E-shop
AI chatbot for e-shops: how to choose one that sells and does not make things up
HiNewton team, IT DEV · Published · 3 min read
Short answer
An AI chatbot for an e-shop is worth it when it reads products, prices and stock live, recommends available products with cards, and answers about order status. For a Greek store it must also handle inflected forms, typos and Greeklish. At HiNewton, e-shop plans start at €79 a month.
How does it differ from a simple FAQ chatbot?
A bot that only reads website text can answer about shipping, but not "I need a gift under €50 that arrives by Friday". That takes a connection to the store's data:
| Capability | FAQ bot | Agent connected to the store |
|---|---|---|
| Shipping, returns, opening hours | Yes | Yes |
| Current price and stock | No, or out of date | Yes, from the store |
| Recommendation by need and budget | No | Yes, with product cards |
| Order status | No | Yes, with an order connection |
| Transfer to a person with history | Rarely | Yes |
Which 10 questions test it on your catalogue?
Vendor examples always show the ideal catalogue. Before the chat opens to customers, test it on your own products and type these the way a customer would:
- Greeklish: write the question in Latin characters, e.g. "exete kati gia...". Does it find products?
- Inflection and typos: misspell a product or use another grammatical case. Does it recognise it?
- Budget: "something under €30". Do all its suggestions respect the limit?
- Sold out: ask for an out-of-stock product. Does it say so clearly and offer an available alternative?
- Describing a problem: describe the need without naming a product. Does it ask a useful clarifying question?
- Comparison: "what is the difference between these two?" Does it rely on the real specifications?
- Follow-up: after a suggestion, ask "is there anything cheaper?" Does it remember the subject?
- Order: ask where a real order is. Does it ask for verification before it answers?
- Outside the catalogue: ask for something you do not sell. Does it admit that, or push something irrelevant?
- Policy: ask about a discount that does not exist. Does it invent one?
Question 8 is also about personal data. A bot that gives the address or contents of an order to anyone who types a number is a security problem, not just a poor experience.
Which mistakes cost sales?
- Long answers. A customer on a phone wants two lines and products, not a paragraph.
- Too many options. A few targeted suggestions help more than a long list.
- Suggesting out-of-stock products. Every product that cannot be bought wastes the customer's trust.
- No route to a person. According to a Gartner survey, 87% of customers consider access to a person essential when a company uses generative AI in customer service.
- Nobody reads the conversations. Unanswered questions show what is missing from your descriptions and catalogue. Check them weekly.
- Counting only conversations. Measure sales from those who talked to the bot against those who did not, bearing in mind that people who open a chat are already more interested.
What does it cost and which plan fits?
At HiNewton the criteria are catalogue size and chat minutes:
| Plan | Per month | Products up to | Chat minutes | Setup |
|---|---|---|---|---|
| E-shop Starter | €79 | 1,000 | 400 | €190 |
| E-shop Growth | €179 | 5,000 | 900 | €390 |
| E-shop Scale | €349 | 15,000 | 1,800 | €490 |
WooCommerce connects as a ready-made, read-only integration, and other platforms through an API. The same agent can also answer the phone for customers who prefer to call. The full breakdown is in the cost guide.
Frequently asked questions
Does it work on WooCommerce?
Yes. The HiNewton agent connects to WooCommerce read-only and answers about products, prices and stock, with product cards and photos in the chat.
Does it understand Greeklish?
A modern language model usually understands Greeklish, but that does not guarantee it finds the right product in the catalogue. Test it with your own searches before you decide.
Will it increase my sales?
No percentage holds for every store, and anyone promising one deserves caution. Measure it in your own store by comparing customers who used the chat with those who did not.
Can the customer talk to a person?
Yes. When they ask, a member of your team takes over the same chat and sees the whole conversation history.