AI shopping agents: 54% of Australian small firms haven't prepared

Australian small businesses are counting on AI shopping agents this Christmas, and most haven't done the groundwork. A Xero survey of 500 Australian small business owners and senior decision-makers, reported on 5 November, found 38% expect AI shopping agents to lift holiday sales, and 22% expect a rise of more than 10%. In the same survey, 54% hadn't taken practical steps to make their business easier for AI shopping tools to find, and only 19% had added structured product information that AI systems can read.

What the survey shows

The AI questions sit inside a wider Black Friday survey, in which 53% of respondents plan to join the sales, up from 43% a year earlier. Xero's ANZ executive Charlie Sheppeard told Inside Retail that "small businesses need to be easy to find this festive season." The 38% figure records what owners expect to happen and says nothing about sales achieved. Coverage of the survey gives no fieldwork dates, margin of error or New Zealand results, so the numbers can't simply be carried across the Tasman. We also haven't seen any published measure of how much revenue AI agents currently bring to NZ small businesses.

The closest local reading comes from Prospa's tracker of 501 NZ SME decision-makers, published last month. AI use reached 59%, up from 50% six months earlier, but only 7% said AI was strategically important across multiple functions. Among non-users, 34% were unsure of the return and 32% didn't know which tools suited them, while 12% cited upfront cost. Prospa's questions cover a firm's own use of AI. Agent readiness is a separate test, because the customer's software does the comparing, and a retailer with no AI tools of its own can still be assessed by one.

Why NZ stores are already involved

Shopify changed the position for many local retailers on 28 September. Its developer changelog says browser agents can now read and update a checkout and submit the order after buyer confirmation, with no merchant configuration needed. The feature uses WebMCP, a standard in a Chrome origin trial since June. Shopify's documentation stops an agent entering new card details or changing line items, and requires the agent to show the buyer the order and total before placing the order. The same documentation describes no merchant opt-out.

Shopify Payments is available in New Zealand, and Forge Creative's Roger McSaveney argued in October that local stores can be in scope without having made any decision. The same article notes that consumer reach here is thin so far, with Google's auto browse announced only for paying subscribers in the US and India, and Meta's Muse launched in the US.

Google's rollout shows a similar split between discovery and buying. On 16 September, Google made AI performance insights in Merchant Center generally available in markets including New Zealand. The report shows how often a retailer's brand and products appear in AI Mode and AI Overviews compared with other brands. Google's new cart transfer and checkout testing tools are rolling out in the US, with Australia and Canada due early next year and no date given for New Zealand.

Stripe's NZ guide, updated on 28 October, says most current tools here narrow the options and build a cart for a person to approve. Shoppers appear to prefer that arrangement. Accenture research cited by Inside Retail NZ found 32% would let an agent choose within set budget and brand limits, while 9% were open to fully autonomous purchasing, although the article doesn't say which countries were covered. PwC's survey of 1,000 NZ consumers who already use AI when shopping describes AI as a new layer in the buying decision.

What it means for NZ businesses

Our read is that NZ retailers should prepare now for being found and compared by AI, and can treat agent-completed purchases as a later project. An agent that can't read a product's size, compatibility or stock level has little reason to shortlist that product. Structured product data also feeds Google listings and site search, so the work has value even if agent traffic stays small for another year.

The harder part is often internal. Product attributes tend to be scattered across an ERP or inventory system, a website and a supplier's spreadsheet, and those records don't always agree. Fixing the data on the website alone means repeating the fix at every product update, so the correction belongs in whichever system holds the master record.

Security settings carry a less obvious risk. TechCrunch reported in October that anti-bot defences are stopping agents on many sites, sometimes by accident. Walmart, which wants agents to reach its store, said its failures weren't intentional, and a human-verification button appears to be the likely cause. In many NZ firms, the person who sets firewall rules doesn't own the online sales target, so a store can block legitimate agents without anyone choosing to.

Some questions remain open. McSaveney's article says Shopify's documentation doesn't confirm whether merchants can see agent-placed orders or refuse them, which affects fraud checks and returns handling. Businesses that build their own shopping assistant have a further obligation under the Privacy Act 2020. Stripe's guide warns that data collected to fulfil orders may not be usable for training pricing or recommendation models, and that firms stay accountable for automated decisions affecting customers.

Wholesalers have more time, because McSaveney reports that Shopify's agent tools don't appear on B2B checkouts.

Practical next steps

• Open AI performance insights in Merchant Center if you sell through Google, and record a baseline before Black Friday trading starts so January's numbers have something to compare against.
• Pick your 50 best-selling products and check that price, stock, dimensions, compatibility and delivery details are complete and consistent. Correct errors in the source system.
• Rewrite vague returns, shipping and opening hours pages in plain terms. McSaveney notes storefront tools answer customer questions from those pages, so outdated wording can cost a sale.
• Ask your IT provider to review firewall, bot and CDN rules, and to record which automated traffic is allowed and why. On Shopify, ask a developer to use the WebMCP testing flag in Chrome to see which tools your store registers.
• Put written questions to your commerce platform: can you identify agent-placed orders, can you refuse them, and who handles disputes?
• Name one person to own AI discovery, and decide in January whether the traffic justifies more spending. Firms that want help setting that ownership and the wider rules can look at our AI consultancy work.