Skip to content
Three CubesGet the kit
Leader's reviewRetail and consumer8 markets11 min read

Agentic checkout in 2026: a review for Asia-Pacific retail leaders

AI shopping agents can now find, compare and pay across Asia-Pacific, but each market is arriving by a different route. This review translates the global story into what to do in your markets, and who should own each decision.

Quick verdict

OverallPrepareAct now in Australia and India, and on the marketplaces your customers use
Quick verdict

Agentic checkout is reaching Asia-Pacific by three routes: card networks piloting market by market, national payment rails in India, and marketplace agents in Japan and South-East Asia. Make your catalogue agent-readable where your customers already shop, test in one market first, and keep the till for partners whose returns and liability terms you have agreed.

Get the prompts and market inputs to test this in your own AI

Best for

  • Large retailers in Australia and India with clean product data
  • Brands selling through Rakuten, Shopee or Lazada, where agents already sit inside the platform
  • Repeat, low-value purchases such as groceries

Not for yet

  • Fully autonomous checkout in Japan, where shoppers want the final say
  • Retailers with unreliable range, price or stock data
  • Anyone without a returns policy written for agents

What I like

  • Marketplace agents are paying off in Japan: Rakuten reports order value up about 26%, and Yahoo! Shopping says about 20% of order value now goes through its agent
  • You can test one market and one door at a time

What I don't

  • No market has settled who pays when an agent gets it wrong
  • Comfort with agents buying ranges from 31% of AI-using shoppers in New Zealand to 1% in Japan

How agentic checkout lands in each market

Act nowStart this quarterPrepareDecide within 6 to 12 monthsWatchRevisit next year

AustraliaOpen-web agents and cards

Act now

Google's first Asia-Pacific market for agentic checkout, live since June 2026 for single-item purchases (Ragtrader), while Bunnings and Kmart build their own agents. Trust lags: 64% are uncomfortable letting AI complete a purchase (IT Brief).

IndiaNational payment rails

Act now

UPI agent payments are in pilot for ChatGPT, and Razorpay has tied up with Zomato, Swiggy and Zepto and is running pilots with 15 to 20 merchants (Economic Times). Grocery and quick commerce move first.

SingaporePayments test bed

Prepare

Banks are ready before shoppers: DBS and UOB are live on Mastercard Agent Pay (Mastercard), but only 14% of shoppers use generative AI while shopping (BriefAsia).

JapanMarketplace agents

Prepare

Agents live inside the marketplaces: about 20% of Yahoo! Shopping order value went through its AI agent by July 2026 (Yahoo! News Japan). Yet only 1% would let AI buy without checking first (Visa Japan), so design for agent-assisted, shopper-confirmed checkout.

New ZealandOpen-web agents and cards

Prepare

31% of shoppers who already use AI are comfortable letting an agent complete a purchase (PwC New Zealand), and the four largest banks are testing agent payments with Visa (CFOtech).

MalaysiaCards and marketplaces

Prepare

Malaysia and Singapore were the first South-East Asian markets for Mastercard's authenticated agent transactions, tested with UOB (Mastercard). Shoppers can now find Shopee products inside ChatGPT here too.

IndonesiaMarketplaces and chat

Watch

The agent is the marketplace or the chat thread: Shopee discovery inside ChatGPT, and Meta's sales agent inside WhatsApp (TechWire Asia). Keep marketplace listings clean; your own checkout can wait.

ThailandCards and marketplaces

Watch

The first authenticated agent transaction came in April 2026, an airport ride booked with Krungthai Card (Mastercard). Shoppers mainly use AI to compare prices before buying.

Found your markets?The kit includes these eight market profiles as a sourced file you can attach to Copilot, Claude or ChatGPT.

Get the kit

How it rates for your organisation

Large multi-market retailer

Act now

Agents are already reading your range in Australia and India. Fix catalogue data once, then choose a lead market for trolley partners.

Marketplace-led brand

Act now

On Rakuten, Shopee and Lazada the agent is the platform's. Your listing data is now your shelf position.

Mid-size specialty retailer

Prepare

Make product data agent-ready in your home market and track which platforms your customers use. Building your own agent can wait.

Grocery and everyday needs

Prepare

Groceries are India's expected first use case (Reuters). Elsewhere the first effect is price comparison. Protect margin before you open the trolley.

Non-profit and charity retail

Watch

Low risk of being bypassed, and donor trust matters more than conversion. Revisit next year.

Asia-Pacific is getting agentic checkout by three routes, and none of them is the US route

The short answerIn the US and Europe, agentic checkout is led by open-web assistants paying by card. In Asia-Pacific it arrives three ways: card networks piloting market by market, national payment rails in India, and agents built into marketplaces in Japan and South-East Asia. The route in each market decides which door you need to open first.

Card networks, market by market. Mastercard has completed authenticated agent payments in Australia, New Zealand, Singapore, Malaysia, India and Thailand this year (Mastercard). Visa has more than 50 issuers across ten Asia-Pacific markets testing agent transactions (Visa).

National rails. India’s payments operator is preparing a protocol that lets registered agents make small UPI payments without approving each one (Reuters). It is likely to need Reserve Bank approval first (Business Standard).

Marketplace agents. Rakuten Ichiba’s agent draws on its marketplace, fashion, resale and books services, and users’ average order value is up about 26% (Rakuten). About 20% of Yahoo! Shopping order value went through its agent by July (Yahoo! News Japan). Shopee products can now be found inside ChatGPT in seven Asian markets, with checkout still on Shopee (Sea).

