We opened the insurance panel at the RiskSpan Summit earlier this month with an interesting statistic: according to Moody’s, almost a third of the $6 trillion in cash and invested assets held by US life insurers is now allocated across private credit sub-asset classes. Nancy Mueller Handal of Bayview confirmed it tracks with consideration to private placements, commercial mortgages, and infrastructure debt alongside direct lending. Bill Moretti of Equitable added a useful caveat: the industry still doesn’t have a settled definition of “private credit,” and ABF has only recently been pulled under that umbrella. However you count it, the scale is real and it’s happened fast.
I was joined by Nancy, Bill, and Larry Yang of Global Atlantic / KKR — three practitioners operating through fundamentally different models. What made the conversation worth having was exactly that diversity of structure. Here’s what stuck with me.
You may not know you’re in a bad vintage while you’re living through it.
This was the sharpest exchange of the panel. The group flagged the current non-QM environment specifically: massive appetite, compressed spreads, and layered risk that has never been tested in a declining home price environment. Non-QM has existed entirely within a rising housing market. There was even speculation that we may be experiencing a vintage risk event in data centers right now – and the industry won’t know for years. The most uncomfortable version of vintage risk isn’t the one you can see. It’s the one you’re inside.
The data problem is harder than the analytics problem.
One panelist observed that in structured finance, “you could get to a point where you don’t know what you own.” Larry described building proprietary end-to-end infrastructure — including custom waterfall models — because off-the-shelf systems aren’t granular enough. Nancy’s observation was the most memorable: her research team has grown as capabilities have improved, not shrunk. More tools enable more questions. The data was always there. The limiting factor has always been the ability to extract, manage, and act on it in real time.
AI is a co-pilot, not a substitute.
The panel’s most grounded take: “Seasoned people know what questions to ask — and know when the answer is wrong.” Junior analysts prompting a model without that judgment aren’t getting the same output. The firms making real progress are the ones embedding AI into durable processes, not just re-prompting the same task each month.
And on why they’re still leaning in: Nancy called it “the most exciting market out there.” Larry cited the breadth: in ABF, you’re always encountering asset classes you’ve never seen before. The undisputed line of the day: “Insurance companies are now sexy.”
Hard to argue with that.
—
The RiskSpan Summit 2026 brought together practitioners across insurance, asset management, and structured finance.
AI is already disrupting parts of the U.S. labor market. The more contested question for residential mortgage investors is how and where. The dooms-day camp sees broad structural displacement where white-collar knowledge workers are replaced by AI agents. The optimists counter that AI makes workers more productive, that falling labor costs will expand demand leading to net employment gains. Both camps agree on one thing: the impact will not be uniform. Disruption will concentrate in pockets — specific occupations, metro centers, and income bands.
That specificity is precisely what makes this a mortgage credit problem. Local labor shocks have always transmitted to local housing markets through a familiar mechanism: income impairment reduces demand which pressures home prices, and price declines erode the collateral cushion that protects investors. What varies is the speed, the depth, and whether prices recover once the shock stabilizes.
Early indicators are consistent with the thesis. A Wall Street Journal analysis found the unemployment rate for recent college graduates ages 20–29 at 7.1% as of October 2024 — nearly double the 4.1% rate for the general population — with computer science graduates at 6.1% unemployment and computer engineering majors at 7.5%. The Federal Reserve of NY also published data showing labor market conditions worsened for recent college graduates at the end of 2025, with the unemployment rate climbing to about 5.7% in Q4 2025 — and underemployment rising to 42.5%, its highest level since 2020.
The occupations most exposed to AI displacement cluster in precisely the metros where high-balance prime mortgages are most concentrated, changing the traditional metrics that investors rely on for early warning. AI-driven labor disruption is not a macro thesis to be managed with national averages. While mortgage borrower occupation data is not complete or granular enough to drive AI job loss assumptions, property location data is. The risk needs to be identified, stress-tested, and monitored at a granular level — before it shows up in delinquency data.
The good news is that the data, models and historical reference points exist. With the growth of Non-QM lending, RTL and loan products that carry credit risk without an insurance wrapper, treating AI-driven labor disruption as a present-tense portfolio risk is not optional. To move from thesis to numbers, we ran a portfolio-level stress test on the universe of securitized Non-QM mortgages – starting with how AI exposure is measured at the local level.
Key Findings for investor consideration:
- AI labor disruption is expected to have a higher disproportional impact on white-collar professional jobs that are concentrated in certain “knowledge-work” centers.
- The largest job markets and the deepest credit pools are systematically the most AI-exposed, suggesting AI risk and portfolio risk overlap rather than diversify.
- Stress-testing the $196B universe of securitized Non-QM mortgage loans shows the magnitude is meaningful: under an AI-driven stress scenario calibrated to GFC-level shocks but re-allocated to AI-exposed geographies, RiskSpan’s credit model projects universe-wide cumulative defaults to nearly double and losses more than quadruple.
- Securitized deals benefit from built-in geographic diversification, which dampens the absolute impact across deals. Whole-loan investors, who construct their own geographic mix, face concentration risk that accumulates silently if exposure isn’t an explicit input into selection and surveillance.
- Act now — put the infrastructure in place to use granular data to monitor unemployment, HPI and loan performance and apply scenarios where relevant.
Measuring AI Exposure at the Local Level
Leveraging research published by the Brookings Institution, RiskSpan used their geographic AI exposure scores as the foundation for a credit stress scenario. The resulting county- and metro-level scores follow a similarly clear geographic pattern to that published by Brookings. Highly educated, high-paying knowledge-work centers — Santa Clara (42.8% exposure), King County WA, New York County, DC, Boston — show the highest exposure. Smaller industrial and rural counties score considerably lower. As a sanity check on the methodology, we aggregated and plotted state-level AI exposure against the share of each state’s adult population with a bachelor’s degree or higher (a well-understood proxy for knowledge-work concentration). The two track closely, which is the result one would expect — it confirms the Brookings measure is capturing what it intends to and isn’t producing arbitrary geographic rankings.

Unlike earlier automation, generative AI concentrates in exactly the metros that house the largest mortgage markets: as the chart below shows, the most AI-exposed MSAs are also among the deepest non-agency RMBS markets.

One important caveat before moving on to how we used this data. As Brookings is careful to note, “‘exposure’ does not speak only to the displacement of workers; it also may involve their ‘augmentation’ through rapidly improving AI tools.” The same coder whose role is highly exposed could be made dramatically more productive rather than displaced; the same financial analyst could spend less time on rote modeling and more on judgment-intensive work. Whether exposure plays out as net displacement, net augmentation, or some mix will vary by occupation, employer, and how quickly the technology evolves. For purposes of this exercise, we focus on the displacement story: the scenario where AI exposure translates into labor market shocks that transmit to housing markets and mortgage credit. That’s not the only possible outcome, but it’s the one investors need to be prepared for.
Translating that exposure into a portfolio-level credit stress test on the Non-QM universe, the largest segment of the non-agency RMBS market, required two steps.
Step 1: Construct an AI-driven stress scenario calibrated to the GFC.
Rather than forecast AI displacement magnitude — a genuinely uncertain exercise — we sized the stress to a known reference point: the total HPI decline and unemployment increase observed during the GFC. The GFC offers a useful calibration episode because the empirical relationship between local unemployment shocks and HPI declines is strong and well-documented, regardless of which direction the causation primarily ran. The chart below shows that relationship across MSAs — a clear negative slope linking the spike in unemployment rates to the depth of the HPI decline.

The critical observation for our purposes is what happened in 2008: the metros most exposed to AI today (DC, Boston, San Francisco, Seattle) cluster toward the middle of that scatter plot. They experienced more moderate impacts than the hardest-hit metros, because the speculative and underwriting excesses that drove the GFC were concentrated elsewhere — Las Vegas, Phoenix, Miami, the Inland Empire. An AI-driven shock would invert that geography.
We took the GFC’s national-level shock magnitudes and redistributed them, concentrating the stress in MSAs with high AI exposure. Starting from Brookings’ occupation-level AI exposure scores rolled up to MSA and state-level, we scaled both the unemployment and HPI shocks to each MSA in proportion to its AI exposure share. The result is a per-MSA stress profile that adds up to GFC-magnitude pain at the national level but lands very differently on the map.
The chart below translates that scaling into implied HPI shock magnitudes across the metros where non-agency RMBS exposure actually sits. The metros at the bottom end of both axes (high portfolio balance and severe implied HPI shock) are where the AI thesis would hit hardest if it materializes.

Step 2: Run the loan universe through two scenarios.
We took the $196B universe of Non-QM mortgage loans (~500K loans across 810 active Non-QM deals as of April 2026) and ran these loans through RiskSpan’s proprietary credit model under two scenarios: a baseline and the combined AI-driven Unemployment + HPI stress scenario described above. Every loan received a projected cumulative default rate and cumulative loss rate under each scenario, isolating the AI-driven impact at the loan level.
Findings
Under the AI-driven combined stress scenario, universe-wide cumulative defaults nearly double (3.6% → 7.4%) and cumulative losses more than quadruple (0.5% → 2.2%). The steeper loss response reflects the familiar non-linearity between defaults and severities under HPI stress. The mechanism behind those aggregate numbers becomes clearer when you look at it geographically.
Finding 1 — AI exposure drives MSA-level credit performance.
This finding is by construction — we built the stress scenario by re-allocating shocks to AI-exposed MSAs, so the model output should track AI exposure. The MSA-level chart confirms it does. MSAs at the high end of AI exposure (DC, San Francisco, Boston, Seattle) see over an 8% percentage point (pp) increase in cumulative defaults compared to MSAs at the low end. Riverside, near the lower end of the AI exposure range at just under 30%, sees only a 2.6% pp increase in cumulative defaults. There’s MSA-specific noise from collateral mix, but the geographic risk concentration the AI thesis predicts shows up cleanly in the model output. This is the core mechanism: re-allocate the stress to AI-exposed geographies, and the credit impact tracks the re-allocation.

Finding 2 — The aggregate impact is shaped primarily by leverage.
Decomposing the universe results by current LTV bucket explains where the headline numbers come from. The first thing worth noting is what the LTV distribution looks like: less than 5% of total non-QM balance sits above 80% LTV, and roughly half is below 60%. That’s a meaningfully larger equity cushion than the borrower base of the mid-2000s, when the proliferation of low- and no-down-payment products left investors with thin protection against any HPI decline. Today’s Non-QM universe starts the AI stress scenario with a lot more skin in the game.
Within that distribution, stress impact scales monotonically with leverage, exactly as one would expect: low-LTV loans (≤40%) see a 1.3pp increase in defaults (from 0.96% to 2.24%), while high-LTV loans (>80%) see a 6.7pp increase (from 5.97% to 12.62%), though the relative impact on cumulative defaults is broadly stable across LTV buckets, with defaults about doubling in each bucket. The same pattern holds for losses, where the highest LTV buckets see losses rise by over 3pp under stress versus 0.16pp for the lowest LTV bucket. FICO declines slightly as LTV increases (from 756 for the lowest LTV bucket to 745 for the highest), so the projected cumulative default differences across LTV buckets are driven, at least partly, by underlying borrower credit as well.

Finding 3 — Deal-level concentration matters, but only in relative terms.
When we rolled up by deal rather than by MSA, the picture got more interesting. The absolute (pp) impact across deals shows essentially no relationship with the deal’s AI exposure share — high-AI-exposure deals don’t show systematically larger pp increases. But the relative impact (stress / base − 1) is clearly correlated with AI exposure.
One plausible explanation is that deal-level base default rates vary widely based on traditional credit characteristics — LTV mix, FICO mix, doc type (Full Doc vs. DSCR vs. Bank Statement), occupancy. That variation in base rates may be dominating the AI signal in absolute terms while leaving it visible on a relative basis. Either way, the relative measure is the cleaner lens for isolating AI exposure’s marginal contribution to deal-level risk.
There’s also a more reassuring story embedded in these charts, at least for securitized mortgage assets. Compare the deal-level scatter to the MSA-level scatter and you’ll notice the deal-level points cluster in a meaningfully narrower range — both on the AI exposure axis (X) and on the impact axis (Y). That tighter clustering is geographic diversification at work: even the most AI-concentrated deal in the universe pulls in loans from enough different MSAs that its blended AI exposure sits closer to the universe average than the most exposed individual MSA does. Diversification within deals is doing what it’s supposed to do — dampening the variance of outcomes — even though the unit of analysis is smaller than an MSA. The magnitude of the differential across deals is bounded by how concentrated any individual issuer is willing to get.


The implication for whole-loan portfolios is sharper. Whole-loan investors don’t get the benefit of pre-built deal diversification — they’re constructing their own geographic mix one loan or pool at a time, and concentration risk accumulates silently if it isn’t an explicit input into selection and surveillance. AI exposure is one such risk, and a relatively new one, but it sits alongside more familiar concentrations — natural disaster exposure, regional employment dependencies, single-industry metros — that warrant the same lens. The toolkit is the same in each case: granular geographic data, stress scenarios calibrated to defensible reference points, and loan-level credit modeling. Whether the goal is initial portfolio construction or ongoing surveillance of a seasoned book, the ability to identify these concentrations early — before they show up in delinquency data — is what separates active risk management from passive observation.
What This Means for Investors
History offers instructive analogies including the energy-patch metros of the mid-2010s when oil prices collapsed — Houston, Oklahoma City, and Calgary experienced home price corrections in the range of 5 to 15 percent. Regional manufacturing declines showed a similar story over a longer time period — the long contraction of auto-sector employment in Detroit, for example, played out in local housing markets over years while national HPA remained positive. In neither case did the national trend warn you that specific MSAs were in distress.
Acting now means two things in practice. The first is tracking — building the early-warning infrastructure before you need it. That means monitoring MSA-level unemployment, not just the national rate; watching labor force participation shifts; and tracking occupational mix changes in the markets where your collateral is heaviest. These indicators move before delinquency does. The second is scenario preparation — running stylized AI displacement shocks against the portfolio today using MSA-level or county-level HPA overrides and occupational exposure overlays. The goal is not a forecast but rather a map of your exposure — which corners of the portfolio are most sensitive to particular labor market stress, and what the loss distribution looks like if the more adverse scenarios begin to materialize. That map is worth building now, while the cost of being wrong about the timing is low.
RiskSpan is releasing a new Non-QM-specific credit model in the coming weeks, estimated directly on Non-QM performance data and designed around the underwriting features (DSCR, bank statement, expanded LTV) that distinguish these segments. Re-running this scenario and other stress scenarios on the new model is on the near-term roadmap and may sharpen some of the findings reported here. For investors who want to evaluate their own exposure, the analytical infrastructure described here can be applied to specific portfolios on request.
Note: The Brookings Institution has been tracking AI’s labor market impact since 2019. A 2024 Brookings piece analyzed data from OpenAI to measure occupational AI exposure as the share of tasks where AI could reduce human completion time by at least 50%. AS one would expect, the cognitive/manual divide is evident in the rankings, with computer and mathematical work atop the list at 75% while construction sits at the bottom at 5.6%.
Their 2025 piece extends that work to geography by multiplying each occupation’s exposure rating by its share of local employment in each county or metro, then aggregating. Counties with large workforces in highly exposed occupations (software developers, financial analysts, lawyers, marketing professionals) score higher than those whose employment skews toward construction, agriculture, or in-person services.
Key References:
Generative AI, the American worker, and the future of work | Brookings
There Is Now Clearer Evidence AI Is Wrecking Young Americans’ Job Prospects – WSJ
www.newyorkfed.org/research/college-labor-market
At SFVegas 2026, RiskSpan had a front-row seat to one of the most consequential conversations happening in our industry right now. I moderated a session on Agentic AI and the Securitization Lifecycle, while our CEO, Bernadette Kogler, participated in a panel on AI applications in structured finance. Across both rooms, the message was the same: we are done experimenting. We are redesigning workflows.
The distinction that framed everything was simple but important. Copilots assist. Agents orchestrate. We are moving from AI that enhances what humans do to AI that executes complex, multi-step workflows — with humans supervising the process rather than driving every step. That shift changes everything about how we think about structured finance operations.
And structured finance, it turns out, is almost perfectly suited for this moment. It is document-heavy, multi-party, process-driven, and data-intensive. From origination through surveillance, the lifecycle is fundamentally workflow-based. Panelists across both sessions shared real examples already in production: legal document review compressed from days to hours, AI-powered loan tape scrubbing before cash flow calculations, prompt-driven scenario generation replacing manual model configuration, and surveillance scaled across hundreds of deals per month. One striking observation: the future interface for structured finance may not be a UI at all — it may be entirely driven by prompts.
But the panels were equally clear-eyed about what has to come first. Sixty-five percent of financial services firms are actively using AI, yet only 13% have deployed it across production processes. The gap between those two numbers is largely a data problem. The most sophisticated AI cannot overcome poor inputs, inconsistent loan tapes, or legacy system constraints. Firms that want to lead need to fix the foundation before layering on orchestration.
Explainability matters just as much. With the EU AI Act and US fair lending enforcement raising the stakes, auditable, transparent models are not optional. And governance is shifting from “human in the loop” to “human over the loop” — a subtle but meaningful difference that requires defined accountability, model drift monitoring, and clear operational guardrails.
The one-year outlook from both panels was notably concrete. Expect AI agents managing defined surveillance workflows, deeper cross-platform integration, and a sharper divide between early movers and everyone else. The competitive advantage will go to firms that clean their data, build explainability into their models from day one, and embed AI into operations — not just into pilot programs.
Structured finance has always rewarded process discipline and deep domain expertise. That doesn’t change in an agentic world — if anything, it becomes more critical. The quality of an AI agent’s output is only as good as the prompts and parameters guiding it, and designing those well requires people who understand cash flow waterfalls, covenant structures, and credit risk at a fundamental level. Agentic AI doesn’t replace that expertise. It amplifies it. The firms that understand that will define the next chapter.
At RiskSpan, we have spent years building at exactly this intersection — combining deep structured finance domain knowledge with purpose-built analytics infrastructure. That foundation is what makes it possible to deploy AI that actually works in production, not just in demos. The opportunity in front of our clients right now is significant, and we are focused on helping them capture it.
Private asset-backed securities don’t follow the same rules as public securitizations. The structures are more diverse. The triggers are more nuanced. The collateral is more diverse. And yet, most market participants still try to manage these instruments with tools designed for a different world entirely—or worse, with disconnected spreadsheets that multiply risk with every manual handoff.
RiskSpan’s Private ABF platform was built specifically to solve this problem. Not as an adaptation of existing tools, but as a purpose-built data architecture designed from the ground up for the unique demands of private structured finance. The difference isn’t incremental—it’s foundational.
Built for the Full Lifecycle
RiskSpan’s Private ABF platform isn’t just a database—it’s a relational architecture where every entity maintains its identity and relationships across the entire deal lifecycle. The platform connects deals to their tranches, tranches to their collateral, collateral to individual loans, and all of these to the triggers, waterfalls, fees, and reserves that govern cash flow distribution.
This matters because private ABS transactions aren’t static. Collateral performs. Triggers trip. Waterfalls redirect. Reserves build and release. A platform that can’t maintain these relationships in real time isn’t managing deals—it’s creating snapshots that are stale before they’re finished.
Loan-Level Depth That Powers Real Analysis
At the heart of any ABS transaction is the underlying collateral. RiskSpan’s Private ABF platform maintains comprehensive loan-level data with over 200 attributes per asset, tracking everything from origination characteristics through current performance status. This includes credit metrics like FICO scores and debt-to-income ratios, property and collateral details, payment history and delinquency tracking, modification and loss mitigation status, and ARM reset schedules and rate mechanics.
The platform currently manages nearly half a billion loan records across active transactions. Each loan maintains its full history—not just current state, but the complete trajectory that informs forward-looking projections. When you run a cash flow model, you’re not working from aggregated pool statistics. You’re working from the actual loans, with their actual characteristics, generating projections that reflect real collateral behavior.
Depth for Esoteric Structures
RiskSpan’s Private ABF platform currently supports over several collateral types and structures. Each asset class has its own performance characteristics, prepayment behaviors, and loss dynamics. The platform’s architecture accommodates this diversity without forcing artificial standardization.
But diversity in collateral is only part of the challenge. Private ABS triggers represent some of the most complex conditional logic in structured finance. Unlike standardized agency delinquency tests, private deal triggers can involve multi-step calculations with lookback periods, cure provisions that allow temporary breaches, step-rate adjustments that phase in over time, and early amortization events with distinct severity levels.
The platform models triggers as executable logic, not static documentation. When a trigger breaches, the system knows what happens next—which waterfall priorities shift, which reserve requirements change, which reporting obligations activate. This is the deal’s immune system, and it needs to function in real time.
Time-Series Without the Chaos
Every entity in RiskSpan’s Private ABF platform exists in time. The platform maintains separate performance histories for tranches, collateral pools, fees, and reserves—each keyed by reporting date to enable point-in-time reconstruction of any deal state.
This architecture solves one of the most persistent problems in private ABF operations: answering the question “what did we know, and when did we know it?” Whether for regulatory compliance, investor reporting, or internal risk management, the ability to reconstruct historical deal states isn’t a luxury—it’s a requirement that spreadsheet-based approaches simply cannot meet reliably.
The Waterfall as Working Code
Cash flow waterfalls in private ABF can run to dozens of steps with conditional branches, pro-rata splits, and priority reversions. RiskSpan’s Private ABF platform models these waterfalls as executable payment sequences—not flowcharts, but actual logic that routes cash from sources to destinations based on current deal state.
Each waterfall step defines its priority in the payment sequence, its source of funds and destination, its allocation basis (pro-rata, sequential, or targeted), and the conditions under which it activates or suspends.
When combined with the platform’s cash flow engine, these waterfall definitions become working models. You can project payments under any scenario, stress collateral performance, and see exactly how cash moves through the structure period by period.
An Architecture Built for AI
The same architectural principles that make RiskSpan’s Private ABF platform effective for traditional analytics make it exceptionally well-suited for artificial intelligence and machine learning applications. This isn’t a coincidence—it’s a consequence of building a data model that prioritizes structure, relationships, and semantic clarity.
AI systems thrive on clean, well-organized data with explicit relationships. RiskSpan’s Private ABF platform delivers exactly this: normalized entities with consistent identifiers, clear hierarchies from deals down to individual loans, and temporal versioning that distinguishes current state from historical snapshots. When an AI model needs to understand a transaction, it doesn’t have to infer structure from unstructured sources—the relationships are already defined and traversable.
The platform’s semantic richness enables natural language interfaces that actually work. Because every field has meaning within a consistent schema, AI can translate questions like “show me deals where the senior OC trigger is within 50 basis points of breach” into precise queries without ambiguity. The loan-level depth means AI models can identify patterns across hundreds of millions of records—finding correlations between origination characteristics and performance outcomes that would be invisible to traditional analysis.
Time-series architecture is particularly critical for AI applications. Machine learning models for credit risk, prepayment prediction, and loss forecasting require historical sequences, not point-in-time snapshots. RiskSpan’s Private ABF platform maintains this temporal context natively, enabling training datasets that capture how loans and deals evolve over their lifecycles.
RiskSpan is already deploying AI capabilities on this foundation: automated anomaly detection that flags unusual performance patterns, intelligent document extraction that populates deal records from offering documents, natural language querying that makes complex analytics accessible to non-technical users, and predictive models that leverage the full depth of loan-level history. The platform doesn’t just store data—it organizes knowledge in a form that AI can reason about.
Designed for the Enterprise
RiskSpan’s Private ABF platform operates as a multi-tenant platform with granular access controls. Every record carries client and user identifiers that enable sophisticated permission model—issuers see their deals, investors see their positions, servicers see their portfolios, and rating agencies see what they’re authorized to review.
This isn’t just about security (though it is that). It’s about enabling collaboration across the structured finance ecosystem without compromising confidentiality. A single platform can serve all participants in a transaction, each with their appropriate view of the data.
Built for What Comes Next
Private ABS is an evolving market. New asset classes emerge. New structures get tested. New regulatory requirements arrive. RiskSpan’s Private ABF platform accommodates this evolution through flexible schema design that allows custom attributes without requiring database modifications.
When a client brings a novel structure—say, a securitization of an asset class we haven’t seen before—the platform can ingest and model it without waiting for a software release. This extensibility is what allows a platform to stay current with market innovation rather than constantly playing catch-up.
What This Means in Practice
The architectural decisions in RiskSpan’s Private ABF platform translate directly to operational capabilities. For deal structuring, you can model waterfall variations and see their impact on tranche economics before going to market. For pricing, you can run scenarios against actual loan-level collateral, not simplified pool assumptions. For risk management, you can monitor trigger proximity and project breach timing under stress. For surveillance, you can track every metric that matters, with full audit trails and historical reconstruction.
