Case study · AI applied to ecommerce
An assistant that sells with real data
AI shopping chat grounded in a real store's catalog: it answers about ingredients, allergens and budgets — without making anything up.
The problem it solves
90% of online store chatbots frustrate users: generic answers, made-up information, or a menu of options disguised as conversation. The challenge isn't adding AI — it's making it answer only with the catalog's truth.
This assistant is connected to the real data of the performance case study's store: every ingredient, price, rating and description. When a fact isn't in the data, it says so and refers to the store's contact.
Real conversation
What it can do
- Allergens and ingredients quoted literally from the catalog — nothing invented
- Budget recommendations with the math done right
- When a fact isn't in the data, it admits it and refers to the store contact
- Politely declines any topic unrelated to the store
- Replies in the customer's language
How it's built
- Full catalog injected as model context — single source of truth
- Server-side proxy: the API key never reaches the browser
- Model fallback chain: if one fails or saturates, the next takes over
- Per-session and per-request abuse limits
- Swappable AI provider (Gemini, Claude, OpenAI) by touching one file
- Performance cost: +2 KB — the store keeps its 96/100 mobile score
Want this in your store?
This same pattern works with any catalog: products, services, pricing or documentation. If your store — or your clients' — needs support that sells without making things up, let's talk.
Let's talk