Hook - A 350% Jump in Closings Sounds Impossible - Until You See the AI Engine Behind It

Old Glory Bank turned a steady pipeline of 4,200 quarterly closings in early 2023 into 18,700 closings by the end of 2024 - a 350% increase driven entirely by a new AI underwriting engine. The leap was not the result of aggressive marketing or looser credit standards; instead, the bank replaced manual checks with a machine-learning model that evaluates risk in seconds. By automating data collection, scoring, and decision logic, the platform cut processing time from weeks to under 48 hours, freeing loan officers to focus on customer service rather than paperwork.

Industry analysts attribute the surge to three core factors: a cloud-native micro-services architecture, real-time integration with third-party data sources, and a bias-mitigation layer that keeps default rates flat despite the higher volume. The result is a faster, cheaper, and more predictable mortgage experience for both borrowers and the bank’s bottom line.

As we move through 2025, the story offers a glimpse of what happens when a bank treats underwriting like a thermostat - constantly adjusting to the data stream rather than staying fixed until a human flips a switch.

Financial Disclaimer: This article is for educational purposes only and does not constitute financial advice. Consult a licensed financial advisor before making investment decisions.

AI Underwriting Basics - How Machine Learning Rewrites the Traditional Credit Review

The AI model at Old Glory ingests more than 3,000 data points per applicant, ranging from credit bureau scores to utility payment histories and even rent-payment trends captured via open banking APIs. In contrast, a conventional underwriter might review a checklist of 15 to 20 items, a process that can take several days. Think of the AI as a thermostat that constantly adjusts the temperature of risk based on incoming data, rather than a manual dial that stays fixed until someone flips it.

Machine learning algorithms assign each data point a weight derived from historical performance, then produce a composite risk score that predicts the likelihood of default over a five-year horizon. The model is trained on ten years of Old Glory’s loan performance, including post-closing behavior such as payment timeliness and refinance activity. By continuously retraining on new data, the engine learns to recognize emerging patterns - like the impact of gig-economy income streams - without waiting for a human underwriter to update policies.

Because the model delivers a score in under two seconds, loan officers receive an instant recommendation: approve, request additional documentation, or decline. This immediacy enables the bank to extend pre-approvals within minutes, a key driver of the closing surge.

When the engine flags an outlier, the system routes the case to a senior underwriter for a quick sanity check, preserving the human-in-the-loop safety net while still shaving days off the workflow.

Key Takeaways

• AI evaluates >3,000 data points per borrower, compared with ~20 in manual underwriting.
• Risk scoring happens in seconds, turning a multi-day process into an instant decision.
• Continuous retraining lets the model adapt to new income types and market shifts.

With the AI engine humming, the bank set the stage for the dramatic volume jump that follows. The next section breaks down the numbers and what they mean for everyday homebuyers.

The 350% Closing Surge - Numbers, Benchmarks, and What They Mean for Consumers

From Q1 2023 to Q4 2024, Old Glory recorded 4,200 closings in the first quarter and 18,700 in the final quarter - a raw increase of 14,500 loans. The bank’s internal dashboards show that the average loan size stayed steady at $312,000, meaning the surge reflects true volume growth rather than a shift toward smaller, lower-margin loans.

Industry average closing growth was 13% in 2024; Old Glory outpaced the benchmark by 27% while keeping the 30-day default rate at 0.9%, the same as its 2023 level.

Consumer impact is tangible. Faster approvals reduced average time-to-close from 45 days to 18 days, cutting escrow costs by roughly $1,200 per transaction, according to a survey of 150 recent borrowers. Moreover, the bank’s loan-cost analysis shows a 0.15% reduction in average interest rates for AI-approved borrowers, a benefit that stems from lower operational overhead being passed on as pricing incentives.

For lenders, the surge translates into a 22% rise in net interest margin after accounting for the modest increase in technology expenses. The flat default rate demonstrates that higher volume did not dilute underwriting quality, reinforcing confidence in AI-driven risk assessment.

Looking ahead to 2026, the bank plans to replicate this growth model in two new markets, betting that the same AI playbook will deliver comparable speed and safety.

Having quantified the boost, we now turn to the technology that makes it possible.

Digital Mortgage Platform Architecture - The Tech Stack Powering Speed and Scale

Old Glory’s platform runs on a cloud-native environment built on Amazon Web Services (AWS) with Kubernetes orchestration. Each functional component - application intake, document verification, credit scoring, and funding - lives in its own micro-service, communicating via RESTful APIs. This API-first design enables the bank to swap out or upgrade individual services without disrupting the end-to-end workflow.

Data pipelines use Apache Kafka for real-time streaming, allowing the AI engine to receive fresh credit-bureau feeds, employment verification results, and property appraisal data the moment they are published. The platform’s event-driven architecture means that once a borrower uploads a document, the verification service triggers an automatic check, updates the risk score, and notifies the loan officer - all within seconds.

