Home/Blog/How Much Does an AI Chatbot Cost? Build and Monthly Prices

// AI AUTOMATION

How Much Does an AI Chatbot Cost? Build and Monthly Prices

Three ways to pay for an AI chatbot, what each one costs at your volume, the costs that never appear in the quote, and how to tell whether it will pay for itself.

How Much Does an AI Chatbot Cost? Build and Monthly Prices

Intercom charges $0.99 every time its AI agent resolves a customer conversation. A support employee who costs you €12 an hour and answers the same question in four minutes costs €0.80. In Bulgaria, the chatbot with the most famous price tag can cost more per answer than the person it is meant to relieve.

That is why "how much does an AI chatbot cost?" has no single answer. The same bot can be the cheapest tool you buy this year or a monthly bill that grows faster than your sales. It depends on how you pay for it: a flat subscription, a fee per resolved conversation, or a one-off custom build with small running costs.

This guide gives real prices for all three, taken from the vendors' own pricing pages in November 2026, plus the indicative ranges we see for custom builds in Bulgaria. Vendor prices are in US dollars because that is how they are listed. Build prices are in euros.

The short answer

For a small or mid-sized business, an AI chatbot that answers questions from your own content costs roughly this:

  • Flat subscription (you set it up yourself): $40 to $500 a month, depending on how many messages you need.
  • Pay per resolution (AI inside a helpdesk): about $1 for each conversation the bot closes, plus a monthly fee for every staff seat.
  • Custom build by a studio: €3,000 to €15,000 once for most projects, then €100 to €500 a month to run.
  • Two-week proof of concept before a custom build: €1,000 to €2,500.

The custom figures are indicative ranges for the Bulgarian market, not a price list. The rest of this article explains what moves you up or down inside each range, and which of the three models is cheapest at your volume.

One thing first. All of these prices are for a chatbot that reads your documents and answers from them, the kind usually called a RAG chatbot. An old scripted bot with buttons and fixed replies is cheaper and does much less. If the difference is new to you, start with RAG chatbot vs regular chatbot.

What off-the-shelf chatbots cost

Off-the-shelf tools are the fastest way to get a bot on your site. You sign up, upload your documents or point the tool at your website, paste a snippet of code into your pages and you are live the same week. They bill in one of two ways.

Flat subscriptions with a message allowance

Chatbase is a typical example. Its pricing page lists three paid plans on monthly billing: Hobby at $40 a month with 700 message credits, Basic at $150 with 4,000 credits and Business at $500 with 15,000 credits. Paying for the year takes 20% off.

Read the allowance carefully. A credit is a message, not a conversation, and a customer rarely stops at one question. If a typical chat takes four or five replies from the bot, 700 credits cover about 150 conversations a month. So the $40 plan fits a quiet site, and a shop with a few hundred chats a month is really looking at the $150 plan.

What you get for that money is a bot that answers from your content inside a chat window. What you usually do not get on the lower plans: your own branding on the widget, more than one or two team members, and a connection to your order system or CRM.

Pay per resolved conversation

Helpdesk platforms price their AI differently. Intercom charges $0.99 per outcome for its Fin AI agent, on top of helpdesk seats listed at $29, $85 or $132 per person a month depending on the plan. You pay only when the bot closes the conversation without a human.

That sounds fair, and at low volume it is. The arithmetic changes as you grow:

  • 300 resolved conversations a month: $297
  • 1,000 resolved conversations a month: $990
  • 3,000 resolved conversations a month: $2,970

Seats come on top. The bill also rises with every improvement you make: the better the bot gets at closing conversations, the more you pay. For a company in London, where the same answer from a person costs several times more, $0.99 is a bargain. For a company in Plovdiv or Varna it is close to what the person costs, as the sum at the top of this article shows.

Per-resolution pricing still has a place. If you already run your support on one of these helpdesks and your volume is a few hundred conversations a month, switching on the built-in AI is the least work and a sensible first step.

Before you sign up for any off-the-shelf tool, ask three questions. What counts as one unit of the allowance: a message, a conversation or a resolved conversation? What happens when the allowance runs out? Can you export your conversations and your content if you decide to leave? The answers are on the pricing page or in the terms, and they tell you more about the real cost than the headline price does.

What a custom AI chatbot costs to build

A custom chatbot is built for your content, your systems and your site, and you own it. You pay once for the build and then a small amount each month to run it. Published estimates online run from a few thousand dollars to several hundred thousand, which tells you very little. Most of that spread is enterprise software for banks and telecoms.

For a small or mid-sized business in Bulgaria, the projects we see fall into four groups. Treat the figures as orientation. A real quote depends on your content and your systems.

Proof of concept: €1,000 to €2,500

Two weeks of work on a sample of your real documents. The result is a working bot you can question yourself, plus a short report on how often it answers correctly and where it fails. It is the cheapest way to find out whether your content is good enough before you commit to the full build. Often it is not, and it is better to learn that for €1,500 than for €8,000.

