Central Retail
genAI shopping assistant and visual product search
Central Retail (Central group, Thailand) deployed Vertex AI Search visual search and the Chefbot assistant on Gemini 1.5 Flash, obtaining a 10% uplift in conversion rate, a 94% reduction in search time, and product pages generated up to 50% faster.
Key points
- Central Retail (Central group) deployed Vertex AI Search visual search, the Chefbot assistant on Gemini 1.5 Flash, and catalog enrichment.
- Visual search: search time for price/promo/stock categories cut by 94%, conversion up 10%.
- Product pages generated up to 50% faster; Chefbot maps recipes to more than 50,000 Tops SKUs.
- Level B evidence: a quantified Google Cloud case study, corroborated by the Google Cloud press release; deployment launched in October 2024, status confirmed.
Objective
Help customers find and buy faster online, free up personal shoppers for cross-sell and up-sell, and industrialize product page production on a catalog that grows by 400 new products a day.
The deployment
Central Retail, filiale du groupe thailandais Central, exploite Tops, la plus grande chaine de supermarches de Thailande, et un service de personal shoppers qui font les courses pour les clients en ligne. Le groupe a deploye trois usages Google Cloud. Chefbot, lance le 9 octobre 2024 sur le compte officiel LINE de Tops Thailand, est un assistant genAI sous Gemini 1.5 Flash: il genere des recettes achetables selon les preferences du client, puis mappe leurs ingredients a plus de 50 000 SKU Tops pour construire une liste de courses que le personal shopper execute. Sur le site du grand magasin, une recherche visuelle sous Vertex AI Search analyse une image uploadee par le client et la met en correspondance avec le catalogue en temps reel, les resultats etant partages avec le personal shopper. Enfin, un outil d'enrichissement de catalogue sous Vertex AI genere des fiches produit optimisees SEO en anglais et en thai, la ou les merchandisers ecrivaient a la main pour 400 nouveaux produits par jour.
Results Proof B
Quantified Google Cloud customer case study (vendor, T2) reporting conversion up 10%, search time down 94%, and product pages up to 50% faster, corroborated by the Google Cloud Press Corner release (T2) that documents the Gemini 1.5 Flash and Vertex AI Search stack. The metrics measure the AI tools themselves. No financial results source, so B and not A.
How it works
Documented architectureThe stack in detail
- llm Gemini 1.5 Flash Model that powers Chefbot: it maps the ingredients of a recipe to more than 50,000 Tops SKUs to generate a shopping list for the personal shopper.
- plateforme Vertex AI Search Visual search: analyzes images uploaded by the customer and matches them to the catalog in real time; also serves as a grounding layer for Chefbot.
- plateforme Vertex AI (LLM d'enrichissement de catalogue) Generates SEO-optimized product pages in English and Thai from existing descriptions, extracting name, brand, and price.
How it runs, concretely
For ops teams-
1Customer request client
The customer asks Chefbot for a recipe on LINE, or uploads a product image on the department store site.
-
2Matching IA
Chefbot (Gemini 1.5 Flash) maps the ingredients to the 50,000+ Tops SKUs; visual search (Vertex AI Search) matches the image to the catalog in real time.
-
3Execution by the personal shopper humain
The shopping list or search results are passed to the personal shopper, who picks the products and advises the customer.
-
4Catalog enrichment IA
The Vertex AI tool generates SEO-optimized product pages in English and Thai from existing descriptions, following Central Retail content rules.
The match between the customer request (text, recipe, or image) and the Tops catalog (inventory, SKU, stock, promo). If the catalog and stock are not up to date in the grounding source, shopping lists and visual search return the wrong products.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- A structured product catalog with up-to-date SKUs, stock, and promotions
- Recipes or editorial content to map to products for the shopping assistant
- Existing product descriptions to serve as the base for enrichment
Org prerequisites
- Clear product content rules (languages, tone, SEO)
- A link between the AI assistant and a human service (personal shopper) for execution
- An existing high-traffic customer channel (messaging, site) to connect the assistant to
Possible stack
- Managed LLM for the assistant and page enrichment
- Semantic and visual search engine grounded on the catalog
- Real-time connection to inventory and stock
The plan, step by step
- Step 1Structure the catalog (SKU, stock, promo) and expose it to a semantic search layer.Deliverable: Product search grounded on the catalog in real time.
- Step 2Add visual search: match an uploaded image to the catalog and surface results to the seller or the customer.Deliverable: Image search in production on one channel.
- Step 3Connect a shopping assistant that turns an intent (recipe, need) into a list of products mapped to SKUs.Deliverable: Assistant that generates executable shopping lists.
- Step 4Industrialize multilingual, SEO product page enrichment from existing descriptions.Deliverable: Product page generation pipeline at scale.
First step: Pick a high-traffic customer channel and ground a search assistant on the catalog in real time before adding recipe or page generation.
The same pattern, proven in other industries
Amorepacific
genAI beauty advice agent built on a skin diagnostic
Booking.com
genAI trip planning assistant
Burberry
genAI copilot for consultants (search, image recognition, complete-the-look)
Sources
- S1 Central Group: Pioneering the future of retail with AI Interested party archive pending
- S2 Central Food Retail Group Debuts Generative AI-Powered 'Tops Chef Bot' with Central Retail Digital and Google Cloud Interested party archive pending
- S3 First in Thailand! Central Food Retail Group collaborates with Central Retail Digital and Google Cloud to debut 'Tops Chef Bot' Primary archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.