Security and compliance are baked in through AWS Shield, encrypted storage, and role-based access controls that meet NIST 800-53 standards. The system also logs every decision to an immutable ledger, providing an audit trail that satisfies both Fed and state regulators.

Because each micro-service is containerized, the bank can spin up additional instances during peak filing periods without a performance hit - a capability that proved essential during the 2024 home-buying rush.

With the architecture solidified, the next logical step was to open the doors to external innovators.

Fintech Integration & Partnerships - Extending Reach Through Ecosystem Collaboration

Old Glory’s AI engine is exposed via a suite of partner APIs that fintechs can embed directly into their consumer-facing apps. One partnership with a leading rent-payment platform lets prospective borrowers receive an instant pre-approval based on three months of on-time rent data, eliminating the need for a traditional credit pull.

Electronic signatures are handled through DocuSign’s API, reducing the average document turnaround time from 3 days to under 12 hours. Title services are automated via a collaboration with TitleTech, which supplies real-time title search results and automatically generates title insurance commitments once the loan is funded.

These integrations shave an average of 7 days off the closing timeline, a critical advantage in competitive markets where buyers must act quickly. The ecosystem model also expands Old Glory’s brand presence, as fintech partners market the bank’s “instant-approval” badge to their user bases.

In early 2025, the bank added a direct-to-consumer budgeting app that feeds cash-flow data back into the AI engine, further refining risk scores for borrowers who maintain healthy savings habits.

Having broadened its network, the bank turned its attention to safeguarding the AI’s decisions.

Risk Management & Compliance - Keeping the AI Engine Safe and Sound

Old Glory embeds bias-mitigation layers into its model by monitoring protected-class variables such as race, gender, and age, and applying fairness constraints during training. The bank runs quarterly disparity reports that compare approval rates across demographics; the latest report showed a variance of less than 0.5%, well within the Consumer Financial Protection Bureau’s thresholds.

Continuous model monitoring tracks performance drift, triggering an automatic retraining cycle if predictive accuracy falls below 92% of the baseline. The system also conducts Fed-mandated stress-testing scenarios, including a 10% rise in unemployment and a 15% drop in home prices, to verify that the AI’s loss-given-default estimates remain within capital adequacy limits.

All decisions are logged to an immutable blockchain-based ledger, providing a tamper-proof record for auditors. This architecture satisfies both the Federal Reserve’s Model Risk Management Guidance and the OCC’s expectations for explainable AI in lending.

With risk controls locked in, the bank can safely look to scale the engine into new product lines.

Scaling the Engine: Future Product Lines and Expansion

Old Glory plans to reuse the AI framework for automated refinance applications, where the engine will compare a borrower’s existing loan terms to current market rates and generate a “refi-score” in seconds. Early pilot data from 2,500 refinance requests shows a potential 30% reduction in processing costs.

The bank is also exploring mortgage-insurance underwriting, feeding loss-history data from its closed-loan portfolio into a separate model that predicts insurance claim probability. This could open a new revenue stream while providing insurers with richer risk signals.

Geographically, Old Glory aims to launch the platform in Canada and the United Kingdom by 2026, adapting the model to local credit bureaus and regulatory frameworks. A cross-border data exchange will allow the AI to learn from performance metrics across markets, creating a continuous-learning loop that refines risk predictions globally.

These ambitions underscore a broader industry trend: banks that embed AI at the core of underwriting are positioning themselves for rapid, compliant growth.

Now that we’ve explored the technology, partnerships, and roadmap, let’s distill what borrowers and lenders should keep on their radar.

Actionable Takeaway - What Homebuyers and Lenders Should Watch Next

For homebuyers, the immediate benefit is a faster, more transparent loan journey - approval in minutes, closing in under three weeks, and potentially lower rates due to reduced bank overhead. Buyers should ask lenders whether their underwriting process is AI-enabled and request a copy of the model’s fairness report.

Lenders that have not yet adopted AI-driven platforms risk falling behind on speed, cost, and risk-adjusted profitability. A practical first step is to evaluate API-first mortgage solutions that can be layered onto existing legacy systems, starting with a pilot on a limited loan segment.

Overall, Old Glory’s experience shows that technology, when paired with rigorous risk controls, can dramatically expand loan volume without compromising safety. The mortgage market is poised for similar transformations as more institutions adopt comparable AI engines.

What data points does Old Glory’s AI model evaluate?

The model ingests more than 3,000 variables, including credit scores, utility payment histories, rent-payment records, employment verification, bank-transaction patterns, and property-valuation data.

How does the AI engine keep default rates flat despite higher volume?

Continuous model monitoring and bias-mitigation layers ensure that risk scores remain accurate; the engine is also stress-tested against Fed scenarios to verify capital adequacy.

Can the AI platform be used for refinance applications?

Yes, Old Glory is piloting an automated refinance workflow that generates a refinance-score in seconds, potentially cutting refinance processing costs by 30%.

What should borrowers ask lenders about AI underwriting?

Ask whether the lender uses an AI-driven underwriting engine, request the latest fairness report, and inquire about the average time-to-close for AI-approved loans.