A customer writes at nine in the evening: “Do you still have it in a 38?”. The manager reads the message the next morning, replies politely — and the customer has already bought elsewhere. They did not leave over price or quality. They left because they waited.
That is what a good chatbot does in an online shop: it does not sell for you, it stops you from losing customers between office hours.
The questions that repeat every single day
Open the last week of conversations and count. In most shops the same handful of questions covers the bulk of the messages:
- is it in stock, in which sizes and colours;
- how much delivery costs and when it arrives;
- which payment methods you accept;
- where my order is right now;
- whether it can be returned and on what terms;
- which of these two models suits me.
A manager answers them dozens of times a day, in the same words. What is lost here is not the sale but the time — and in the evening, when nobody answers, the sale goes too.
What the bot takes on without a human
The bot replies instantly, in Romanian and Russian, on Instagram, Telegram and the website — wherever the customer writes, not where it suits you. Specifically:
- checks stock and says honestly when something is unavailable, offering alternatives from the catalogue;
- explains delivery and payment based on the town and the basket value;
- helps choose: asks two or three questions and narrows the catalogue down;
- takes the order details and passes them on, with nobody retyping anything;
- answers at night and at weekends, when your team is asleep and so is your competitor.
What has to stay with a human
This is the part agencies skip and we put in the contract. A bot has no business handling:
- complaints and angry customers — an upset customer wants a person, not a correct sentence;
- exceptions to the rules: a return past the deadline, a discount beyond the usual policy, an urgent delivery;
- large purchases, where the customer decides after a conversation, not from a product page;
- any situation where a mistake costs money or reputation.
The bot recognises these cases and hands them to a human along with the conversation so far, so nobody has to ask again what was already said.
Where the bot gets its answers
A bot invents nothing. It answers from what you give it: the catalogue with stock levels, delivery terms, the returns policy, the frequent questions. If something is written down nowhere, the bot will not know it either — it will say so and call a human.
That is why the first stage of a rollout is not technical but organisational: we gather in one place what your team already knows but keeps in its head. This is usually where the contradictions surface, too: the site says one thing, the direct messages say another.
What the first weeks look like
Week one: we read the real conversations of the past months and pull out the recurring questions and the sources of the answers. Skip this and the bot comes out generic, and nobody uses it.
Week two: the bot answers in test conversations and you correct it like a new hire. Once the answers look right, we put it on your channels and watch it for a few days.
For an ordinary shop, two to four weeks usually pass between the first conversation and a bot that answers customers.
How to tell whether it is worth it
Do not guess, count. How many messages arrive a day, how long one reply takes, what share of them are those same few questions. Multiply the hours per week by the cost of an hour: that is the price of the routine. Add the customers who write in the evening and never come back in the morning.
If the number looks small, you do not need a bot. We say that plainly at the audit as well — we do not sell automation where it does not pay for itself.
You can talk to a demo bot right now: it is the real bot of an electronics shop, ask it something as a customer would. And if you want to work out what fits your shop, we run a free process audit: 30 minutes online, no strings attached.