Category: Technology & AI

AXA is building a global AI hub and says an initial version is used across five entities. EIOPA reports that 65% of insurers in its survey already use generative AI and another 23% intend to. Those figures tell us how widely the technology has spread. They do not show whether any particular process now works better.

For a leasing company, I would count fewer agents and measure more work: time per application, errors in document checks, the quality of the credit memo and the time until a useful question reaches the customer.

Every pilot needs a baseline and a comparison. A draft produced in two minutes offers little benefit if correcting it takes twenty. Detecting missing documents early and reducing repeated requests would be a better outcome for customers and colleagues alike.

Scaling begins when speed and quality can be measured together.

Which measure would show genuine progress in your organisation?

Sources: https://www.axa.com/press/press-releases/axa-and-publicis-sapient-collaborate-to-deploy-global-ai-hub-at-scale
https://www.eiopa.europa.eu/publications/scaling-ai-finance-systemic-risks-resilience-and-european-sovereignty_en

Leave a Comment

The Monetary Authority of Singapore and industry partners have published a paper on safeguards for agents in finance. Its SAFR framework proposes checks at the point of action: what is the agent authorised to do, what must be tested before execution and what record must be kept? It is a proposal from Singapore, not a Swiss rule.

In leasing, the distinction is practical. A document agent may read files and draft a request for missing evidence. It has no reason to access payments. A KYC agent may highlight contradictions. It should not independently approve a customer.

Each agent therefore needs a defined mandate: purpose, data sources, permitted actions, limits, an accountable owner and a way to stop it. Critical actions should be checked against that mandate at the time they are attempted.

Technical ability does not confer business authority.

Which action in your process would you explicitly refuse to delegate to an agent?

Source: https://www.mas.gov.sg/news/media-releases/2026/mas-partners-industry-to-develop-safeguards-for-ai-agents-in-finance

Leave a Comment

ESMA has made digital innovation a European supervisory priority from 2027 and explicitly includes AI. European supervisory authorities have also highlighted dependencies on external technology providers and cyber risks. These are European signals; they do not by themselves create a new rule for a Swiss leasing company.

The operational questions are already immediate. Which documents reach the model? Who can see them? Where are prompts and outputs stored, and for how long? What happens when a provider or an interface fails?

A firm that waits until after a pilot to answer may have to rebuild a working process. I would map the data flow, permissions and fallback arrangements before using real customer files.

A policy alone cannot enforce a control. The control must work in the system and leave evidence that it did.

Could you draw the route a customer document takes through your AI workflow today?

Sources: https://www.esma.europa.eu/press-news/esma-news/esma-sets-new-supervisory-priority-digital-innovation-2027
https://www.eba.europa.eu/publications-and-media/press-releases/esas-call-vigilance-over-external-dependencies-cyber-threats-and-private-credit-risks

Leave a Comment

Citi describes Arc as infrastructure for agents across its businesses with common controls. AXA is developing a global AI hub, with an initial version in use across five entities. The examples raise an organisational question: who manages access, permissions, logs and quality once separate experiments become daily operations?

A leasing company does not need to copy a global bank. Yet ten isolated assistants, each with its own path to customer data, soon become difficult to govern.

I would establish common foundations first: approved data sources, roles, approval points, test cases and a record of what each agent has used. Specific tools for applications, KYC and contract checks can then share that foundation.

The platform has no value by itself. Its value appears when a successful pilot can move into production without rebuilding its controls from scratch.

Which control does your organisation currently recreate for every AI pilot?

Sources: https://www.citigroup.com/global/news/perspectives/2026/competitive-differentiator-david-griffiths-building-worlds-most-ai-empowered-financial-institution
https://www.axa.com/press/press-releases/axa-and-publicis-sapient-collaborate-to-deploy-global-ai-hub-at-scale

Leave a Comment

Deutsche Bank has deployed an agentic AI solution for source-of-wealth work in its private banking centres in Singapore and Hong Kong. It researches approved sources, prepares documentation and identifies gaps. People review the assessment.

A Swiss leasing company faces different questions, but much the same problem of fragmented evidence. An application, company register, identity checks and information on beneficial owners must agree. Someone needs to spot conflicting names, missing records and facts that require clarification.

An assistant could prepare that comparison for the operations team. It should not convert an unresolved match into an approval. In a difficult case, the useful result is a clear explanation of what remains uncertain and who must review it.

The aim is a more complete, auditable file, with less time spent moving between systems.

Which recurring question delays onboarding most often in your organisation?

Source: https://wealth.db.com/en/about-us/news/2026/deutsche-bank-private-bank-launches-agentic-ai-kyc-processes.html

Leave a Comment

Mastercard is building Agent Connect to link merchants, platforms and authorised AI agents. The Bundesbank has examined how agents might prepare or initiate payments under a mandate. These are developments in infrastructure, rather than a ready-made payment process for every financial firm.

For a leasing company, the first useful question comes before execution. Does the supplier on the invoice match the approved contract? Has the asset been accepted? Has the beneficiary account changed? Are all conditions for disbursement documented?

An agent could assemble those answers and draw attention to exceptions. I would keep a firm boundary around the payment itself: software prepares; authorised people review and release. Any move beyond that deserves a separate decision with precise limits and a means of stopping the agent.

The more an agent can do, the more clearly its identity, mandate and permissions need to be defined.

Could your organisation show who authorised every automated step in a payment today?

Source: https://www.mastercard.com/global/en/news-and-trends/press/2026/september/mastercard-gives-merchants-a-simpler-way-to-build–connect-and-s.html
https://publikationen.bundesbank.de/content/1007712

Leave a Comment