AI in business is moving from assistant mode to action mode. DBS and Visa are now piloting a model where AI agents can not only suggest products but also complete purchases under bank-controlled rules.
That shift matters because this is no longer about productivity inside a chat window. It is about real operational transactions.
What changed
In the DBS pilot, AI agents can handle routine purchases using tokenised payment credentials and issuer-controlled approval logic. In simple terms: the AI can initiate, but the bank still controls whether money moves.
This gives companies a practical architecture for AI-driven commerce:
- Agent layer: finds options and initiates actions
- Policy layer: applies customer limits, categories, and consent rules
- Bank control layer: verifies identity, risk, and authorisation before settlement
Why this is useful for business teams
For operations and finance leaders, the value is clear when use cases are repetitive and bounded:
- subscription renewals
- routine procurement
- travel rebooking within policy
- low-risk repeat purchases
The real adoption challenge
The key blocker is not model quality. It is trust, governance, and dispute handling.
Before scaling, teams should define:
If these controls are weak, automation risk rises faster than productivity gains.
A practical rollout pattern
A low-friction path is:
- Start with one narrow, low-risk workflow
- Keep strict issuer and policy controls in place
- Measure approval rates, exception rates, and user trust
- Expand only after you can explain every failed and successful transaction
The DBS pilot is a strong signal that AI-native payments are entering real business environments. Teams that prepare governance now will be in a better position to scale safely later.
Ready to see where your team can apply AI safely and practically? Start here: https://dakik.co.uk/survey
