Fiuu warns AI shopping raises payment permission questions

Fiuu has warned that AI-led purchasing will raise new questions about how customer permission is established in payments, an issue it says is becoming more relevant as AI adoption rises among businesses in Singapore.

Enterprise AI adoption in Singapore rose from 4.3% in 2023 to 23.5% in 2025, according to figures cited in Fiuu's commentary. The shift is moving discussion beyond search, recommendations and productivity tools toward the role AI could play in completing parts of a purchase journey on a customer's behalf.

Often described as agentic commerce, the model would allow AI tools to place orders, initiate payments or complete bookings within parameters set by a customer. For payment providers and merchants, the central issue is whether a transaction can still be shown to reflect the customer's intent when an automated system makes some of the decisions.

"Customers still need confidence that every transaction is authorised, secure, and reliable. As AI plays a greater role in facilitating purchases, the important thing is ensuring that it acts with the customer's permission and within clearly defined limits," said Eng Sheng Guan, Chief Executive Officer, Fiuu.

The concern goes beyond standard authentication, which verifies who a customer is. Businesses will also need to understand what an AI agent was permitted to do, and where those limits end if prices change, stock becomes unavailable or a booking no longer matches the original instruction.

Permission limits

In a conventional online purchase, the buyer typically moves through each stage directly, selecting a product and then approving payment. Under a more automated model, a customer might instead authorise an AI tool to buy within a set price range, choose from pre-approved alternatives or complete a transaction only if certain timing and cost conditions are met.

That creates a distinction between confirming identity and confirming authority. A payment may be technically approved, but merchants and platforms may still need a way to determine whether the AI acted within the customer's original instructions.

"The first consideration should be customer permission. Customers need to understand what they are allowing an agent to do and have the ability to change or withdraw those permissions for future purchases. Identifying the customer and obtaining payment approval are important, but neither establishes the full scope of what the agent was permitted to do," said Eng.

This becomes particularly important when circumstances change during a transaction. If a product costs more than agreed, a preferred option disappears or terms shift materially, the system may need to stop and return the decision to the customer rather than complete the purchase automatically.

Keeping it simple

The answer is unlikely to be a checkout process with more visible steps for consumers. Instead, businesses may need to build more of the complexity into background systems, such as secure credentials, contextual risk checks and records showing what was authorised and what happened during the transaction.

Automation should not make it harder for customers to understand what they approved or to seek help if something goes wrong. That may be more difficult when merchants, platforms and payment providers all play a role in the same transaction chain.

"When something goes wrong, customers should still have a clear route to assistance. Where several parties are involved, merchants, platforms, and payment providers need to understand their respective responsibilities and have a reliable record of what happened," said Eng.

For businesses considering where automated purchasing may emerge first, Fiuu pointed to repeat purchases and some travel services as likely early use cases. In those areas, customers may be more willing to set rules in advance, such as acceptable price ranges, preferred timing and fallback options, allowing AI tools to act without manual input at every stage.

Even in those cases, businesses will need to define the boundary between acceptable variation and a change that requires a fresh decision from the customer. That boundary could shape how much of the buying process can be delegated to software while still leaving the customer in control.

Regional complexity

The issue may be more complex for Singapore-based businesses operating across Southeast Asia. Automated transactions could have to work across a mix of cards, wallets, bank transfers, QR payment systems, currencies and approval rules that differ from one market to another.

What works in one payment environment may not work in another, raising questions about interoperability as well as authorisation. For companies selling across the region, that means the debate over AI in commerce is also a debate about whether payment systems can support consistent rules and records across multiple methods and jurisdictions.

Fiuu, which says it processed more than USD $13 billion in total payment volume in its 2025 financial year, operates across several Asian markets including Singapore, Malaysia, the Philippines, Thailand, Indonesia, Taiwan, Vietnam and Hong Kong. It supports more than 110 payment methods across online and offline transactions.

"The long-term value of agentic commerce will not come from automation for its own sake. It will depend on whether we can make these new experiences easier to support while keeping customer permission, control, and payment security clear throughout the transaction," said Eng.