in one place, it becomes much easier to pinpoint where underwriting is too loose, too tight, or misaligned with strategy.

What this looks like in practice:

A mid-sized lender brought on PAG to build out a reporting suite after seeing the kind of segment-level deterioration described above. Once we had a consolidated view, we discovered something the client hadn’t seen: a fraud issue that had been quietly impacting delinquency for months and was about to hit chargeoff. After a deep analysis, PAG identified the source of the fraud and shut it down, reducing delinquency on new vintages by 30% and saving 12% on vintage chargeoff curves.

A large credit union client told a similar story. They were using different LOS systems for each of their consumer lending products, some with more sophisticated data sources and capabilities than others. PAG combined the data streams and showed that the systems for secured and unsecured consumer loans were underperforming compared to the credit card LOS systems. PAG rebuilt the underwriting programs for the loan portfolio, resulting in a 17% reduction in Year 1 losses while increasing growth by 8%.

2. Refresh Cutoffs and Pricing with Current Data

Many cutoffs and pricing grids were set in an earlier environment and haven’t kept pace with how the market has evolved. If they haven’t been revisited in 12 to 24 months, they’re guesses about how today’s portfolio will behave.

A data-driven refresh uses rolling performance cohorts and current market dynamics to recalibrate. You tighten where losses are no longer being compensated, and selectively open up in segments that consistently outperform.

The opportunity is to move away from broad, one-size rules toward more precise criteria aligned to observed behavior.

What this looks like in practice:

PAG rebuilt the underwriting program for a large credit union by increasing APR and adjusting lower credit spectrum cutoffs through augmented data enhancements. The result: losses dropped 22% while overall yields declined by just 36 basis points. Growth dipped a moderate 4%, but the improved loss curve allowed for a $12 million release from loan loss reserves.

3. Turn Manual Overrides into a Learning System

Overrides capture information that models may miss, especially in nuanced or relationship-driven situations. They become dangerous when they’re opaque and unmanaged.

The goal isn’t to eliminate overrides. It’s to instrument them.

Systematically capture who overrode what, for which reason, and how those accounts actually performed. Then analyze patterns against model-only decisions.

That allows you to codify override patterns that consistently add value, discourage those that don’t, and identify where additional data could improve the model itself. Instead of overrides being a blind spot, they become an input into better underwriting.

This kind of override analysis naturally feeds into the performance command center described below. When you can see override performance alongside vintage curves and segment-level metrics, you get a complete picture of where human judgment is adding value and where it’s adding risk.

4. Stand Up an Underwriting Performance “Command Center”

You don’t need more raw data. You need a clear, recurring view of the signal.

Right now, critical underwriting insights are often buried in specialist reports or one-off analyses. An underwriting performance “command center” is a governed reporting layer that surfaces a small set of leading indicators, reviewed monthly or quarterly.

Metrics we typically include on an underwriting-health dashboard:

• Approval rate: Are you saying yes at the right frequency for your risk appetite?
• Book-to-look and book-to-approve ratios: How efficiently are you converting applications into funded loans?
• Net interest margin (APR minus losses minus cost of funds): What’s your real return after credit costs?
• Credit loss rate: Where are you relative to plan, and how is the trend moving?
• Vintage loss and delinquency curves: Are newer cohorts performing better or worse than older ones?
• FICO/VantageScore migration and distribution: Is your incoming credit quality shifting?

Reviewed on a regular cadence, these metrics create a shared language between credit, finance, and operations about what “good” looks like now, not last year. That rhythm makes it possible to make smaller, proactive adjustments instead of reacting to surprises.

5. Use Analytics to Test and Scale New Segments Safely

Tightening across the board is the bluntest possible response, and it often hurts long-term growth more than it protects against loss. A better move is to use segmentation and analytics to run controlled tests into new or previously constrained segments.

Design small pilots with clear cutoffs, pricing, and loss expectations, then feed performance data back quickly. If a segment performs within appetite, you scale. If it doesn’t, you adjust or shut it down.

Institutions that adopt this test-and-learn mindset can say “yes” more often where the data supports it, while still protecting the downside. In a competitive landscape, those pockets of safe growth compound.

What this looks like in practice:

A credit union client had been offering near-prime and prime credit cards exclusively to existing members. PAG helped them expand into a new market: non-customer acquisition through affinity-branded cards. Using our data resources (bureau data and premium attributes), PAG built the initial profit model, marketing model, and underwriting program. Year 1 growth targets exceeded projections by 7%, with losses and delinquency coming in 2% below forecast.

Where Predictive Analytics Group Fits

Most organizations don’t have a modeling problem as much as a visibility and execution problem. The models, data, and reports usually exist somewhere. But they’re not stitched together in a way that gives you a timely, trustworthy view of underwriting performance.

Predictive Analytics Group helps you close that gap. Through our GOBLIN enterprise data platform, we consolidate data from multiple legacy systems, build practical reporting that leaders actually use, and turn analytics into specific changes to policies, cutoffs, and strategies.

What this looks like at scale:

We helped a large card client explore the mid-credit market (620–680 FICO scores) by building segmentation models that targeted customers in those ranges while avoiding adverse selection. The machine learning model also improved performance in their existing near-prime and prime portfolios through more effective segment targeting. That freed up room to take on riskier segments below 680.

The result: 35% portfolio growth with only a 5% increase in credit losses. The bank has since grown its card portfolio, internal operations, and vendor network to the point where they’re now closing in on top-15 credit card provider status.

Next Steps

If you see some of these symptoms in your own portfolio, the next step may not be a brand-new model. It may be a sharper view of what your current underwriting is really doing to risk, return, and growth, and a focused effort to align decisions with that reality.

If you’d like to talk through what this looks like for your portfolio, here’s where to start: