AI Employment Risk Scoring: How DataStars Measures Occupational Displacement

AI Is Rewriting Borrower Risk — and Most Portfolios Can't See It

Every mortgage lender evaluates a borrower's ability to pay. Every insurer prices a policyholder's ability to sustain premiums and likelihood of filing a claim. Both depend on the same assumption: that the borrower's or policyholder's income will persist over the life of the obligation.

That assumption is breaking.

The occupations that have historically anchored the strongest loan files and the most profitable insurance books — financial analysis, legal services, marketing, software development, accounting — are now the occupations most exposed to generative AI displacement. The 800-credit-score borrower with an 8-year tenure at a fintech company looks pristine on every traditional metric. DataStars scores that borrower's occupation at HIGH — with a Distress Duration of 22 months before income disruption forces a default.

No traditional credit model captures this. No actuarial table prices it. DataStars built the scoring layer that does.

The Research Foundation

DataStars' scoring methodology is built on a composite of four peer-reviewed academic indices, each measuring a different dimension of AI occupational exposure:

C-AIOE (Complementarity-Adjusted AI Occupational Exposure) — developed by researchers at NYU and the IMF. Measures how exposed an occupation's core tasks are to current AI capabilities, adjusted for whether AI complements the worker or replaces them.

GENOE (Generative AI Occupational Exposure) — developed by the International Labour Organization. Estimates the probability of significant task automation within five years — the time-horizon index that answers "how soon?"

Webb Index — developed at Stanford. Tracks active AI research and development capital flowing toward automating specific occupation task descriptions. This measures where the money is going.

Frey & Osborne — the foundational Oxford study on automation probability. Predates generative AI but provides a structural anchor that confirms or challenges the newer indices.

These four indices are combined into a proprietary composite using DataStars' weighting methodology, then applied to Canada's NOC 2021 and the U.S. SOC classification systems. The composite covers approximately 500 Canadian occupations and maps to U.S. equivalents for cross-border portfolio analysis.

The result is a score from 0.00 to 1.00 for each occupation, classified into risk tiers: LOW, MODERATE, HIGH, or CRITICAL.

Beyond the Occupation: Borrower-Level Intelligence

An occupation score tells you the job is at risk. It does not tell you whether this borrower at this employer will be displaced on a timeline that matters for the file.

An accountant at a Big Four firm and an accountant at a six-person crypto startup have the same occupation code — but very different risk profiles. A software developer at a hospital system is in a fundamentally different position than a software developer at an AI company that is automating its own engineering function.

DataStars layers real-time employer intelligence on top of the base score to produce a Borrower Volatility Score — a borrower-specific risk assessment that accounts for the employer's financial health, AI adoption posture, recent workforce changes, the borrower's career mobility within their sector, and their estimated financial runway in a displacement scenario.

The headline metric is Distress Duration: the estimated number of months a borrower could sustain mortgage payments — or insurance premium payments — after an AI-driven employment disruption. This is the number that maps directly against court timelines, mortgage maturity dates, policy terms, and settlement windows.

Three Scoring Branches

Not every borrower or policyholder is a salaried employee at a public company. DataStars uses three scoring branches to handle the full range of employment types:

Branch A — Employed. Occupation-based scoring with employer-specific intelligence layered on top. Handles the majority of traditional mortgage borrowers and insurance policyholders.

Branch B — Self-Employed / Business Owner. Occupation scoring doesn't apply to "Owner, 3-Unit Rental Portfolio." This branch assesses business continuity risk: industry AI exposure, operating history, revenue concentration, and client base stability.

Branch C — Gig / Platform Workers. Income depends on the platform's AI trajectory, not the worker's individual task automation. A rideshare driver on a platform deploying autonomous vehicles faces a fundamentally different risk than a freelance designer on a marketplace with no automation roadmap. This branch scores the platform, not the person.

What This Means for Mortgage Portfolios

The risk that AI displacement creates for mortgage lenders is not aggregate price collapse. It is clustered default in borrower segments that traditional models treat as low-risk.

Concentration risk is invisible to standard metrics. A MIC with 200 loans may have 40 borrowers in HIGH or CRITICAL occupations without knowing it. The credit scores are clean. The debt-service ratios work. The properties are in strong markets. But 20% of the book is underwritten against income that has a measurable probability of disappearing within 5 years — and the lender's standard risk reporting does not capture it.

Displacement creates non-performing loans that don't resolve through traditional channels. When a borrower in a cyclical downturn loses a job, they find a new one. The mortgage goes 90 days delinquent and then self-cures. When a borrower in a structural displacement loses a role that no longer exists in their sector, the non-performing loan doesn't self-cure — it escalates to Power of Sale, enforcement, and loss.

The "switch lanes" problem extends distress timelines. A displaced financial analyst cannot transition to nursing or electrical work without years of retraining. Reemployment timelines for structurally displaced white-collar workers are longer than traditional unemployment assumptions — which means the borrower is non-performing for longer, carrying costs accumulate faster, and the lender's recovery position deteriorates further.

What We Disclose and What We Don't

DataStars publishes the academic foundations of the scoring methodology — the four peer-reviewed indices and their institutional sources — the three scoring branches, the risk tier classifications, and the regulatory framing.

The proprietary composite weighting, the employer intelligence pipeline, the modifier calibrations, and the Distress Duration model are not disclosed. These are the product. They are built on a scored employer intelligence dataset that grows with every file processed and cannot be replicated without equivalent file volume over equivalent time.

DataStars tracks 69 market indicators across labour, housing, distress, macro, AI risk, income, and legal categories — updated daily, weekly, and monthly from primary sources including StatsCan, TRREB, CMHC, CanLII, Bank of Canada, and the Ontario Superior Court.