Episode 13: Marketplaces, Acronym Soup and Practical AI

Ask most people in this industry where Australia sits and you get an answer about lag. A few years behind the UK and US on ecommerce maturity, on fulfilment and delivery networks, on technology adoption, on consumer expectations.

That is broadly accurate.

On AI adoption it is exactly backwards. Australian businesses are as far along as anyone we speak to, and in places further. Not despite arriving late, but because of it. Watching a hype cycle play out somewhere else first turns out to be an advantage.

We spent two weeks there recently, at Retail Fest and in customer meetings. Three things came up in nearly every conversation.

The catalogue arms race

Australian retail loves a marketplace. Most of the large retailers are either setting one up or already running one, using their digital storefront as a shelf for third party sellers who list and drop ship directly. The standard model, arrived at a few years after the UK and US went through the same evolution.

The result is an endless aisle and a catalogue size arms race. On some Australian retail sites you can now buy more or less anything from more or less anyone.

Delivery has moved with it. Three or four years ago, ordering something in Australia meant waiting, sometimes a week, even from Amazon. Next day delivery now covers most states and the major cities. Neither of those shifts works without the other.

What has not kept pace is onboarding. Two or three of the retailer meetings we had were entirely about the same problem: they cannot introduce marketplace products into their ranges fast enough. Seller data arrives incomplete, or the process to take it in does not run at the speed the commercial ambition requires.

The barrier is the portal, not the seller

There is a comfortable assumption underneath a lot of marketplace strategy. The seller owns the listing. The seller wants their products live and selling. So the seller has every incentive to provide good data, and the problem solves itself.

It does not, and the reason is not motivation.

What we saw repeatedly is that willingness was never the constraint. The technology is. Spreadsheets and old-school mechanisms for getting data in, validation loops between retailer and seller over whether the right attributes are present and the listing is complete, and a portal experience that makes all of it harder than it needs to be.

Sellers are not refusing to onboard. They are being asked to do it through tooling that makes it slow.

Acronym soup

The PIM market in Australia is less saturated than in the UK, Europe or the US. Fewer vendors, less noise.

More interesting is what that has done to the conversations. Roughly 90% of the discussions we had were not about needing a PIM. They were about needing better ways to onboard and enrich data. The market appears to have learned something by watching the UK and US work through their PIM cycle first. Brands and syndication is still a real use case there, particularly on the FMCG side, but the reflex purchase is noticeably absent.

Where the acronyms get tangled is assets. Onboarding means data and assets together, and once you hold a master asset you have to prepare it for every downstream use: resizing, background removal, format conversion. A lot of PIMs do not do that. You still need somewhere to store master assets, but storage is not manipulation.

It is the same distinction we keep making about data. A database holds your product data. It does not enrich it. A DAM holds your assets. It does not transform them.

Stop comparing acronyms. Or at least understand which one you are buying, because a PIM is many things to many businesses.

A real example

The best conversation of the trip was with Australia's leading bonsai specialist. A genuinely niche catalogue, and a problem familiar to anybody buying from smaller suppliers: for many products, no supplier imagery exists at all.

So they photograph the products themselves, in the warehouse, and use AI image models to build the product records from there. A footwear business at the same event was doing exactly the same thing.

Now hold that against the other end of the market. We spoke to a large multinational that leans heavily on one of the big four consulting firms. Strip out what is actually being delivered for the fee, and a substantial part of it is people keying data into a PIM.

One business has a few thousand SKUs, a small team and very little budget. The other has all the budget there is. The first has automated a job the second is still paying consultants to do by hand.

The constraint was never resources. It was permission to change the process.

Three kinds of AI vendor

The marketing hype phase is over. Nobody is impressed by AI appearing on a product sheet any more. What is left, judging by the vendor floor at Retail Fest, sorts into three groups.

Tools that embed AI into a specific operational workflow. A defined job, done end to end, with the model inside the process rather than bolted onto it. Product data operations is one of these, because onboarding and enrichment are unusually easy to describe as an AI use case. It is mostly natural language and structured attributes, and people grasp it immediately.

Agentic workflow builders. Whatever you need, built to order. There were several, and most were closer to consultancies than products. The honest version of their pitch contains the interesting question: you could probably use Claude or a similar model to do a lot of this yourself, so the value has to sit in industrialising it into something lean and reliable.

Customer service agents. A crowded category, and the one most people picture when they hear AI in retail.

The group worth being careful about is the middle one. Given how fast the frontier models are moving, a lot of AI startups sitting between the model and the workflow will struggle to stay ahead of the thing they are built on.

The takeaway

The pattern across all three topics is the same. Where Australian businesses are behind, it is on infrastructure and maturity. Where they are ahead, it is on judgement.

Distance does some of that work. It is hard to sell into Australia from the other side of the world and harder to support it, so vendors who go have to commit properly, and buyers know it. That filters out a lot. The market is small, reasonably close-knit, and blunt in a way that shortens sales cycles. You find out quickly whether somebody is interested, rather than nine months in.

Constraint does the rest. Most of the businesses we work with there are mid-size retailers and distributors, 10,000 SKUs and up, with small teams. They lean on automation because they have no realistic alternative. Nobody is running a proof of concept to see whether it is interesting.

Which is the part that travels. The businesses making the best product data decisions are not the ones with the most budget or the newest platform. They are the ones sceptical enough to ask what problem they are solving, and constrained enough that the answer has to work.

Listen to the full episode

Episode 13 of Product Data Weekly is available now. For more episodes and the weekly newsletter on operational issues inside product data and ecommerce teams, visit productdataweekly.com.