How to build an online store chatbot in one evening

First, decide what you want the bot to do

The most common mistake is choosing a platform before defining your goals. Review your customer conversations from the past month and list the questions that come up repeatedly. For most stores, they’re the same: availability and sizes, price with delivery, shipping times, installment payments or cash on delivery, and returns.

• Consultations — answers to questions about products, delivery, and payment.
• Product recommendations — suggest 2–3 products from the catalog based on the buyer’s description.
• Orders — collect a name, phone number, and branch, then issue an invoice.
• Handoff to a person — negotiations, complaints, and unusual requests go straight to a manager.

If the bot handles even the first two tasks, your managers will save hours each day. If it handles all four, some sales will happen without a person at all—even at night.

Step 1. Give your bot knowledge: website and catalog

Without knowledge, a bot makes things up or gives generic answers. So the first step isn’t setting up flows—it’s training. LEO Chat’s scanner makes this easy: enter your website address, and in minutes it collects products from your site’s microdata into a catalog, and pages about shipping, payment, and returns into a knowledge base. The first 50 pages scanned are free.

If you have a large catalog, it’s more reliable to connect it directly: via an XML/YML feed, CSV, or your store’s API—Prom.ua, Horoshop, or WooCommerce. Prices and stock levels will then update automatically, so the bot won’t sell anything that’s out of stock.

Step 2. Test the answers before launch

Before showing the bot to customers, ask it the questions you listed at the start. Your dashboard includes a sandbox where you can see the answer, its sources, and the cost. If an answer isn’t accurate, add a short response to your knowledge base (FAQ) — the bot will use it right away.

Tip: test “awkward” questions — about discounts you don’t offer, competitors, and requests to “make it cheaper.” A good bot doesn’t make promises it can’t keep and passes negotiations to a manager.

Step 3. Connect the channels where customers already message you

Your bot should be where shoppers already expect to message you. For most stores, that’s your website and Telegram; for some, it’s OLX. The website widget takes one line of code, and you can open it on a test page and show it to colleagues before adding it to your site. Connect a Telegram bot in a minute using a token from BotFather.

• Website chatbot — a 24 KB widget that won’t slow down your page.
• Telegram chatbot — catalog, cart, and payment right in the messenger.
• OLX auto-responder — replies to listing inquiries using your catalog.

Step 4. Set up handoff to an agent

A bot shouldn’t replace people entirely—it should take routine work off their hands. Decide what the bot should pass to a manager: negotiations, complaints, wholesale inquiries, and anything where a shopper asks for a “person.” Conversations from every channel are collected in one “Inbox,” so managers can see the history instead of starting with “Hello, how can I help?”

Step 5. Measure results in revenue

Message volume doesn’t tell you how your business is doing. Track the orders and revenue that came through the bot. Learn how to measure them and where to find the data in our article on how to calculate how many sales your chatbot brought in.

How long does it take?

It takes minutes to learn from your website, half an hour to test in the sandbox, and another half hour to set up the widget and Telegram. Realistically, you can have your first live conversation in one evening. Fine-tuning answers to real customer questions takes longer than setup — and continues during the first few days.