Bunnings Doubled Online Conversion. Pay Attention

The most useful Australian AI story this year is not coming out of a lab or a ministry. It is coming out of a hardware chain.

On Thursday, iTnews reported that Bunnings' AI assistant "Buddy", built on Google's Gemini and launched in April, has more than doubled the retailer's online conversion rate and lifted basket sizes, and is now being extended to PowerPass trade and commercial customers, with New Zealand to follow later in 2026. Earlier in the month, iTnews also reported that Bunnings would begin selling through Google's AI Mode, with agentic commerce extending across the Wesfarmers portfolio, and that customers arriving via agentic search were converting at higher rates with larger baskets.

The usual caveat first, because anti-hype cuts both ways: these are Bunnings' own reported numbers, via a trade outlet, not an audited study. Treat "more than doubled" as a strong directional signal rather than a promise about your business. Even heavily discounted, though, a conversion claim of that size from Australia's largest retailer, three months after launch, is worth taking apart properly.

What Bunnings did right

They scoped it to their own ground. Buddy answers questions about Bunnings products, projects and stock. A customer standing in the decking aisle, or browsing at 9pm, asks a specific question and gets a specific answer drawn from the retailer's own catalogue and content. The assistant was pointed at a bounded domain the business controls, which is precisely where current AI is strongest and hallucination risk is weakest.

They attached it to a number that matters. Conversion rate and basket size were the yardsticks from day one. Notice what that changes: within three months of launch, Bunnings could tell a journalist, with a straight face, what the assistant had done commercially. Most AI pilots I see cannot answer that question after a year, because nobody chose the number before they started.

They earned the right to expand. April: consumer launch. June: measured result, then extension to trade customers, the higher-value segment. The sequence ran prove, then scale. Not scale, then hope.

If that sequence sounds familiar, it should. It is the same discipline Bain's ROI research has been begging enterprises to adopt, executed in public by a retailer.

The part that affects you even if you sell nothing online

The quieter half of the story is Bunnings selling through Google's AI Mode. That points at a structural shift worth understanding now: increasingly, the "customer" arriving at your digital shopfront will be an AI agent shopping on a human's behalf, comparing, checking stock and completing purchases.

Agents do not admire your homepage. They read your product data. Which flips the question every business with a catalogue, a service list or a booking page should be asking, from "does our website look good?" to "can a machine accurately read what we sell, what it costs, and whether it is available?"

For most mid-market businesses the honest answer today is no. Product information lives in PDFs, pricing is "call us", availability is tribal knowledge. Every one of those is invisible to an agent, and being invisible to agents will quietly become being invisible to the customers who use them.

A four-step version for a business without Wesfarmers' budget

You do not need Gemini Enterprise and a data science team to run the Bunnings play at your scale.

Step one: baseline the funnel you already have. What share of website enquiries become quotes, and quotes become orders? If you do not know, that is the first fix, because without it you will never know whether anything you deploy worked.

Step two: get your product and service data machine-readable. Structured, current, one source of truth: names, specifications, prices, availability. This is unglamorous data hygiene, it is the prerequisite for everything agentic, and it makes every future AI project cheaper.

Step three: deploy one assistant, scoped to that data. An assistant that can answer "do you have X, what does it cost, when can I get it" from your own catalogue, on your own site. Bounded scope, grounded answers, no ambitions of general intelligence.

Step four: judge it on the baseline. Ninety days, conversion and enquiry-to-quote rate against step one's numbers. Expand if it pays. Kill it without sentiment if it does not.

Bunnings' number will get quoted in a hundred vendor decks by spring, mostly to sell platforms. The better use of it is as a benchmark of method: bounded scope, a commercial yardstick chosen upfront, expansion only after evidence. That method transfers to a 40-person distributor just as well as to a national chain. The budget does not need to.

Where does your business stand? The free AI Scorecard takes three minutes and shows you. If you want a straight steer from a person, book an AI Opportunity Call.