Online retail is expanding into a multi-dimensional ecosystem, with the modern storefront evolving to serve two distinct customer types: humans and AI agents.
Although 39% of Australian retail SMEs are already using AI in some capacity, much of that investment is focused on improving the human experience. Meanwhile, the architecture required to attract the agentic customer is either absent or, worse, being built in silos.
It’s a trap we’ve fallen into before. Post-COVID, many retailers treated their surging online channels and physical stores as disconnected ecosystems – a mistake that is now costing millions to reconcile.
But there’s a lesson here. Rather than treating human and agentic commerce as isolated channels, successful retailers will integrate new AI interfaces with existing touchpoints and build a foundation that serves both the human eye and the machine algorithm.
Side one: Hyper(human) personalisation, redefined by AI
Let’s start with the half that retailers are more used to talking about: the human customer.
For years, the industry promised personalisation. In practice, production economics forced us to settle for broad audience segments. A campaign team could create five variations of a hero banner, maybe ten. The data existed to do more, but the cost made it impossible.
With generative AI, that constraint has now been removed. GenAI has reduced the marginal cost of individualised marketing and dynamic interfaces to near zero. We are moving from segments to a ‘segment of one,’ where engagement is continuously tailored to each person’s exact context, intent, and history with the brand.
Publicis Sapient research finds that 32% of Australians already use and find value in personalised discounts, and 29% appreciate tailored recommendations when shopping online. Those numbers will only climb as the experiences get better.
This is driving the emergence of generative user interfaces, and with it, a fundamental shift in how retail experiences are built and delivered.
For years, organisations invested heavily in the “golden path”, designing ideal customer journeys through fixed screens. But the future online storefront won’t be a static homepage. It will be a blank canvas, assembled in real-time based on user intent.
Side two: Optimising for the machine customer, not eyeballs
Here is where it gets uncomfortable.
While retailers invest in making human experiences more engaging, a second type of customer is emerging that does not care about any of it.
With the introduction of frameworks such as Google’s Unified Commerce Protocol (UCP), the foundations for machine customers are now live, and agents are able to browse, compare and purchase on behalf of consumers. These agents do not respond to visual design or storytelling. They respond to API latency, structured data and machine-readable product catalogues. If your infrastructure is not optimised for them, you will not just rank lower, you will not exist.
Signs of this shift are already underway. Google AI Overviews are now reducing organic click-through rates by 58%, weakening the traditional search-driven path to a retailer’s website. Platforms are beginning to restrict crawler access and are actively litigating against AI agents that transact without authorisation.
The commercial and legal battles are just beginning, but the trajectory is clear: a growing share of commerce will be mediated by agents, not humans.
This will demand something most retail technology teams are yet to seriously consider: building an entirely separate ecosystem for a fundamentally different type of user.
Enter the “dark storefront,” a high-speed API gateway designed exclusively for AI to shop from. No visual design. No brand experience. No emotional engagement. Just clean, structured, data detailing your product, pricing, availability and fulfilment options in real time.
This is a starkly different technology problem, calling for new architecture, performance benchmarks and even ways of thinking about your product offering.
In this context, your product is not the item on the shelf, but the data about the item on the shelf, and how easily that data can be accessed, interpreted and transacted on by an autonomous agent.
The question retailers need to answer
The split between the human and agentic customer, and the very different approaches they require, will challenge retailers to rethink their strategy.
Retail has traditionally been built on owning the customer relationship. Brand, channel, experience and data all support that goal. Every dollar spent on a loyalty program, mobile app, or beautifully designed e-commerce experience has been justified to bring customers closer and keep them there.
The dark storefront upends that logic entirely. If an AI agent is going to buy your customer’s products, does the customer need to visit your website? If the agent is making the recommendation, do you need to own the discovery experience at all? And if not, do you need to maintain the expensive channels you have built to facilitate it?
Faced with these challenges, one option is to double down on owning the channel. Invest in experiences that are personalised and adaptive enough to remain compelling for customers. Build stickiness through value that only a direct relationship can deliver.
The alternative: let go of the channel. Accept that for a significant and growing share of transactions, you will be relegated backstage. Invest in making your catalogue, inventory and fulfilment data clean, fast and easy for agents to use so you become the default supplier, even if the customer never sees your brand. You may not own the customer, but you’ll still make the sale, while shedding the cost of maintaining expensive consumer-facing channels and gaining access to a vastly broader user base of agents shopping on behalf of those who would never have discovered you otherwise.
Both paths have merit. In truth, there may not even be a choice between them.
By Harshu Deshpande, Group Vice President, Product and Engineering at Publicis Sapient
This article was first published by RetailBiz