Customers write when it suits them: late in the evening, on a Sunday, in a lunch break. If your answer arrives the next morning, a fair number of them have already asked someone else. An AI assistant on Telegram, WhatsApp or your website can answer straight away. It won't replace your staff, but set up properly it takes a real chunk of the routine off them.
Start with your own inbox
Before you do anything else, read the last hundred messages you received. In most businesses the same few questions keep coming up: opening hours, prices, the address, whether something is in stock, how long delivery takes, how to book. Those are exactly what an assistant should handle.
What it's good at
It answers from the information you give it: prices, services, terms, hours. It copes with questions asked in odd ways, with typos, or in a mix of Romanian and Russian. It can collect what an enquiry needs (a name, a phone number, what the person wants) and it can put a booking in a calendar or take down a simple order. And it doesn't sleep.
Where a person takes over
An assistant can be wrong, especially when a question falls outside what it was told. So we set it up not to improvise: when it doesn't know, it says so and passes the conversation to someone on your team. Complaints, haggling over price, unusual cases and anything that costs money go to a person too.
The rule is simple. The assistant answers what's written down. Anything that needs a decision goes to a human, with the whole conversation attached.
Romanian and Russian in one chat
Here a customer might start in Romanian and finish in Russian. Current AI models handle both, and the assistant can reply in whatever language it's addressed in. Still, test it on real questions in both languages before you switch it on. You'll find gaps.
Connect it to your CRM
An assistant that only answers is useful. One that also records every enquiry in your CRM, with the name, the number and a short summary, is a lot more useful. In the morning your team sees everything that came in overnight, and nothing is left sitting in a chat history. We wrote about what a CRM does in a separate article.
Is it paying off?
Watch four numbers: how many conversations the assistant finished on its own, how many enquiries arrived outside working hours, how long people wait for a first reply, and how many conversations went to a person, and why.
For the first two or three weeks, read the conversations yourself and fill in whatever the assistant was unsure about. We build assistants like this as part of our AI agents and automation service.