Private ABF deserves purpose-built infrastructure. RiskSpan’s Private ABF platform delivers it.
As generative AI becomes a powerful tool in Private asset-backed finance (ABF), the need for precision and transparency is more critical than ever. At RiskSpan, we’re applying Large Language Models (LLMs) to automate and accelerate private ABF deal modeling and surveillance. But speed is only half the battle—accuracy is non-negotiable.
That’s where Human-in-the-Loop (HITL) validation plays a vital role. While the RiskSpan platform incorporates sophisticated AI guardrails, we believe the right blend of automation and expert oversight ensures results that are not just fast—but reliable, auditable, and production-ready.
The AI-Powered Workflow: What’s Automated
Our private ABF modeling and surveillance system uses LLMs to handle several critical tasks:
- Data Extraction: Parsing offering memos, indentures, and loan tapes to extract structural and financial data.
- Deal Code Generation: Producing executable waterfall models based on extracted rules.
- Database Ingestion: Uploading validated deal terms and triggers into the RiskSpan system of record.
- Surveillance Automation: Running periodic deal performance analyses and compliance checks.
But What About Hallucinations?
Generative models are powerful but imperfect. Without the right controls, they can fabricate securitization tranches or fees that are not present in the deal documents. They can also misinterpret waterfall rules or omit critical override conditions or generate semantically incorrect code for cashflow models. To address these challenges, RiskSpan employs a multi-layered safeguard framework, combining asset class based extraction; LLM-as-Judge; Rule-Based Guardrails and Inline Human Review
Humans in the Loop: Three Layers of Oversight
We’ve embedded human validation at three key points in the deal lifecycle:
- Pre-Modeling Validation: before LLM-generated outputs are finalized, RiskSpan analysts review extracted terms and model prompts—correcting anything misaligned with the source documents.
- Inline Oversight: during waterfall code generation, humans validate AI-generated logic in context, ensuring correct treatment of subordination, triggers, caps/floors and other.
- Post-Deployment Monitoring Surveillance: outputs are reviewed both by the RiskSpan team and client-side structuring or credit teams. Feedback is looped back into model tuning and prompt optimization.
Looking Ahead: RAG and Continuous Improvement
We’re actively exploring Retrieval-Augmented Generation (RAG) to reduce hallucinations even further. By grounding AI responses in deal-specific material such as offering documents, trustee reports, and internal risk memos—we aim to: 1) eliminate off-topic responses. 2) increase trust in model-derived outputs and 3) enable deeper customization per issuer or asset class.
The Takeaway
AI is transforming how private ABF deals are modeled and monitored—but it must be grounded and guided by human expertise and built for institutional rigor. At RiskSpan, we’re not just accelerating workflows—we’re raising the bar for accuracy and trust in AI-assisted private structures. Human-in-the-loop is not a fallback—it’s a strategic pillar. Want to see how our AI platform works in action? Reach out to schedule a demo or contact your RiskSpan representative to learn more.
Limited Partners (LPs) continue to demand better data, faster, and with full transparency. At this week’s Private Credit Tech Summit in New York, I moderated a panel of industry leaders for a discussion on where LP expectations are heading and the challenges managers face trying to meet them.
My fellow panelists included:
· Charlie Tafoya, Co-Founder and CEO of Chronograph
· Marios Tsiptis, Senior Portfolio Manager
· Arnab Mazumdar, Partner at Pantheon
We explored the evolving expectations of LPs, the operational hurdles General Partners (GPs) face, and the technology shaping the next frontier of data transparency in private credit.
What follows is where we landed.
LP Expectations Are Outpacing the Status Quo
Quarterly performance reports and aggregate numbers used to be enough. But Marios Tsiptis explained that today’s LPs (particularly insurance companies) want detailed, timely insight into exactly what they own and what risks they’re carrying.
From Pantheon’s fund-of-funds vantage point, Arnab Mazumdar laid out three foundational data pillars:
1. Performance metrics
2. Operating and credit-level data
3. Consistency across managers
Lagging information and inconsistencies across investment structures, especially between SMAs and feeder funds, create significant friction. Structure matters, and data must be complete, accurate, and delivered on LP terms.
The Operational Reality: Bridging Ambition and Execution
Charlie Tafoya, who works closely with GPs via Chronograph, provided a sobering view of the day-to-day realities. Many managers are eager to modernize, but they’re grappling with:
· Delays in data delivery
· Data quality and validation issues
· Fragmented internal processes
There’s just no getting around the fact that investment teams simply must be embedded in the data flow. Too often, front-office insights are siloed from the operations teams responsible for reporting. “The investment team is the source of truth,” Charlie noted, making their engagement essential to any successful data transformation.
Meanwhile, Arnab called for a convergence of internal monitoring and external reporting. Aligning what GPs see internally with what they share externally could yield benefits across the board—but cultural and technological hurdles remain.
