--- title: "An AI Agent in Retail: Availability, Advice and Orders" description: "Customers do not ask about the product they can see on the website. They ask what the website cannot tell them — whether the shop has it and whether it suits them." canonical: "https://interfase.sk/en/news/an-ai-agent-in-retail-availability-advice-and-orders" lang: "en" last_updated: "2026-11-09" --- # An AI Agent in Retail: Availability, Advice and Orders > Customers do not ask about the product they can see on the website. They ask what the website cannot tell them — whether the shop has it and whether it suits them. Published: 2026-11-09 - Most retail enquiries revolve around three things: availability, comparing products and order status. - The agent is exactly as good as the stock data is accurate — wrong availability is worse than no answer. - Advice on choosing has to come from specifications and descriptions, not impressions; otherwise the agent recommends the wrong thing. - The agent should not push for a sale. "This will not suit you" builds trust faster than an extra recommendation. "Do you have it in the Bratislava shop?" "What is the difference between these two?" "Where is my parcel?" Three questions that make up most customer communication in retail. They repeat every day, the answer is in a system every time, and a person still goes looking for it. ## Availability is the most wanted and the most risky A question about availability in a specific shop is the most common one, and also the one where most can go wrong. If the agent says "we have it" and the customer makes a wasted trip, the damage is worse than not answering at all. So **the quality of the answer follows the quality of the stock data**, not the model. What that means in practice: - Answer from **the live state**, not from an overnight export. - Distinguish **free from reserved** quantity. - Phrase low stock carefully — "last item, we recommend reserving it" rather than "we have it". - If the system does not track per-store availability reliably, **this feature does not get deployed** until it does. The accuracy of stock data connects to [automating stocktaking and inventory movements](/blog/automatizacia-inventury-a-skladovych-pohybov). ## Help with choosing This is where the agent adds the most value and where failure comes easiest. It works when answers come from **product specifications and descriptions** — dimensions, material, compatibility, what the product is for. The agent compares, explains the difference and flags a limitation. It does not work when the model fills in properties the catalogue does not contain. A customer takes "it is waterproof" as binding and the company carries the liability. So: what is not in the product data, the agent does not claim. Incidentally, this task exposes the quality of a product catalogue faster than any audit. If the agent cannot compare two products, neither can the customer. ## Order status and delivery The simplest part with the fastest return. Where the parcel is, when it will arrive, what to do about a delivery problem. The data sits with the carrier and the answer is unambiguous. What matters is that the agent can tell an ordinary delay from a real problem and hands the second to a person — a principle we describe in [an AI agent in logistics](/blog/ai-agent-v-logistike-sledovanie-zasielok-a-vynimky). ## Complaints belong elsewhere An enquiry that opens with a description of damaged goods is not an availability question. The agent should gather the material and hand the case on, not decide on entitlement. There is a separate piece on [an AI agent for complaints and returns](/blog/ai-agent-pre-reklamacie-a-vratenie-tovaru). ## Do not push for the sale The temptation to have the agent attach a recommendation to every answer is strong. In practice it lowers trust and raises returns. The opposite works better: an agent that says **"this is not suitable for what you need"** keeps a customer for a long time. Sales do not fall as a result — what falls is the number of bad purchases that come back anyway. ## What to measure - **Share of enquiries handled without staff.** - **Accuracy of availability answers** — how often a customer made a wasted trip. - **Return rate** on products the agent advised on, against the rest. - **Number of enquiries handed to a person**, and whether they were the right ones. ## Where to start With order status. The data is accurate, the risk is nil and the agent goes live in days. Help with choosing comes next, once it is clear what the product catalogue looks like — and per-store availability last of all, once the stock data holds up. If you are working on this, [get in touch](/kontakt). More on [AI and automation](/riesenia/ai) and examples in our [case studies](/referencie). --- Web version: https://interfase.sk/en/news/an-ai-agent-in-retail-availability-advice-and-orders llms.txt: https://interfase.sk/llms.txt