Does a business need AI?

Whether a business needs AI is usually decided after someone else's video: a robot answers customers, a robot calls, a robot wakes the old database, and it also replaces the team. Meanwhile yesterday's "how much" is still sitting in your own Direct, and the manager got to it at lunch.

AI does not replace a sales team. There are four jobs. The first meets a new message. The second writes to people who already came and went quiet. The third listens to calls the managers already made. The fourth, the one that calls by itself, we do not sell yet as a finished call centre. Piling that into one button called "turn on AI" gets expensive and annoying.

When a business does not need AI

If a few messages come in a day and the owner sees them, the bot is extra. A client writes "aka, do you have the size?", and gets an answer from a phone in ten minutes. Putting a model, a price list, and handoff rules between them does little.

The same with a database of thirty cards. "Revival" is a loud word for a list the manager already remembers. Calling is worse: hiring a voice to phone twenty people costs more than calling them yourself.

Maybe for a wholesaler with three regular buyers AI really is unnecessary. Then again, no. Even there a forty-second voice note lands in Telegram with a list of items, and the manager listens to it a second time from a job site because he did not write it down the first time. Calling is still too early for them.

An AI chatbot: what it is for in sales

A chatbot is for the place where the same questions arrive faster than anyone can answer them: price, availability, address, hours, delivery. People write in Instagram, WhatsApp, Telegram, and on the site. Half an evening on a commercial account looks like "how much" and "do you have the size?". Long negotiations are rare in that pile.

An evening in Direct looks like this. At 21:10: "grey sneakers, size 42, delivery to Chilanzar?" The price and the size are on the list, the district is in the delivery terms. The bot names the price and the time, and puts the size on the card. At 21:14 another chat: "the courier did not show up yesterday, give the money back". That is a manager. The bot does not promise a refund that is not in the script. A third chat, a voice note: someone lists five items while walking. In the morning the card has the text of the list. The manager does not replay the street noise.

Whatever the client already said lands on the card: size, branch, pickup, district. The manager opens the chat and sees a short summary. "Hello, how can I help" after ten messages sounds like nobody read them.

The bot calls a person in clear places. "Get me a manager". A complaint, a refund, the courier did not come. An invoice, a contract, "let's close it". A discount that is not on the price list. A question the materials do not answer. If it misunderstood twice, one apology, the line "I am an AI assistant", and the handoff. No new product selection at that moment.

At night it does not promise "a manager will reply in five minutes" if the schedule says the person is asleep. The time comes from the working hours you set. A "how much?" comment under a post can be closed with a line from the price list, and the conversation moved to Direct. A complaint is better not unfolded in the feed: strangers read it. Where that line sits in Instagram we wrote up separately: what the bot answers itself, and what it hands to a manager.

The client picks the language. Russian, Uzbek, Tajik, and English, often two of them in one message. The bot answers in the one they wrote in.

A separate fear shows up in demos before the price list does. "It will learn our customers and leave for a competitor". The conversation stays in your CRM. The model only receives the text it needs to answer, and it is not trained further on your chats.

You pay for the chatbot separately from the CRM plan. A dialogue is the conversation with one client, however many messages are inside it. You can take a simpler model, DeepSeek handles consultations fine, or ChatGPT if you need it to stick to the script more closely. On the free period the bot is set up on your price list, and you see the answers on live messages before you pay. How it looks on the screen: the AI chatbot.

Reviving the database

Over the years a CRM collects what you already paid advertising for. They wrote in Direct, asked the price, disappeared. They bought once and have been quiet for six months. The manager does not go back to them: there are fresh leads, and the plan is on those. The old database is always "when there is a free minute".

Reviving the database, we call it waking the base, is about these people. You take a segment: no reply for more than three months, or the refusal was about price, or the deal was for a certain amount. The filter is built from fields on the card, not from the mood of "let's write to everyone".

Then you need a reason. A new delivery, a different price list, a course that ran out, time for a repeat visit. The same text to a thousand numbers looks like spam, and in WhatsApp it ends in complaints fast. The agent writes in the messenger where you already talked to that person: Telegram, WhatsApp, or Instagram. Not one paragraph for everybody. It remembers how the last conversation ended.

Before anything is sent, you can see the campaign. You approve the list and the sample messages. Without that confirmation nothing goes out. You can stop the campaign. This is the detail people ask to see twice on a demo: they remember a contractor who once "just sent it".

If the client answers with a question, the bot continues from the price list and does not wait for the manager to be free. When the interest is back, a task appears on the card. The manager gets the people who replied. He does not call a thousand silent contacts by hand.

On a few dozen cards the idea is weak. It starts to make sense from a few hundred contacts. Below that, it is more useful not to lose fresh messages with a chatbot than to wake people there are so few of. To see who in your database is even worth touching: the database revival agent.

The AI that calls

The request "let the robot call them" comes up often. Calling runs into people: find them, train them, pay them, replace them after they quit. Half the conversations end on "not relevant", and spending a strong seller on those is a waste. An auto-announcer that says "press one" gets dropped in the first second.

The voice agent we are building calls itself, introduces itself, speaks from a script, and hears the answer. If someone is interested, it transfers them to a manager. For the rest it sets a status on the card, with the transcript next to it. The list of who to call is approved before the launch, the same way a database campaign is.

This module is not for sale now. It is in a closed test: we connect a limited number of companies by hand and keep writing the scripts. We do not sell it as a finished AI call centre, and we will name the price at launch. The request for early access is here: the AI autodialer.

What already works with calls is a different job. AI listens to recordings the managers already made, and marks where they skipped a question and where they gave a discount straight away. The manager does not have to listen to the whole month to see the hole. That is call review, not autodial. Mixing them up is expensive: one can be turned on for the calls you already have, the other cannot be bought as a product yet.

Where to start

On demos people usually poke the chat, not the calling. The hole in Direct is visible the same evening: unanswered messages sit in a list. They ask for the database next, once fresh leads are no longer getting lost and the old thousands are just lying there. A voice call makes sense last, and only when there is someone to take the transfer. Until that module exists, call review is the useful thing to do with calls.

You can take the agents one at a time. They are not included in the CRM plan and are billed separately, on any plan, from Basic to VIP. Messenger channels are included in the CRM plan. Setting the chatbot up on your materials takes one or two days, not a quarter with a programmer.

These jobs, and the note that calling is not for sale yet, are on one page: AI in the CRM.

If "how much?" piles up while you are on a job site, start with the chatbot. If the database has thousands of silent contacts and nobody to write to them, look at revival. The robot that calls instead of a call centre can wait until it is actually for sale.