Technology as Enabler, Not Panacea
Tools that support data interoperability, real-time reporting, and workflow automation are rapidly maturing. But challenges persist around integration with legacy systems and data standardization.
From the LP side, Marios painted a picture of an ideal future. Intuitive dashboards, seamless access across portfolios, and fully integrated delivery pipelines all featured prominently in this future. But the road leading there is still under construction.
According to Arnab, the next few years will hinge on industry standardization and early adopters gaining a competitive edge, while laggards risk being left behind.
Transparency builds trust—but it requires real operational change.
More than just a tech problem, this requires a cultural shift. GPs and LPs need to work as true partners in designing data ecosystems that are not only robust and scalable, but also reflect the growing sophistication of the private credit space.
As the private markets evolve, so too must the infrastructure that supports them. Those who succeed in this transition will not only meet the expectations of today’s LPs, but also shape tomorrow’s.
Register here for next month’s call: Friday, June 20th, 2025 (pushed back one day on account of Juneteenth).
Each month, we host a Models & Markets call to offer our insights into recent model performance, emerging credit risks, and broader economic indicators. This month, we spotlighted model backtesting updates, macroeconomic conditions, and market analytics that are shaping investment strategies across loans, securities, and private credit.
Click here to listen to the entire 23-minute recording, or continue reading for a summary.
Prepayment Model Performance and Enhancements
RiskSpan’s prepayment model continues to exhibit strong alignment with observed data across all coupon cohorts. During the call, we showcased updated backtesting results for 2022 FN/FH cohorts across multiple coupon bands (1.5s–3.5s, 4.0s–5.5s, and 6.5s), revealing that projected vs. actual CPRs remain closely correlated, even in volatile rate environments.
Additionally, RiskSpan has introduced a Non-QM-specific prepayment model to address behavioral differences in this segment. This is particularly timely, given elevated delinquency trends discussed later in the session.
Our recently enhanced Credit Model 7, leveraging a delinquency transition matrix, is expected to be released by the end of May and will provide a more granular view of credit migration patterns.
Spread at Origination: A Key Risk Signal
Spread at Origination (SatO), the difference between the borrower’s rate and the prevailing PMMS rate at application, is emerging as a critical predictor of refinance activity. Higher SatO values suppress prepayments even in pools with favorable coupons.
Using MBS loan-level data, we illustrated how SatO dynamics impact investor vs. owner-occupied loans, with notable geographic variation. States like CA, FL, and NY show materially different average rates for investor loans, independent of LLPA effects.
As a forward-looking initiative, we are developing a generalized spread model that isolates residual pricing differences not explained by known borrower or loan characteristics. This could further enhance predictive power by benchmarking loans against peer cohorts defined by origination date, FICO, occupancy, and geography.
Macroeconomic Outlook: Sticky Rates and Stable Housing
The economic backdrop remains mixed:
- Mortgage rates hover around 6.95%, with no near-term relief in sight.
- The Fed Funds Rate is projected to stay elevated, with the first potential cut not expected until September 2025. Even then, consensus suggests only a modest decline to 3.75–4.00% by year-end.
- Home prices are largely stable, as reflected in the Case-Shiller Index. Year-over-year appreciation remains positive but muted.
- Unemployment stands at 4.0%, and inflation is moderating but still above target.
This persistent high-rate environment will continue to dampen refinance activity and challenge affordability, reinforcing the importance of modeling spread-driven behavior accurately.
Non-QM Delinquencies Spike
The bad news: Delinquencies are surging within the Non-QM sector, particularly for 2022–2023 vintages:
- DSCR/investor loans are showing delinquency rates an order of magnitude higher than conventional loans.
- This reinforces the need for robust credit modeling, especially in the private credit space where standard agency risk buffers don’t apply.
The good news: RiskSpan’s new NonQM credit and prepay models are now live to support more accurate surveillance of these exposures.
Contact us to learn more or to request a free demo of our platform and models.
As securitization models become increasingly complex and differentiated, validation becomes a critical challenge. We’ve experimented with an innovative approach that leverages large language models (LLMs) as impartial judges to validate models implemented across different platforms.
The Dual-Implementation Challenge
In cash flow modeling, we often maintain parallel implementations—typically in Python for flexibility and Excel for transparency. How do we ensure both versions produce consistent results?
Enter the “LLM as Judge” approach!
A Real-World Case Study: Residential Transition Loan Funding
Consider a portfolio of residential transition loans with a funding structure including:
- 100 loans averaging $275,000 each
- 12-month average terms at 8.75%
- A 75% advance rate
- 2% loss reserve build-up
- Performance triggers based on delinquency rates
We implemented this structure in both Python and Excel, then submitted both models to an LLM for validation.
The LLM Validation Process
The LLM first analyzed the conceptual alignment between models, confirming both followed the same fundamental approach to cash flow projection, default assumptions, reserve mechanics, and triggers.
Next came a rigorous numerical comparison. The LLM detected a $100,000 investor distribution discrepancy in Month 2:
- Python model: $1,790,702
- Excel model: $1,690,702
Through logical analysis, the LLM determined this likely stemmed from differently evaluated trigger conditions. This kind of subtle implementation difference could easily go unnoticed in manual validation, potentially leading to significant valuation discrepancies over time.
