An AI Assistant That Answers Your Customers While You Sleep

The most expensive lead is the one who wrote at 11:40 p.m. and never got an answer. In the morning your manager opens the inbox, sees “is this model in stock?” — and sees that the person has already bought elsewhere. Technically nothing broke: the site was up, the item was on the shelf, the price was competitive. There simply was nobody there at the moment the customer was ready to pay.

That gap is exactly what an AI assistant closes. Not a support widget with a tree of buttons and “choose a section”. A consultant that knows your range, your delivery and payment terms, your exceptions — and talks to the customer in plain human language at any hour: on the site, in Telegram, wherever your customers already write to you.

What “knows your catalogue” actually means

The difference between a toy and a working tool is the source of knowledge. An assistant that talks “about lighting in general” helps nobody — the customer can search the web on their own. Value appears when the assistant relies on your data and only on your data.

In practice that means it can see:

• products, specifications, prices, availability — from the same source the site itself runs on, so it never quotes yesterday’s stock;
• delivery and payment terms, lead times, warranty, return rules;
• the things that are not on the site at all, but that your managers repeat every day: whether the item fits a stretch ceiling, whether you deliver to a village, whether the paperwork for a sole trader will be issued, whether the colour can be changed.

The second half of the job — the half people rarely talk about — is the limits. The assistant must clearly know what it does not know. When a question falls outside its data, the correct answer is “let me check with a manager”, not a plausible invention. An assistant that improvises a price or promises a lead time that does not exist costs a business more than having no assistant at all: you pay not in money but in trust.

Which questions it closes on its own

In most small and mid-sized businesses, incoming messages boil down to a few dozen repeating questions. They eat the manager’s working day — and they are precisely what gets lost at night.

• Availability, price, the difference between two similar models.
• Selection by parameters: size, power, compatibility, budget.
• Payment and delivery terms, shipping cost, timing.
• Order status and the Nova Poshta tracking number.
• Taking the order itself: items, city, branch, contact details.
• The small objections that otherwise make a person quietly close the tab.

The assistant holds dozens of such conversations at once, does not get tired by the evening, does not forget about the current promotion, and does not tell two customers two different things. There is also a side effect that owners often end up valuing even more than the extra sales: every conversation is recorded. After a month you have an honest picture of demand — what people ask most, what they look for and fail to find, where your site confuses them. That is a ready-made list of what to fix.

Where the line is — and why you must draw it deliberately

An AI assistant does not replace a person. It takes the routine off a person so that the person can do what people are genuinely better at.

A human is needed when:

• the conversation is about bargaining, individual discounts or terms outside the standard rules;
• the customer is upset: a complaint, a defect, a missed deadline — this needs empathy and the authority to actually fix something;
• a complex project has to be calculated, measured or agreed on site;
• it is a large deal with a contract, documents and liability.

How the assistant behaves at that line matters more than the line itself. A good assistant does not keep pretending to be human until the last moment — it hands over: it summarises the conversation, captures the contact and passes both to a manager together with the context. The customer does not have to explain everything twice, and the manager starts the conversation already prepared. Both sides feel the difference.

And one more thing: honesty. The assistant should introduce itself as an assistant. Passing a bot off as “Olena from the sales department” is discovered sooner or later, and it damages exactly the trust the whole thing was built for.

What the business needs to provide

The technical side is our job. What we need from you is what nobody but you knows:

• A single source of truth about your products or services. A catalogue, a price list, a spreadsheet, your accounting system — even an export from 1C works. What matters is that there is one such source and that it is current.
• Answers to the questions your managers hear every week. The simplest way to collect them: ask the managers to write down what they are tired of repeating.
• The rules of what may and may not be promised. Lead times, discounts, returns, exceptions. This is the hardest part — and the most useful one: many businesses put their own rules into words here for the first time.
• The channels. The site, Telegram, messengers — wherever your customers already are.
• A person to escalate to, and an agreement on how quickly they respond.

The launch itself is rarely the bottleneck; the third point usually is. Once the rules are spelled out, a working version appears faster than people expect. After that comes the first month of watching: we read the real conversations, sharpen the answers, remove the places where the assistant gets confused. Going live is not the finish line — it is the start of tuning.

In short

• A lost lead is not the one you turned down, it is the one you answered too late. Nights and weekends are a quiet hole in the budget.
• “Knows your catalogue” means training on your data — and an honest “I don’t know” beyond it.
• The assistant closes the repeating questions: stock, price, selection, delivery, order status, taking the order.
• The line is drawn deliberately: bargaining, conflict, complex projects and large deals go to a human, with the context handed over.
• From the business we need current data, the typical questions and clear rules. The rest is on us.

Our own AI assistant already works on this site — you can simply ask it a few questions about your own situation and see how it feels from the customer’s side. If afterwards you want to work out whether an assistant like this makes sense for your business, write to us. We will be equally honest whether the answer is yes or no.