4 Reasons Why AI Agents Are Transforming Your E-Commerce Business

AI agents are already helping to determine which products are purchased. Most companies don’t know whether they’re being taken into account.

AI-powered product recommendations now account for up to 35% of e-commerce revenue. Recommended products are about twice as likely to be selected as non-recommended ones. 39% of consumers have already purchased a product recommended by an AI assistant.

The crucial question isn’t whether that’s true. It’s: When was the last time you checked whether your products are even being recommended by Gemini, ChatGPT, or Perplexity when someone searches for something that falls within your product range?

Most companies have a very clear understanding of their visibility in traditional search engines. Rankings, click-through rates, and impressions are tracked weekly. Visibility in AI assistants, however, remains a blind spot for most. This is problematic because that’s exactly where pre-selection takes place today, before a human even gets involved.

Visibility in AI systems follows different rules than traditional SEO. No ad auctions, no keyword density. Instead: structured data signals, product attributes, review quality, availability, and delivery times. If you don’t play by the machine’s rules there, you won’t show up.

It’s not the best product that gets recommended—it’s the one described most thoroughly.

This is the most troubling finding from current trends, because it means that brand awareness and product quality alone are no longer enough.

Products with complete, structured information achieve a purchase share of up to 16% in AI-powered recommendation systems. Products with incomplete or difficult-to-interpret data account for 1.6%. Higher data quality increases actual purchase volume by up to 38%. These aren’t marginal differences—it’s a whole different league.

Specifically: A running shoe whose cushioning, running style, and availability by size are available as structured data points outperforms the same shoe whose description is embedded in continuous text and whose size chart is an image. Not because it’s better. But because the machine understands it.

Today, this is still an advantage for those who do it right. In 24 months, it will be a basic requirement. Structured product data requires time, processes, and governance. The gap that companies are creating today cannot be closed in the short term.

What percentage of your product range is currently machine-readable and fully structured? In practice, the answer is usually sobering.

AI assistants in customer service save costs. Poor AI assistants cost customers.

The efficiency gains achieved through AI in customer service are impressive: 60–70% of routine inquiries are now resolved autonomously by AI, and service costs are reduced by about 30% thanks to AI-powered systems. For many companies, this is the main reason to invest in agentic commerce.
What’s often overlooked is this: efficiency is not the same as trust. Consumers tolerate human errors because they know people make mistakes. From a machine, they expect precision. One wrong suggestion, one inappropriate result, one assistant that doesn’t understand the issue—and trust is gone. Often permanently.

What trust means in concrete terms: After the purchase, Gemini explains to the customer: Cushioning suited to her running style, size 44 available, delivery tomorrow, 30-day return policy, 4.6 stars out of 847 reviews. Not just a recommendation, but a clear rationale. Explanability, relevance, and control determine whether customers trust the agent or not.

For companies, this means a specific task: ensuring that AI assistants justify their recommendations, address individual concerns, and provide a human contact person in case of errors. Those who fail to address this save on service costs at the expense of customer loyalty.

Agentic Commerce doesn’t scale through platforms. It scales through standards.

Anyone who wants to be a player in the world of AI agents must be interoperable on three levels: protocols, product data, and payments.

When it comes to protocols, it’s becoming clear which standards will be crucial in the future. Google’s Universal Commerce Protocol (UCP) is a promising example: it’s already integrated into Google AI Mode and Gemini and is in use with partners like Shopify, Walmart, and Target. Those who aren’t connected there will still appear in search results, but their products won’t be shoppable. The Model Context Protocol (MCP) also connects a merchant’s own store to AI agents via a single standardized interface. When it comes to product data, the same principle applies as in Thesis 2: An agent only recommends what it understands. A size chart as an image or the term “stability shoe” in the body text aren’t matters of style—they’re deal-breakers. And when it comes to payments, new standards are emerging that enable audit-proof transaction chains between machines. For the Swiss market, this is still a trend to watch—but one that shouldn’t be overlooked.

Lesson from Q1 2026: OpenAI’s Instant Checkout was discontinued in March 2026. Insufficient merchant adoption, lack of tax integration, and no multi-item shopping cart. This isn’t an argument against Agentic Commerce. It shows that not every trend takes off, and that it makes a difference to bet on the right standard.

The key question is: Is your commerce architecture currently designed so that you can integrate at all three levels—protocols, product data, and payments – without having to rebuild your core systems?