Retailers are losing direct control of the customer journey as shoppers hand the path to purchase over to autonomous AI agents. Salesforce’s Fourth Edition State of Commerce report found that consumer reliance on AI assistants as the first stop for shopping grew 200% year over year. Traditional search traffic to brand owned channels fell 15%, and product discovery through those channels dropped 7%. Newer discovery channels, including AI assistants, social feeds, and delivery apps, grew 38% over the same period.
In South Africa, where ecommerce has reached a tipping point on the back of rapid digital adoption and aggressive marketplace expansion, this shift is felt closely. Retailers have long competed with search engines and digital marketplaces for visibility. AI agents raise the difficulty considerably, because they do more than surface options for a shopper to browse. They assess those options on their own and, increasingly, complete the transaction for the buyer.
Trust now has two audiences to satisfy. There is the human shopper, who judges the finished experience, and there is the AI agent, which forms its judgement of the brand well before that. Commerce leaders expect this to matter quickly: 86% believe large language models will be essential to product discovery within the year. This means a retailer’s product data needs to be accurate, consistent, and structured well enough for an agent to act on it.
Discovery is a data problem as much as a marketing one
Many South African boardrooms still treat discovery mainly as a merchandising task, one to be solved with better creative campaigns, paid placements, and standard search keywords. That approach falls short once agents enter the picture, since an agent does not browse a page the way a person does. It works through structured data to decide what to recommend.
Consider a shopper who asks an assistant to find the most reliable inverter friendly washing machine with next day delivery in Johannesburg. The agent has little interest in a banner ad. It goes straight to the back end data feeds. Some brands have already started building their catalogues with this kind of query in mind, describing products in terms of what they are, who they suit, and why they are worth choosing, rather than optimising purely for old style search keywords.
Salesforce projects that AI agents will generate a fifth of all ecommerce traffic this coming holiday season. Businesses that treat this as a matter of data readiness, worth fixing properly, tend to be the ones an agent chooses to recommend.
Consistency is what earns the repeat recommendation
Retailers who keep earning recommendations from AI agents tend to get the basics right every time: accurate pricing, real time inventory, and fulfillment that holds up.
That consistency is harder to find than it should be. Only 27% of commerce organisations report that their customer data is fully unified across sales, service, marketing, and commerce. Just 32% have fully defined success metrics for their AI efforts. Among businesses with fragmented data, 41% report inconsistent pricing and promotions, and 40% say their inventory is not synchronised in real time.
When a South African shopper follows an AI recommendation and it turns out to be wrong, perhaps a price that does not match at checkout or stock that is not actually available nearby, the retailer absorbs the reputational cost. The algorithm, for its part, quietly learns to stop suggesting that retailer. Leaders who are ahead of this problem have put executive accountability behind a single source of truth for their data, which protects loyalty and gives their AI investment a fair chance of paying off.
A branded agent is really a statement of confidence
Once a retailer has its underlying data in order, the benefits start to show. Early adopters who launch their own branded, on site shopping agents are effectively telling the market that their data infrastructure can handle real customer interactions in public, in real time.
Among businesses further along in unifying their data, the payoffs most often mentioned are closer alignment between sales, marketing, and commerce teams, better outcomes from AI and automation, and stronger customer retention.
The agent itself is not really the point. It is a visible sign of the architectural work that happened long before the launch.
Physical retail is still the sternest test of trust
AI agents have not made the physical store less important. If anything, they have raised what is expected of it. Around the world and in South Africa, 77% of consumers still prefer physical retail for key shopping moments. Meanwhile 79% use their phones while walking the aisles, and 12% already ask an AI assistant for buying advice while standing in the shop. Separately, 88% of B2C respondents say customers now expect the same personalisation in store that they get online, a bar most physical operations were never really built to meet.
When a shopper walks into a store in Cape Town or Sandton, an earlier conversation with an AI assistant has usually already shaped what they expect to find. If the shelf price or the stock on hand does not match what the assistant promised, trust falls away almost immediately. Retailers who extend real time data down to the shop floor turn that earlier credibility into revenue. Those who do not tend to watch customer lifetime value slip.
The operational work retailers have been putting off
A little over 70% of commerce leaders agree that scaling across channels depends on AI and expanding their use of it sits at the top of their strategic priorities. Even so, 59% admit their current technology cannot support real time, personalised experiences at the scale they need.
Closing that gap pays off in different ways depending on the business. B2C organisations report their biggest gains in development speed and personalisation. B2B organisations see theirs mostly in revenue growth, employee productivity and operational efficiency.
Trust in a brand and trust from an AI agent now rest on the same technical foundation. The South African retailers that earn an agent’s recommendation today are likely to be the ones shoppers choose for themselves tomorrow.
- Linda Saunders, Country Manager & Senior Director, Solution Engineering Africa at Salesforce
