The problem
Tires Club is a tire and wheel store in Zaporizhzhia with its own fitting centers. Customers message with the same questions over and over: “What do you have in 205/55 R16 for winter?”, “Which wheels fit my car?”, “When can I book fitting?”, “Where are you and are you open now?”.
To answer, a manager has to collect the parameters, check stock, pick options and take contact details. Evenings and Sundays nobody answers, and some customers go elsewhere.
The store didn’t need a scripted bot. It needed an assistant that knows the real stock, talks like a person and takes the conversation all the way to an order or a booking.
The solution
We built an AI assistant in Telegram that is wired into the store’s processes.

What the bot does
- Matches tires by width, profile, diameter and season, taking driving style and budget into account. It offers in-stock items for pickup first.
- Matches wheels by type, diameter, PCD and fastening, and sends photos straight away — for wheels, looks matter.
- Understands real-world input: “195 65 15 winter budget” instead of a full size, misspelled brands, Cyrillic instead of Latin.
- Helps when the customer doesn’t know the size: suggests the typical one for their car and asks to check the sidewall marking.
- Places orders for pickup or delivery across Ukraine in 2–3 days.
- Books tire fitting: collects name, phone, car and preferred date; a manager confirms the slot.
- Gives up-to-date store information from Google Maps: addresses, phones, opening hours with live open/closed status, rating and a map link.
- Hands over to a human what a human should handle: discount requests, complaints, “call me” — the bot collects contacts and instantly notifies the managers’ work chat.
How it works
- Stock syncs every hour from the store’s spreadsheets, so the bot never offers what isn’t there.
- The AI uses tools instead of making things up. The model calls 23 functions: tire and wheel search by parameters, brand, size and season reference data, order placement, fitting booking, store details and manager escalation. Prices and availability come only from the database.
- Web admin panel with an archive of every conversation: short and full view of each chat, user access management. The store sees what customers ask, and we use it to keep improving the bot’s guidance.

- Separate flows for tires and wheels, because the matching logic differs.
Results
| 199 | customer conversations |
| 41 | orders placed through the bot |
| ≈ 1 in 5 | conversations ended in an order |
- Customers get an answer instantly, including evenings and weekends.
- Managers receive qualified orders and bookings with contacts and parameters, not just “message me”.
- Complex cases don’t get lost: the bot hands them to a person with the reason attached.
Tech stack
Go · PostgreSQL · Telegram Bot API · OpenAI (GPT-4.1, function calling) · Google Maps Places API · Docker
Want an assistant that knows your stock and closes the conversation with an order? Discuss your project