Whichever route reaches your market, the decision is still the same three doors.

The three doors of agentic commerceThree doors in sequence: the catalogue, owned by merchandise and operations; the trolley, owned by commercial and pricing; and the till, owned by the executive team and board.DOOR 1The catalogueCan agents read your range?DOOR 2The trolleyCan outside agents build baskets?DOOR 3 · BOARDThe tillWho owns the customer?
Three doors, three owners. In marketplace-led markets, door 1 is your platform listing, not your own site.

The assessment against five board questions

Value potential

Early results are strong, from Rakuten's order values to Bunnings' conversion, though both are company-reported.

Readiness of the technology

Payment rails are ahead of retail. Banks in six markets have run live agent payments; most retailers have not.

Cost and effort

Most of the work is product-data quality, which you need anyway. Each extra market adds language, payment and platform work.

Risk managed today

Liability, loyalty and disputes are unresolved everywhere. 45% of Asia-Pacific shoppers want stronger payment security before trusting agents (Visa).

Ease of reversing

Easy for the catalogue and trolley. Harder once a marketplace agent owns the relationship in your category.

Four options compared by cost, time and risk

OptionCostTimeRiskMy take
Wait and watchLowNoneRisingAgents and marketplaces will describe your range anyway, possibly badly.
Open the catalogue
Recommended
Low to medium1 to 3 monthsLowMake range, price and stock agent-readable on your site and every marketplace you sell on.
Partner on checkoutMedium3 to 6 months per marketMedium to highOne market, one partner at a time, with returns and data terms agreed first.
Build your own agentHigh6 to 12 monthsMediumKeeps the customer relationship. Worth it at Bunnings' or Rakuten's scale, rarely below it.

What works today and what does not yet

What works

  • 74% of Asia-Pacific shoppers already use AI to discover or track products (Visa).
  • Real-time rails such as UPI, PayNow, DuitNow and PromptPay suit small, frequent agent payments.
  • You can decide market by market and door by door.

What does not yet

  • In South-East Asia, 24% use AI while shopping, and creators are still trusted more (Marketing-Interactive).
  • 21% of Japanese shoppers want to handle final payment themselves, against 14% to 18% in the other markets surveyed (Adyen via PaymentNavi).
  • Two-thirds of retail leaders surveyed don't expect shoppers to fully embrace agent purchasing before 2028 (Deloitte).

The final verdict

Ideal adopter
A retailer with clean product data, a lead market in Australia or India, and a returns policy it is willing to extend to agents.
Main drawback
Nobody has settled who owns the customer, or who pays, once an agent is doing the buying.
Bottom line
Open the catalogue in every market now, choose one lead market for trolley partners, and treat the till as a board decision.

One move for Monday. Ask an outside agent to shop your top 20 products in your two largest markets, and note everything it gets wrong.

The Monday kit

Run Monday's test in your own AI

Everything you need to do Monday's move this week with Copilot, Claude, ChatGPT or Gemini, inside your organisation's own AI.

  • Market inputs packEight market profiles with sourced data points, ready to attach to your AI
  • Five tested promptsAgent shopping test, catalogue audit, door owners, partner terms, market priority
  • Experiment scoper skillA Claude skill and Copilot agent instructions that draft your one-page experiment scope
  • One-page scope templateHypothesis, market, door, metric, owner, 30-day test and stop rule

Free. One email: click confirm and the files download. Nothing from your organisation goes into the kit; you run it inside your own AI. Privacy

Check your inbox

We've sent one email to . Click confirm and the kit downloads.

Two optional choices, so the kit arrives in the right format:

Your organisation's AI
Your markets

How I assessed it

I reviewed announcements, consumer surveys and trade reporting across eight markets published between February and October 2026, and checked each figure against its source on 17 October. Each source carries a label for how far it can be relied on.

How reviews and evidence labels work

Questions leaders ask about agentic checkout in Asia

What is agentic checkout?

Agentic checkout lets an AI agent find a product, build a basket and complete the purchase for a shopper, within limits the shopper sets.

Is agentic checkout live in Asia-Pacific?

Yes, in parts. Google made Australia its first Asia-Pacific market for agentic checkout in June 2026, for single-item purchases in Search, AI Mode and Gemini. Mastercard has completed authenticated agent payments in Australia, New Zealand, Singapore, Malaysia, India and Thailand, and India is piloting agent payments on UPI.

How is agentic commerce in Asia different from the US and Europe?

In the US and Europe it is led by open-web AI assistants paying by card. In Asia-Pacific it is arriving through three routes: card networks piloting market by market, national real-time payment rails such as India's UPI, and agents built into marketplaces and super-apps in Japan and South-East Asia.

Should retailers block AI shopping agents?

Blocking every agent is a short-term choice. Most retailers should make their catalogue readable now, decide case by case which agents can build baskets, and keep checkout for partners whose returns and liability terms they have agreed.

Who is responsible when an AI agent buys the wrong thing?

There is no settled answer yet in any market reviewed. Reporting on India's planned agent payment framework says liability for wrong or unauthorised payments will still need to be addressed through regulation, so agree returns, refunds and dispute handling with each agent partner before enabling checkout.

About the author

Dan McMahon

Dan helps leaders across Asia-Pacific turn global AI shifts into results in their own markets, starting with the value already in their people, process and technology. He founded Three Cubes.