Malin Gaertig.
Malin Gaertig, Founder MARK - Build For Selection, Agentic Commerce Strategy
Ask an AI assistant what makes a good pair of running shoes and it'll answer from memory. Ask it to find you a pair, in your size, in stock, arriving before Friday and it has to go and check.
That difference, answering from memory versus actually checking, will decide which brands benefit from AI and which ones lose revenue without ever knowing why.
Deloitte Digital puts AI-agent-influenced consumer spending in Australia at close to $1 trillion annually. That's the size of the channel. At this point, the majority of that channel is still organic which means big media budgets don't give you a leg up here the way they do everywhere else.
Two very different games
Memory mode is the AI model answering from what it already knows about a brand based on its training data. It rewards what's always mattered, being well known, well reviewed, talked about a lot.
Grounded mode is the AI model actually going and checking live prices, live stock, live availability, because the answer has to be up to date, not remembered. This is the mode agentic commerce runs on. An agent buying on someone's behalf can't work off old training data. It needs a source it trusts enough to query right now, and a product it can actually locate.
What's surprising is a brand can look completely healthy in memory mode and be practically invisible in grounded mode. I tested that gap directly across 11 Australian retailers. I tried to locate a specific product, the starting point any agent needs before it can check a price, confirm stock or buy anything. 60% of the time I never reached one. The reasons varied but always came down to not-fit-for infrastructure.
When the agent can't find what it’s looking for, it fills the gap by guessing a price, listing a competitor and giving a flat statement that the product doesn't exist. That’s one lost sale, but if this happens repeatedly the agent will start prioritising the brand.
Content gets all the love. Infrastructure gets ignored.
The layer everyone's optimising, content, is the layer that matters less once an agent is actually shopping. What decides whether your products get found, though, is the unglamorous stuff sitting underneath the website e.g. whether your product feed is accurate and current, whether your product pages actually contain readable product data rather than a pretty shell, whether pricing and stock update in real time instead of once a day. If content is the cool kid, infrastructure is its unglamorous stepsibling nobody wants to sit next to at the table.
The bright side is the playing field is wide open and the winners won't be decided by media budgets, but by who has fast, accurate, machine-readable product data an AI agent can actually rely on.