In a speech on 18 September, FSI Chair Fernando Restoy discussed AI’s effects on banks’ customers and business-model resilience alongside banks’ own use of the technology. Published by the BIS, the speech contributes to supervisory debate. It is neither a new binding rule nor a forecast of defaults in particular industries.
Event / source publication date: 18 September 2026The following is Human Fintech’s analysis
Past repayment does not explain future demand.
At a renewal or limit increase, a lender can ask what the borrower’s customers are paying for and whether that proposition is changing. Payment history remains useful evidence, but it reflects an earlier revenue model. A borrower facing pressure on the price of routine work may need a different assessment from one using automation to deliver a more valuable service.
Avoid treating an industry or occupation as an automatic risk verdict. Two businesses with the same classification can differ in contract durability, customer relationships and responsibility for outcomes. The relevant analysis concerns those observable differences. A broad technology label without borrower-specific evidence can lead to poor decisions and obscure companies that have credible plans to adapt their offering.
Trace the transition through the cash schedule.
Consider an illustrative service provider whose customer requests a lower price after introducing automation. The credit discussion should follow the effect on margins, staffing commitments and collections rather than stopping at a revenue haircut. It should also ask how management could respond and how quickly. This is a scenario design exercise, not a prediction that the same shock will affect every provider.
Adaptation can require expenditure before savings emerge. Training, integration and quality control may consume cash during the transition even if the eventual operating model is stronger. A facility structured around immediate cost savings could therefore create avoidable pressure. Separating investment timing from the expected benefits helps the lender assess liquidity needs and an appropriate repayment profile under alternative outcomes.
Use evidence to improve monitoring.
A manageable starting point is a sample review of borrowers with different business models. Examine which indicators are available and meaningful: major contract renewals, repeat orders or changes in realised pricing may be useful in some cases. Information should be obtained through appropriate channels and assessed within the lender’s established decision process, with accountable human review of material conclusions.
The output may be revised underwriting questions, a more relevant contact schedule or additional downside scenarios. It should not become an automatic portfolio restriction based on a single assumed exposure to AI. A sound monitoring process connects observed business changes to credit decisions and gives borrowers an opportunity to explain the evidence and their response before conclusions are drawn.
White Fintech.
Parts II–III of White Fintech cover lending economics, funding and risk management: a foundation for further reading about the lending business.
Complete English edition, PDF + EPUB. Read the first 30 pages free, with no sign-up.