Shopping by Conversation, Not Search
Pick n Pay, South Africa's second-largest retailer by revenue, is launching an artificial intelligence shopping assistant named Penny inside its asap! delivery app, with a rollout beginning July 6. Built on Google's Gemini models, Penny lets customers build a grocery order conversationally rather than by hunting through categories. Shoppers can speak a voice note in any language, type a request, or upload a photo, including a picture of a handwritten shopping list, a recipe, or a product they want to buy, and the assistant assembles the basket. It is a deliberate attempt to strip the effort out of online grocery, which remains a chore of search boxes and exact product names.
The framing from Pick n Pay is telling. Enrico Ferigolli, the retailer's omnichannel executive, put it plainly, saying that on-demand delivery changed how people shop and that AI is now changing how they order. In a separate remark he added that the next disruption in e-commerce is removing the effort from shopping itself. That is the right way to think about the opportunity. The value of a grocery assistant is not novelty. It is friction removed from a repetitive, low-joy task that customers do every week, and friction removed is a durable reason to keep coming back.
What Gemini Actually Does Here
The technical substance is in the multimodality. Because Penny is built on Gemini, it can accept voice, text, and images interchangeably, and it is tuned around South African shopping habits and languages rather than a generic global template. A customer can photograph a scrawled list on the back of an envelope and have it turned into a basket, or describe a meal and receive the ingredients. Penny also suggests recipes, recommends substitutions, helps with meal planning, and offers budget-conscious options, moving from a search tool toward something closer to a shopping companion that understands intent rather than keywords.
Crucially, Penny personalizes through data the retailer already owns. By integrating with the Smart Shopper loyalty program, it tailors recommendations to a customer's past purchases, preferred brands, quantities, and price sensitivity, so each interaction is meant to be more relevant than the last. This is where an incumbent grocer holds an advantage that a pure technology company does not. The assistant is only as good as the context it can draw on, and years of loyalty data is exactly the context that makes recommendations feel personal rather than generic. The model is the interface, but the proprietary data is the moat.
A Companion, Not an Autonomous Buyer
For all the talk of agentic commerce, Pick n Pay has drawn a deliberate line. Penny helps assemble the basket but does not autonomously complete purchases. It hands off to the checkout, leaving the final confirmation to the customer. That is a conservative and, we think, sensible choice at this stage. Grocery orders involve substitutions, budgets, and dietary needs where a wrong autonomous decision is not an abstract error but a missing dinner or an unwanted charge. Keeping a human at the point of purchase preserves trust while the assistant proves itself on the lower-risk work of discovery and basket building.
This restraint is a useful counterpoint to the industry's more breathless agentic commerce rhetoric. The vision of fully autonomous shopping agents that buy on your behalf is real and coming, but the near-term value is in assistance, not automation. We would argue Pick n Pay has the sequencing right. Win customer trust on the tasks where the stakes are low, learn from real behavior, and expand the assistant's authority only as reliability is demonstrated. Ferigolli has signaled that more AI features will follow in the coming months, which suggests a staged path from helper toward greater autonomy rather than a leap.
Answering a Rival That Moved First
The competitive context is impossible to miss. Penny arrives just months after Checkers Sixty60, run by larger rival Shoprite's ShopriteX unit, introduced its own AI assistant, Pixie, in April. The two take different routes to the same destination. Penny leads with conversation, starting from open-ended dialogue, while Pixie emphasizes predictive replenishment and a swipe-based smart basket. Ferigolli has acknowledged that both approaches will eventually converge toward similar capabilities, which is a candid admission that the differentiation today is execution and speed, not fundamentally different visions of the future.
We see this as a textbook example of AI raising the competitive stakes in a mature category. Grocery retail is low margin and fiercely contested, and neither player can afford to let the other own the more convenient shopping experience. When one incumbent ships an AI assistant, the other cannot sit out without ceding the perception of being behind. That dynamic tends to accelerate adoption across an entire market, as rivals match each other feature for feature. For South African shoppers, the practical result is that conversational grocery ordering is becoming a baseline expectation rather than a novelty within a single season.
The Turnaround Bet Underneath
Beneath the product story is a business story. Pick n Pay has endured years of weak trading and lost market share to Shoprite, and it is in the middle of a turnaround in which strengthening its online offering is central. Penny is not only a feature. It is a lever in that recovery, an attempt to make the asap! app compelling enough to win back digital shoppers and defend the ones it has. For a retailer under financial pressure, choosing to invest in a flagship AI experience is a statement that digital convenience is where it intends to compete, even while it repairs the rest of the business.
The lesson generalizes well beyond one grocer. AI capabilities are increasingly deployed not as speculative innovation but as instruments of competitive survival, especially by incumbents fighting to hold ground. Pick n Pay is wisely building on the assets it already has, its loyalty data, its delivery network, and a partnership with Google for the underlying models, rather than trying to become an AI company itself. That is the pragmatic pattern we expect to see repeated across retail: pair proprietary data and existing logistics with a frontier model provider, and aim the result squarely at the customer experience that decides who wins the weekly shop.
