Journal / Kirill Naumov
An independent look at Central Group’s retail AI work in Thailand: shoppable recipes, visual search, catalogue quality, human handoffs and measures worth checking.
By Kirill Naumov ·
“Let’s add a chatbot” is a popular opening line in retail AI. It can also produce a chatbot that explains the returns policy beautifully while the customer is still wondering whether the blue chair will fit through the door. A better starting question is: where do customers or colleagues lose time, confidence or context?
A Google Cloud case study of Central Group in Thailand offers a useful example. It describes connected work across Tops supermarkets and department stores: a recipe assistant, AI-assisted bilingual product descriptions, visual search and a customer-insight assistant. This is an independent analysis of a vendor-published case study, not a project carried out by Kirill Naumov or an independently audited result.
At Tops, Chefbot proposes shoppable recipes based on customer preferences. According to Google Cloud, it maps ingredients to more than 50,000 Tops-specific SKUs and prepares a shopping list for a personal shopper. That connection to a real catalogue matters more than the charm of the conversation. A meal idea is useful only if the store can supply the ingredients, handle substitutions and tell the customer what happens next.
The personal shopper remains in the loop. The assistant can prepare a candidate basket; a person can resolve ambiguity, timing and the occasional “I meant the other kind of basil”. This is a design choice, not an admission that the technology failed. The goal is a better completed order, not an impressive answer in an empty chat window.
For department-store customers, the case describes image-based search: a customer uploads a picture, the system matches it with catalogue items and shares results with a personal shopper. Google Cloud reports that searching within price, promotion and stock categories fell from 32 minutes to one or two minutes, with a 10% conversion uplift. Those are the vendor’s reported figures, not a forecast for another retailer.
The transferable question is narrower and more useful: how long does it take to find a suitable, available product before an employee can help someone make a decision? If the image match points to an unavailable item or the next handoff is unclear, a faster search has merely moved the delay.
The case also describes merchandisers preparing English and Thai descriptions for roughly 400 new products a day. An AI catalogue-enrichment tool extracts attributes from existing descriptions and generates copy under Central Retail’s content guidelines. Google Cloud says descriptions can be prepared up to 50% faster. That is interesting, but speed has to be paired with factual control.
A reliable workflow retrieves approved product attributes, drafts within the rules, validates facts and sends uncertain cases to a merchandiser. Size, materials, allergens, compatibility, availability and promotional terms are not creative prompts. A confident error repeated across thousands of product pages is a very efficient way to increase returns.
First, start with a stuck customer moment rather than a technology label. “People cannot find an item from a photograph” is a better brief than “we need computer vision”. Second, connect the assistant to current catalogue facts, stock, prices and the person who owns the next step. Third, measure the full job: time to a suitable product, successful fulfilment, substitution and cancellation rates, conversion, returns and employee rework. A click-through rate is a signal, not a business case wearing a tiny hat.
For a first pilot, choose one category, one customer task and one operating team. Record the current baseline, test the AI-assisted step beside the existing process and audit a sample of matches and generated product facts. Agree before launch what would justify expanding, revising or stopping the pilot. Central Group’s published case is not a plug-and-play blueprint; its useful lesson is to treat trustworthy retail information and the human handoff as part of the product.
Explore the English AI for Retail workshop → · Discuss a retail AI consulting brief
Case-study source: Google Cloud’s Central Group customer story. Its figures are vendor-reported, not independently audited or guarantees of another retailer’s result.