AI assistant for your website

The speed of the first reply decides more than the price: people usually write to three companies and order from the one that answered first.

Look at when people actually write to you. Most of them open your site in the evening, after work, from the sofa, on a phone. At ten at night your manager is home, the “leave a request” form stays silent, and the person has a question right now. In the morning you call back — and they have already arranged things with someone else.

The assistant removes that pause. It replies within seconds at any hour, takes no lunch break and keeps no queue: it can hold twenty conversations at once, and each person feels they are being spoken to personally. Compare that with a standard contact form — a message into the void, where the visitor cannot even tell whether it was read.

• evenings, nights, weekends and holidays — no gaps
• an answer in seconds, not “please wait for our call”
• several conversations at once, with no queue
• peak-season traffic does not slow the replies down

Holidays make the point best. The shop is closed for three days, and those are exactly the days when people calmly pick out a gift, furniture or a renovation. Everything they managed to ask about will be waiting for you on Monday as enquiries with a name and a phone number, instead of lost visits.

To be honest: the assistant answers instantly, but a real person still has to call the lead back. Agree who goes through the overnight enquiries in the morning — otherwise the speed you gained is lost at the very next step.

The assistant does not know “products in general” — it knows your price list, your lead times and your terms, because it was trained on your own materials.

The difference from the well-known chatbots is roughly the difference between a passer-by and your own salesperson. The passer-by will sensibly explain how interior doors are chosen in general. Your salesperson says something else: “this model is in stock, it costs this much, fitting is charged separately, measuring is free”. We build the assistant to be the second one.

To do that we hand it your site pages, the catalogue, the price list, the frequent questions and your working rules — and from then on it checks that base before every answer. In a tool store the assistant holds a catalogue of more than a thousand items and sees what is currently out of stock.

• site pages and service descriptions
• the product catalogue with prices and stock
• frequent questions and your managers’ answers
• the rules: delivery, payment, warranty, limitations

There is an honest limitation here: the assistant knows exactly what you gave it. If a price list from two years ago is still sitting on your site, it will confidently quote the old figure. So before launch we go through the knowledge base together, and for shops we connect the catalogue directly, so prices and stock update themselves.

A practical tip: write down the twenty questions you are asked most often, with honest answers. That is half the knowledge base — and a list of what is missing from the site itself.

A conversation without a contact is worth nothing, so the assistant’s main job is to bring the visitor to an enquiry with a name and a phone number.

The assistant tells the difference between advising and the moment when asking for a contact is appropriate. While someone is working out how one door differs from another, it simply explains. As soon as “and what would that cost me” or “when could you come and measure up” appears, it offers to take a name and a phone number so a manager can quote precisely.

The enquiry goes where you are certain to see it: to your email, to Telegram — so the phone buzzes right away — or into your CRM as its own record. The whole conversation travels with it, so the manager calls back already briefed: they can see the person asked about front doors with fitting within a certain budget, and they do not start from scratch.

• name, phone number and a convenient time to call
• the full conversation attached to the enquiry
• duplicated to email, Telegram and the CRM
• a note on which page the customer came from

The classic chatbot mistake is asking for a phone number in the very first second, before the visitor has asked anything at all. It scares people off: the window gets closed and you end up with neither the conversation nor the contact. So the assistant asks for details at the end of a useful answer, never instead of one.

An enquiry is not yet a sale. Watch not only how many come in, but how many your team actually handled within a day: that is where money is lost most often.

The assistant behaves like an attentive salesperson: it asks questions first and only then suggests an option.

People rarely arrive with a finished order. “I need doors” is not yet a brief: it is unclear whether they mean front or interior doors, what the opening measures, whether fitting is required, what budget they have in mind. A live manager clears that up in a minute, while a website usually just shows a catalogue of a hundred items and leaves the visitor alone with it.

The assistant fills that gap. It asks two or three short questions, filters out the noise and shows a handful of suitable options instead of a hundred. If you have given it a pricing formula, it works out a rough figure right in the chat: for printing, say, that is area, material and print run; for a website, the set of pages and the features needed.

• narrows the task down with two or three questions
• suggests options that fit the stated budget
• estimates the approximate price inside the chat
• explains the difference between similar options

A chat estimate is a guide, not an invoice. We deliberately teach the assistant to say so plainly: roughly this much, with the exact figure confirmed by a manager after measuring up or a short brief. That way the visitor gets a sense of the price range and does not feel misled when the final number differs.

If your prices depend on complicated conditions, do not be afraid to show a range. Someone who has seen a “from and to” leaves an enquiry more often than someone who saw no figure at all.

Most enquiries boil down to five or ten identical questions, and those are exactly what the assistant takes off your people.

Count it up yourself: how many times a day does a manager type “we ship by courier”, “we are open from nine to six”, “yes, it is in stock”. Each reply is short, but the day is eaten away in pieces: every time they have to break off a difficult conversation where a human really is needed.

The assistant absorbs that flow and hands the manager a warmed-up person, with the task clarified and contact details taken. It is not a replacement for staff: tough negotiations, haggling and non-standard orders stay with people. What changes is how their working day begins.

• prices, lead times, delivery, payment, warranty
• opening hours, address, how to get there
• stock levels and product specifications
• order status, when the shop is connected

A useful side effect is the conversation log. After a month you can see that people asked twenty times about collection in person, which is nowhere on the site, or kept getting confused about sizes. That is a ready-made list of what to add to your pages — after which even the assistant gets fewer questions.

Do not set the assistant the goal of “replacing a manager”. The workable version is “free up two hours a day for the people who are already ready to buy”.

The assistant replies in whatever language it is written to — no separate site version and no second bot required.

It picks up the language from the first message. Write in Ukrainian and it answers in Ukrainian; switch to English halfway through and it switches too, without losing the thread. Compare that with a proper language version of a site, which means translating every page and running separate addresses — here the language simply appears by itself.

It shows most where the audience is mixed: orders from abroad, Ukrainians who have moved away, wholesale buyers from Poland or Germany, foreigners looking for a service in Ukraine. Nobody has to switch the site language or type “do you speak English” — they just write the way they normally would.

• the language is detected from the first message
• switching mid-conversation without breaking down
• one knowledge base serves every language
• form of address and tone tuned to your business

“The customer’s language” also means the manner of speaking. The assistant can be taught to sound restrained and formal, as is usual when dealing with companies, or simpler and warmer, as in a small shop. In the same way you can strip out words and promises you do not want made on your company’s behalf.

One honest limitation: the assistant will answer in any language, but your pages and price list stay as they are. If a customer asks for a link, they will get a page in the site’s own language — and that is fine.