A Tashkent Instagram store lives in DMs: price, availability, size, delivery, payment — all in the chat. Once you hit 30–40 enquiries a day, an operator can not keep up and the customer leaves for whoever replied first. An AI employee closes exactly that gap: it replies the same second, in Uzbek and Russian, and carries the chat through to payment.
Below is a working launch path. For an SMB it takes from 2 hours; in a simple case it fits into one evening. What actually stands behind the phrase «AI employee», and how it differs from a chat bot, is covered in AI employee: what it is.
| Step | What you do | What you get |
|---|---|---|
| 1. Channel and catalogue | Connect Instagram and, if you have one, Telegram, upload the product table | Every enquiry in one inbox, answers based on catalogue facts |
| 2. Sales scenario | Describe what to ask the customer, how to calculate delivery and which payment methods you accept | The conversation follows your logic, anything out of scope goes to an operator |
| 3. Part of the traffic | Enable AI on a share of enquiries and review 20–30 real conversations | Refined wording without putting the whole flow at risk |
| 4. Full coverage | Extend AI to the whole DM flow and leave the exceptions to an operator | Instant replies in Uzbek and Russian at any hour |
Step 1. Connect the channel and upload your catalogue
Connect the Instagram account and, if you have one, Telegram — they merge into a single inbox. The catalogue can be uploaded as a table: name, price, availability, photo. The AI sources answers from the catalogue, so “is size 42 in stock” is answered with facts, not guesses.
Step 2. Describe the sales scenario
A scenario is not a tree of hundreds of branches. It is enough to describe how you sell: what you ask the customer, how you calculate delivery, which payment methods you accept — Payme, Click, cash on delivery. The AI runs the conversation in that logic and hands anything out of scope to an operator.
The scenario is written in the language of the business, not of a programmer — “ask the city, calculate delivery, offer payment”.
Step 3. Enable it on part of the traffic and check on real chats
Do not open 100 % of the chats at once. Enable the AI on a share of enquiries, review 20–30 real conversations, refine the wording. After a day or two, once answers are stable, widen the coverage. That way you launch without risk and see the result on your own numbers.
Frequently asked questions
How long does it take to set up an AI employee for Instagram
For a small store, from 2 hours. That covers connecting the account, uploading the catalogue as a table and describing the sales scenario. The rest is watching the first conversations and refining the wording; in a simple case the whole launch fits into one evening.
Do you need a developer or a rollout team
No. The channel is connected through the standard Instagram authorisation, the catalogue is uploaded as a table, and the scenario is written in plain words: what to ask, how to calculate delivery, which payment to offer. The store owner or a content manager can handle it.
Does the AI reply in both Uzbek and Russian in the same chat
Yes. The AI platform answers in the customer’s language and switches along with them inside the conversation, including mixed phrases. For an Instagram store that matters: part of the audience writes in Uzbek, part in Russian, and nobody has to be asked which language they prefer.
What happens with an out-of-scope question
Anything beyond the scenario — a dispute, a return, a custom discount — is handed to an operator together with the chat history. The customer does not repeat themselves, and the employee sees the context and steps in only where a human is genuinely needed.
Where does the AI get prices, stock and sizes from
From your catalogue: name, price, availability, photo. Answers are drawn from that data, so a question about a specific size or remaining stock gets a fact, not a guess. Update the table and the answers in DMs update with it.
Can the AI be enabled on only part of the enquiry flow
Yes, and that is the recommended way to start. Enable the AI on a share of enquiries, review 20–30 real conversations and refine the wording. Once answers are stable, widen the coverage — you launch without risk and see the result on your own numbers.