What is a chatbot and how does it differ from an AI manager

To answer the question of what a chatbot is in short: it is a program that chats with a person in a messenger or in a website chat window instead of a live operator. There are two main types. A rule-based bot responds with buttons and pre-prepared phrases. An AI-based bot understands free-form text thanks to a language model. An AI sales manager goes further: it checks the catalogue, answers questions about the product, and guides the conversation all the way to an order. Below, all three are broken down using a single example: a cosmetics shop selling via Instagram Direct and one manager on shift.

A chatbot is a program that is already sixty years old

The first chatbot was called ELIZA. Joseph Weizenbaum wrote it at MIT and described it in an article in Communications of the ACM in January 1966. The program searched for given words in a person's utterance and, following rules, rebuilt the sentence into a reply: to “you are a great help” it would ask “why do you think I am a great help?” (this is how Wikipedia describes ELIZA's work). There was no understanding there. There were templates and substitution.

Surprisingly, that is exactly how most business bots worked until 2022. The channels and buttons have changed, but the principle remains the same: recognise a word, output a template. Language models broke this pattern, and now under the word “chatbot” lie two completely different technologies. Confusion in terminology costs money, so next we'll look at them in turn.

How a scripted chatbot works

Imagine a menu in Instagram Direct: “Delivery”, “Payment”, “Availability”, “Contact manager”. A female customer taps “Availability”, the bot asks for the article number, finds the row in the spreadsheet, and replies. This is the very answer to the question of what a first-generation chatbot is: a tree of branches that someone has drawn in advance. Every response is written by a human, every transition is anticipated.

Such a bot performs well as long as the customer follows the preset branches. It breaks down at the very first live question: “Is this cream available in a smaller size and is it suitable for dry skin after 40?”. The tree doesn't have such a branch, and the bot replies “I didn't understand, please select a menu option”. The female customer types “manager” and waits. At night, she waits until morning.

At Anthropic, there is a precise word for such systems: workflow. This is a scheme where “language models and tools are orchestrated through predefined code paths” (Anthropic, “Building effective agents”, December 2024). The path is predetermined, which is both an advantage and a limitation. A power grid operator bot that shows outage schedules, or a bank bot that provides a balance by card number, are set up in the exact same way. For customer support, this is ideal. For sales, where questions are not repeated verbatim, this is not enough.

What is an AI chatbot and how does it work

A language model bot reads messages in their entirety. It will understand a question about a dry skin cream for over-40s on the first attempt and respond with a sentence rather than a menu item. By the definition in that same Anthropic article, a system becomes an agent when “the model itself manages its process and use of tools”. That is, it decides what to do: look in the catalogue, clarify the skin type, suggest two options. How such systems are structured and where the limit of their freedom lies is discussed in the article about AI agents.

There is risk in this freedom. An unrestricted model will readily invent a product composition that does not exist, or promise one-hour delivery to a city where you do not deliver. Therefore, only one configuration works for business: the model responds exclusively from the knowledge base, and when there is no confident answer there, it honestly states this and calls for a human. Without these two rules, an AI bot is more dangerous than a button-based one.

How an AI sales manager differs from a chatbot

An AI chatbot and an AI sales manager are built on the same technology. The difference lies in the task. The bot responds. The manager sells, meaning it drives the dialogue to an order and logs it wherever you keep your records.

Sign	Script bot	AI chatbot	AI Sales Manager
Understands free-form questions	no	yes	yes
Knows your product range and stock levels	as a separate table, manually	if you download	reads a catalogue from a website, PDF or Excel and updates prices and stock levels
Creates an order	no	rarely	yes, in the CRM
Fulfilling the complex question to a person	“message the manager”	sometimes	to the operator along with the chat history
Where does it get its facts from	from the script	from everywhere, if not restricted	solely from the knowledge base

The right-hand column describes how it works in Marichka AI: a reply in 10–30 seconds at any time, a catalogue from a website page or file, orders placed straight into the CRM, and an unclear question going to an operator with the context. This setup is not tied to a single product, and that is precisely why you should demand it from any solution being offered to you.

Demanding it is also worth doing because there is a lot of renaming on the market. In June 2025, Gartner called this “agent washing”: vendors take an assistant, RPA or a standard chatbot and call it an agent without adding agentic capabilities. According to Gartner's estimate, out of thousands of “agentic AI” vendors, about 130 actually possess such capabilities (Gartner, 25/06/2025). The check takes two minutes. Write a question in the demo that is not in any FAQ: “I need a cream for a child with atopy, which is under 600 UAH and when will it be in Dnipro”. Then ask to place an order. If a menu is returned in response, you have a scripted bot in new packaging.

Where chat bots are used

In Ukraine, chatbots live where people live: in messengers. According to a Gradus survey in September 2025 (1,100 respondents, cities of 50,000+, ages 18–60), 75 % of Ukrainians get information through messengers, and 44 % spend more than three hours a day in them. Telegram is opened weekly by 92 % of those surveyed, and Instagram by 67 % (Gradus via MediaSapiens).

Hence the map of use cases. The first layer: reference bots for banks and utility providers, where requests are simple and repetitive. The second layer is noticeably smaller in terms of the number of bots, but larger in terms of money: shops and service businesses, from clinics to salons, for which Instagram Direct and Telegram chat have become the main place where a customer asks questions and decides whether to buy or not. Here, showing a schedule is no longer enough. You need to answer about ingredients, compare two products, suggest a size, and not lose the person while they are thinking.

What is a chatbot for small business used for

Let's go back to the cosmetics shop with a single manager. Her shift ends at nineteen hundred. Messages keep coming until midnight and start again at seven. Exactly how many of them would turn into orders if the reply came immediately depends on the niche and the average order value, and only your own monthly statistics will give you an honest figure. But it is already clear from the description of the situation why a chatbot is needed: so that the first message gets a reply in the very same minute, and in the morning the manager sees not thirty “Good afternoon, is this in stock?”, but five placed orders and three dialogues where a human is actually needed.

One example from a public case study. The food delivery service Box Catering integrated an AI manager and, over the course of its operation, generated over 500 leads with a processing time of up to 10 seconds and 500,000 UAH in revenue (the case study is published on the Marichka AI website). The figures from other niches will be different, but the mechanism is what matters: the speed of the first response and the fact that the bot doesn't forget to follow up with a customer who has gone quiet. The second part is usually called a follow-up. And the full list of tasks that the bot takes off a manager in a small shop is gathered in the text about business chatbot.

All of this can be checked without choosing a platform or using comparison tables: open your direct messages and write down the last twenty customer questions verbatim, then divide them into two piles – those that can be answered by a four-button menu, and those that require product knowledge. If the second pile is larger, a rule-based bot won't cope here, and the conversation moves to an AI manager.