Chatbot or AI Assistant: What Actually Separates Them

On a website both look identical: a round button in the corner, a window, someone answering. The difference only shows when a customer asks something nobody expected. One of them asks you to rephrase; the other answers. That difference is where both the price gap and the gap in results come from.

Let us go through both honestly, including the cases where the expensive option is not needed.

A scripted chatbot: a tree of prepared answers

There is no artificial intelligence inside such a bot, although it is often sold as exactly that. There is a tree: question, answer options, next question. Everything the bot says was written by hand in advance.

Recognition works on keywords. If the script contains "opening hours" and the customer writes "are you open on Sunday", there is no match, and the bot replies that it did not understand or offers a menu.

Where it genuinely shines

• Booking an appointment. Choose a service, a specialist, a time - an ideal task for buttons. Buttons are even better than free text here: fewer mistakes.
• Guiding along a short path. Processing a return, checking an order status, picking a collection point.
• Capturing a contact. Name, phone, convenient time - then hand over to a manager.

Where it breaks

On any question you did not anticipate. And anticipating everything is impossible: real people phrase things in endlessly varied ways. The second problem is sprawl. A script with forty branches becomes unmaintainable: six months later nobody remembers why a branch says what it says, and edits start breaking neighbouring ones.

Price from published rates: roughly from 30 thousand roubles for a simple version, from 90 thousand for a bot with sales logic. Running costs are near zero.

An assistant on a language model: understands and composes

Here there is a real language model inside. It understands a question asked in any form and composes an answer rather than picking one. The customer asking about Sunday gets a proper reply.

But there is a fundamental point that determines everything. The model on its own knows nothing about your business: not your prices, not your terms, not the fact that you do not work with companies. If you do not give it that, it will answer in generalities or, worse, invent things. So the main engineering work in a deployment is not "connecting a model" but assembling your knowledge base and teaching the assistant to search it and answer only from what is there.

The question that separates an experienced vendor from a beginner: "where does the assistant get facts, and what does it answer when a fact is missing". The right answer is that it says so honestly and hands over to a human. "The model will figure it out" means that in a month it will promise a customer a discount you do not offer.

Where it is stronger

• Free-form questions. The customer writes as they think, not as you anticipated.
• Long conversations. It keeps context: if a two-room flat was mentioned at the start, that is taken into account later.
• Preparing the inquiry. It clarifies details, so in the morning the manager receives substance rather than a bare contact.
• Change. Updating terms is an edit in the knowledge base, not a rebuild of a decision tree.

Where it is weaker

• Strict procedures. Where an exact sequence of steps matters, freedom of phrasing hurts: buttons are more reliable.
• Price. From 120 thousand for a basic deployment and noticeably more with integrations, plus a monthly charge for model calls.
• The risk of invention. It does not disappear entirely, it gets constrained. Ask exactly how.

How to choose: three questions to yourself

1. Do your customers ask the same things? Collect the last hundred inquiries. If nine in ten fit a dozen typical questions, a scripted bot is enough. If every second question is its own, you need an assistant.
2. What has to happen after the conversation? If it is booking a slot, both will manage. If it is understanding the task and preparing the manager, only an assistant will.
3. Who will maintain it? A script needs someone to add branches for new questions. A knowledge base needs someone to keep it current. If there is nobody for either, any solution goes stale within six months.

How to test a solution before buying

A demo always looks convincing: the vendor shows scenarios they prepared themselves. What you should test is not those but the behaviour at the edges. Ask for access to a test version and put five questions to it.

1. Something off-topic. Ask about what the company does not do. A good answer: "we do not do that, but we do this". A bad one: a cheerful promise to help.
2. A question with typos and no punctuation. That is how half of all real people write. A scripted bot usually gets lost here.
3. Two questions in one message. "How much is it and when can you start" is a typical phrasing that breaks simple solutions.
4. A follow-up three messages later. Ask about price, discuss something else, then say "back to my first question". This tests conversational memory.
5. A question with no possible answer. Ask about a service or condition that does not exist. This is where you find out whether the assistant admits ignorance or invents.

The fifth question matters more than the other four. An assistant that invents confidently is more dangerous than a bot that honestly did not understand: the customer will believe it and arrive with a claim based on a promise you never made.

How long each takes to launch

• Scripted bot: from a few days to two weeks. Most of the time goes not into the technology but into inventing and writing every branch.
• Assistant: from three to six weeks. The largest part is assembling the knowledge base, and that part sits with you, not the vendor.
• Both need a month of running in on real inquiries. Not skippable: that is where you learn what people actually ask.

Five situations and what to take in each

Barbershop, salon, workshop: booking a time

Take a scripted bot. The questions are typical, there is one action, and buttons are more reliable than free text. An assistant here costs more without winning anything.

A clinic: booking plus questions about treatment

A hybrid. Booking on buttons, questions through an assistant, but with a hard ban on any medical advice and mandatory handover to a human at the first sign of urgency.

Construction, renovation, complex services

An assistant. Every inquiry is different, there are many questions, and the real value is that in the morning the manager receives a full picture of the task rather than a line saying "interested in a renovation".

An online shop with a large catalogue

An assistant with access to the catalogue and stock levels. The key here is not conversation but the link to availability: a bot cheerfully selling what is out of stock produces losses, not revenue.

Wholesale and B2B

Careful with both. There are few inquiries and each is valuable, so automation is justified only on first contact: record who wrote and about what, then hand over to a human immediately. Trying to run such negotiations automatically is a way to lose a large client.

Three costly misconceptions

"We will take the cheap one and replace it later"

Replacing means not only new development but a knowledge base rebuilt from scratch, new integrations and retraining your team. If your questions clearly exceed what scripts can handle, the cheap option is not a first step but money thrown away.

"The assistant will learn from our conversations"

In most deployments it will not. It answers from a knowledge base maintained by people. Training on transcripts is separate, expensive work and is usually not done. If you are promised self-learning, ask who checks what exactly the system learned.

"Customers dislike bots"

Customers dislike useless bots. A fast, substantive answer at eleven at night is noticeably more welcome than silence until morning - the same point is made in the piece on lost inquiries.

The hybrid that beats both

In practice the strongest setup is a combination: the assistant handles free conversation and, at the moment of action, hands control to a rigid script.

The customer writes whatever they like, the assistant understands and replies. When it comes to booking, buttons take over: service, date, time. Freedom of phrasing then does not interfere with precision where mistakes are expensive. Ask a vendor whether they can do this - it is a good indicator of maturity.

What not to expect from either

• That it will sell for you. It removes routine and stops inquiries being lost, but the decision to buy is made by a person, and in complex deals it is made in conversation with a person.
• That it will replace a team. It replaces the first line: answers to typical questions and gathering details.
• That it will earn for you. If the offer is weak, a fast answer only speeds up the refusal.

In short

A scripted bot is a tree of prepared answers: cheap, predictable, breaks on the first unanticipated question. An assistant on a language model understands free speech and answers from your knowledge base: more expensive, more flexible, and it requires that such a base exists.

The choice is driven not by what is more modern but by how varied your customers questions are. The price ranges and the scope of work are covered in the pricing breakdown, and what happens to an inquiry afterwards in the piece on connecting to a CRM.