Beyond Discrepancy Detection
The true power of this approach extends beyond finding differences. The LLM also provided:
- Stress testing recommendations tailored to our specific product, including scenarios for rapid defaults, extension waves, and interest rate shocks
- Model risk management insights highlighting documentation needs and suggesting a formal reconciliation process
- Code quality assessment noting strengths and weaknesses in both implementations
Why This Matters
For securitization professionals, this approach offers several advantages:
- Efficiency: Automation of tedious line-by-line comparisons
- Comprehensiveness: Identification of conceptual differences, not just numerical ones
- Regulatory compliance: Better documentation for model risk management requirements
- Objectivity: Unbiased third-party perspective
Private credit is having a moment, as they say, now approaching $7 trillion in global assets, and is poised to double in size over the next decade. As traditional banks tighten lending due to regulatory constraints, private credit is stepping in to provide flexible, high-yield investment opportunities for institutional investors. However, this expanding market brings challenges, including illiquidity, bespoke deal structures, and complex risk assessments.
Chartis Research, in collaboration with RiskSpan, explores these evolving dynamics in a recent report, shedding light on the forces shaping private credit’s expansion and the critical role of technology in mitigating risk.
As private credit markets grow, effective risk management is crucial for investors seeking stable returns. Advanced technologies like AI and machine learning are revolutionizing private credit risk assessment, enhancing cash flow modeling, pricing accuracy, and portfolio diversification. RiskSpan leads the industry with innovative solutions, leveraging loan-level data and cloud-based platforms to provide real-time analytics. Whether you’re an asset manager, institutional investor, or lender, understanding the latest private credit trends is essential for success.
Read the full article to explore how private credit is transforming finance and why technology-driven risk management is the key to sustainable growth.
Contact us to learn more about how RiskSpan’s platform can support your private credit analytics.
We are delighted to announce the release of RiskSpan’s series of Private Credit Primers aimed at providing investors with essential knowledge about the diverse and growing landscape of the loan types that private credit investors are buying. These primers offer at-a-glance insights into the mechanics, performance expectations, and unique features of various asset classes, enabling investors to make informed decisions in this dynamic market.
The first three primers in the series are available now: They focus on Residential Transition Loans (RTLs), Personal Loans, and HELOCs — loan types that are becoming increasingly popular in the private credit space.
Residential Transition Loans (RTLs) (full primer here)
RTLs are short-term loans designed to help borrowers bridge financial gaps during transitional periods in residential real estate. Often used for construction, bridge, and relocation purposes, RTLs typically have terms ranging from 6 to 36 months and feature higher interest rates than traditional long-term financing. These loans play a crucial role for both homeowners and real estate investors, especially in markets where property values fluctuate or where short-term liquidity is needed.
- Important Features: RTLs often involve draw functionality, allowing borrowers to access funds incrementally as projects progress. Another key aspect is the use of the “As-Repaired” Value (ARV) to calculate Loan-to-Value (LTV) ratios, based on the projected value of the property after repairs.
- Performance Considerations: While RTLs have performed well during periods of stable or rising home prices, the primer cautions that these loans are more vulnerable during economic downturns or periods of home price decline.
Personal Loans (full primer here)
Personal loans can be secured or unsecured and are used for a variety of purposes, including debt consolidation, medical expenses, and large purchases. They are repaid in fixed monthly installments over a predetermined period.
- Important Features: The primer highlights key modeling considerations for personal loans, such as static default/prepayment assumptions, which rely on historical data to predict future loan performance based on factors like loan age and borrower profiles.
- Performance Expectations: As of 2024, the delinquency rate for personal loans stands at around 3.38%, with average interest rates hovering around 12.42%, though they can vary widely depending on market conditions and borrower credit quality.
HELOCs (full primer here)
The complexity of modeling HELOCs stems from their sharing characteristics of both a mortgage and a credit card. Reliable assumptions about borrower behavior both during the draw period (when balances can move in either direction at any time) and during the post-draw, repayment-only period are crucial to forecasting correct cash flows.
- Important Features: Understanding regional variations and borrower characteristics can provide deeper insights into HELOC performance, helping to refine risk models and lending strategies.
- Performance Expectations: The delinquency rate for HELOCs can vary based on factors such as economic conditions, borrower credit quality, and market trends. However, historically, the delinquency rate for HELOCs tends to be lower than for unsecured loans or credit cards.
What to Expect from the Series
Each primer in the series will not only break down the mechanics of the loan type but also provide performance insights and modeling considerations. With the ongoing volatility in the financial markets, these primers will explore how various asset classes perform under different economic conditions, such as rising interest rates, declining home prices, or increasing unemployment.
By offering practical, data-driven insights, the Private Credit Primer series will serve as an invaluable resource for private credit investors who are looking to deepen their understanding of these asset classes and navigate potential risks effectively.
Stay tuned for more primers in this series, as we continues to expand RiskSpan’s library of resources for private credit investors!