AI-driven underwriting and claims automation change the math for banks by making fraud a priceable, transferable risk instead of a volatile operating expense.  

AI models score identity and transaction risk up front, reduce false declines and lower incident rates before any loss occurs. When fraud does happen, straight-through digital claims move validated events into a 30-day payment cycle, so insurers and reinsurers carry the volatility while the bank replaces a lumpy expense with a predictable premium. 

Fraud today is a recurring operating cost. Banks absorb direct losses and the indirect expenses that surround them, including investigation time, chargebacks, customer churn and higher servicing efforts. Industry studies have shown that every $1 of fraud can cost more than $4 when all effects are counted. Accounting rules such as Current Expected Credit Losses also pull expected losses forward into allowances, which depresses earnings and ties up capital. 

Insurance for fraud

A new approach is to transfer fraud losses to insurance and treat them like an insured portfolio. In this model, the bank pays a premium that is priced to its exposure. Claims are adjudicated quickly, and the insurer and reinsurers carry the loss volatility. As insured portfolios mature and data improves, insurers can syndicate or securitize parts of the risk to expand capacity. The effect for a large institution is to replace an uncertain stream of losses and operational drag with a predictable premium and faster recovery of cash after an incident. 

Sunil Madhu, founder of Instnt, described the shift plainly: “Instnt’s the first solution in the market to allow businesses to shift these losses that they would otherwise be holding on their own balance sheets.”  

He noted that large institutions “hold hundreds of millions and billions of dollars in Tier One capital — think of it as self-insurance capital — for a rainy day.” His aim is to “pull all of that into a large insurance pool, collateralize the risk, and eventually securitize that risk.” On insurer strength, he said, “The insurance is on A plus paper, we have Munich Re and Swiss Re’s balance sheets behind us.”  

On speed and certainty, he added, “we pioneered insurance payments in 30 days or less, guaranteed. … Issues get resolved at submission; there are no claims denials.”  

The payoff to bank treasury is clear, he said: “Treasury can convert those reserves into working capital for the bank, which the bank can then leverage for growth.” 

The numbers

Play the economics forward.  

Start with a representative large bank that suffers $500 million in annual gross fraud losses. If you apply an all-in cost multiplier of 4.41, which captures operational and revenue side effects, the baseline cost is about $2.205 billion a year.  

Now replace that with an insured construct. Assume premiums priced at 40 to 60 percent of expected gross losses. Assume modern controls and underwriting reduce gross incidents by 10% to 30% before any claim is filed. Assume claims are paid in about 30 days rather than funds being tied up for a half year or more.  

Finally, assume the bank had been informally self-insuring by holding economic reserves equal to roughly the annual fraud loss, and that moving to insurance allows 50% to 70% of that buffer to be redeployed to earning assets.  

If the value of freed capital is marked at 10% and the cost of carry on delayed recoveries at 5%, the base case produces roughly $2 billion in modeled savings against the old baseline and an estimated premium ROI near 8-to-1. Even a conservative case with higher premiums and smaller loss reductions remains strongly positive. 

AI and savings

Where do the savings come from? 

• AI screening reduces incidents and customer friction while increasing approval rates and revenue.  

• Volatility and tail risk move from the bank to the insurer, which converts a lumpy and reputation-sensitive expense into a budgetable premium. 

• Faster claims remove months of working capital drag.  

• Capital that had been functionally idle as self insurance can be put to work in loans and fee businesses while the insurer stands ready for the remaining incidents. 

What needs to be true?

• Coverage definitions and triggers must be precise and auditable.  

• Claims operations need to be digital and straight through.  

• Insurer and reinsurer capacity must be committed at meaningful limits and aggregate covers.  

• Controllers and examiners must be comfortable that expected losses have been genuinely transferred and that allowances and capital reflect the new risk position.  

Over time, as portfolios standardize, insurers can place diversified identity fraud layers into the insurance-linked securities market to add scalable third-party capacity. That pathway is already proven for natural catastrophe risk and can be adapted to well defined, data rich financial crime exposures. 

The result for large institutions is simpler planning, lower earnings volatility, more productive capital and a better customer experience.  

Madhu’s framing captures it: Use AI to prevent and underwrite, shift residual losses off balance sheet, resolve incidents in about a month and let treasury turn former rainy day buffers into growth.  

Jim McCarthy serves as chairman for McCarthy Hatch, which provides data-driven insights for risk management. A founding member of the Consumer Financial Protection Bureau, he is a keynote speaker and fractional CRO/CCO in the financial services industry with more than three decades of experience.