Grocery retail is embracing conversational commerce

Online grocery shopping has followed the same pattern for years: search by keyword, filter by category, adjust quantities. Three recent examples from the grocery sector show how chatbot technologies and AI assistants aim to turn shopping into a conversation.

The three approaches taken by Knuspr, Tesco, and Big C differ fundamentally on a technical level – raising the question of what this means for customers and retailers alike.

German online supermarket Knuspr, part of the Czech Rohlik Group, is taking a bold step: it's bringing its entire product range directly into ChatGPT, in what the company describes as one of the first official ChatGPT integrations in Europe. This lets Knuspr meet customers inside an AI assistant they already use for everyday tasks, help, and inspiration – for example, to look up recipes or plan their weekly menu.

The approach is designed to be thoroughly conversational. Customers can ask ChatGPT to find a "high-quality olive oil," put together every ingredient for a carbonara, create a quick weeknight dinner plan, or even decode a photo of a handwritten shopping list.

However, a test of Knuspr's ChatGPT integration by retail magazine iXtenso found that the AI doesn't yet have genuinely proactive advisory capabilities. Customers still need to tell the bot themselves about preferences, intolerances, or dietary requirements.

Once the cart is complete, customers switch over to Knuspr with a single click to review and finalize their order. That means a platform switch is still required at checkout – but it also means the retailer keeps the actual sale within its own ecosystem.

Notably, Knuspr didn't stop at a single chatbot. The company has built its own MCP (Model Context Protocol) server, allowing more advanced users to connect Knuspr to other AI tools and agents as well.

Tesco is taking the more cautious route, testing in-house first – though at a scale few competitors can match. The UK retail giant is launching a beta test of a new AI assistant built directly into the Tesco app. The initial focus is meal planning: a natural, two-way dialogue that generates personalized recipe ideas based on dietary preferences, then helps turn a chosen recipe into an actual shopping basket by drawing on the customer's past purchase history.

The rollout strategy stands out in particular: before customers ever see the assistant, Tesco is inviting around 280,000 colleagues to test it in their daily lives. Their feedback is meant to help shape features, uncover gaps, and even help name the assistant.

The assistant was developed together with Tomoro AI, a UK consultancy working in alliance with OpenAI, under strict secrecy since last autumn – alongside a recently signed three-year strategic partnership with AI start-up Mistral to expand further AI capabilities.

In Thailand, Big C, one of the country's largest retail chains, shows what's technically happening behind a friendly chat window. The new Shopping Assistant AI Chat, built on Amazon Bedrock, lets 20 million customers across 1,750 stores order in natural Thai instead of using keywords or category filters.

What's truly remarkable is the technology behind it: multiple specialized AI agents work in parallel – one searches Big C's product catalog, a second matches recipe requirements, and a third scales ingredient quantities. Amazon Bedrock AgentCore routes between them and combines the results into a single, unified answer, while Bedrock Guardrails checks every response before it reaches the customer. Type in "tom yum soup for four people tonight," and you get a curated basket of shrimp, lemongrass, galangal, lime leaves, and chili paste – already scaled to four servings.

The business case is backed by early numbers: pilot data points to an estimated basket-size uplift of 5 to 10 percent, as customers discover complementary items they wouldn't have found through a plain keyword search. Order tracking works conversationally too – a simple "where's my order?" pulls up the delivery status.

The three retailers have taken fundamentally different technical paths. Knuspr combines an official ChatGPT integration with a custom-built MCP server and its in-house assistant Maia – one foot in the open AI ecosystem, the other in proprietary infrastructure. Tesco relies on outside expertise, pairing guidance from Tomoro AI (allied with OpenAI) with a fresh strategic partnership with Mistral. Big C goes furthest toward a multi-agent cloud architecture, orchestrating several specialized AI agents on Amazon Bedrock, complete with its own guardrail and development tools.

For customers, one key difference lies in how many platforms the customer journey spans. With Knuspr, the conversation starts in an app many people already use daily, but ends with a platform switch at checkout. Tesco and Big C, by contrast, stay entirely within their own ecosystem – from the first recipe suggestion all the way to delivery tracking, with no break in the experience.

This also affects personalization. A retailer that integrates the AI assistant directly into its own system can immediately connect it to the customer's full purchase history and tailor recommendations accordingly. When the conversation runs through an external, open platform instead, the assistant doesn't yet know the customer's past purchases – preferences, intolerances, or dietary requirements still need to be spelled out by the customer each time. Anyone expecting more personalized recommendations currently benefits more from closed systems built into a retailer's own ecosystem, while open approaches offer greater reach but, for now, work with less deeply integrated customer knowledge.

For retailers, the choice of platform is above all a trade-off between reach and control. Opting for an open AI ecosystem means potentially reaching customers wherever they already are, but it also means giving up part of the customer journey – and part of the data visibility – to an external platform. Investing in a proprietary app instead means keeping all purchase histories and interaction data within one's own system – a clear advantage for personalization and customer loyalty, but one that comes at the cost of significantly higher development effort. The more proprietary the technical path, the greater the investment – but also the more complete the control over processes, data, and the long-term relationship with customers.