Question-and-answer chatbot: €3,000 to €6,000

A bot that answers from your documents, FAQ, policies and product pages in Bulgarian and English. It sits in a chat window on your site in your own colours, cites where each answer came from and passes the conversation to your team by email when it does not know. This covers most clinics, service firms and smaller online shops.

Chatbot connected to your systems: €6,000 to €15,000

Here the bot does more than quote documents. It looks up an order, checks stock or free appointment slots, creates a record in your CRM, or hands the chat to a staff member in real time. Every connection is separate work, and the price depends on how good the other system's API is. Boro, the property assistant we built for Houzez, belongs here: it turns a plain-language request into a search over live listings.

Several channels and several systems: from €15,000

The same assistant on the website, in Viber or WhatsApp and inside your internal tools, with access rights for different roles and connections to an ERP or accounting system. Few small businesses need this in the first year.

Four things decide where you land inside a group:

  • The state of your content. Clean, current documents are quick to load. Scanned PDFs, contradicting price lists and knowledge that lives only in people's heads all add work.
  • The number of integrations. Each system the bot reads from or writes to adds days.
  • How wrong it is allowed to be. A bot for opening hours needs light testing. A bot that explains insurance terms or medical preparation needs a proper test set and a strict rule for when to hand over to a person.
  • Channels and languages. A website widget in two languages is the baseline. Each extra channel has its own rules and its own approval process.

Ask what the quote includes. A complete one covers loading and cleaning the content, the chat window, the handover to your team, a test set of real customer questions with the share the bot must get right, and a period of corrections after launch. If testing is missing, the price is lower because part of the work has been left to your customers.

Our RAG chatbot service page describes how we scope this work. Every project starts with a two-week proof of concept.

What it costs to run each month

People expect the AI itself to be the expensive part. It is the cheapest line on the bill.

Language models are billed per token, a token being a piece of a word. Anthropic's price list puts Claude Sonnet 5.5 at $2 per million tokens going in and $10 per million coming out. A typical chatbot answer sends the model about 3,000 tokens (the instructions, the passages found in your documents and the conversation so far) and gets about 300 back. That is less than one cent per answer. A conversation of five answers costs about five cents, and a thousand conversations a month cost about $50.

Smaller models cost far less. Claude Haiku 5.5 is listed at $0.10 and $0.50 per million tokens, a twentieth of the Sonnet price, and it handles simple questions well. Many bots send easy questions to the small model and keep the larger one for the hard ones.

The same price list shows which way prices are moving: the older Claude Sonnet 4.6 is listed at $3 and $15, the newer Sonnet 5.5 at $2 and $10.

The full monthly bill for a custom chatbot looks more like this:

  • Model usage: €20 to €300, depending on volume and which model answers.
  • Vector database and hosting: €50 to €200.
  • Monitoring and maintenance: €100 to €300 if a studio reviews the conversations, fixes wrong answers and keeps the integrations working.

Most small businesses end up between €150 and €500 a month in total. The important part is how that number behaves. If your conversations double, the model line doubles and the rest stays where it was. On per-resolution pricing, the whole bill doubles.

Set the five cents against the $0.99. The difference is what the vendor charges for the software, the helpdesk and the support around the model. That is a fair thing to charge for. It also means that above a certain volume you are paying every month for software you could have paid for once.

The costs that are not in the quote

Whichever route you take, some costs sit outside the price you were given. None of them is large. Together they explain why a bot that worked well in the demo gives wrong answers six months later.

  • Preparing the content. Someone has to collect the documents, remove the outdated ones and settle which of two contradicting price lists is right. Allow a few days of your own team's time. A bot built on messy content gives messy answers, at any price.
  • Keeping the content current. Every new product, changed delivery fee or promotion has to reach the bot. Somebody must own that task, even if it takes an hour a week.
  • Reviewing conversations. In the first month, read what customers ask and how the bot replies. This is where you find the questions nobody expected. Later, a sample every week is enough.
  • Overage and add-ons on subscriptions. Extra message credits, extra team members and removing the vendor's logo are usually paid extras. Check what happens when you pass the allowance: some tools stop answering, others keep going and bill you.
  • Messaging channel fees. WhatsApp and Viber charge businesses for conversations on their networks. These fees go to the channel and are separate from the chatbot.
  • The human handover. Somebody still answers the questions the bot passes on. If nobody picks them up within a reasonable time, the bot has only moved the queue.

We wrote about the upkeep side in more detail in AI chatbot for customer support.

Subscription or custom build: which is cheaper for you

Over two years the choice comes down to three things: how many conversations you have, whether the bot needs your systems, and how much of your own time you can give it.

A subscription is the cheaper option when:

  • you have up to a few hundred conversations a month;
  • the bot only needs to answer from documents and web pages;
  • someone on your team is happy to set it up and look after it;
  • you are still testing whether customers will use a chatbot at all.

A custom build is the cheaper option when:

  • volume is in the high hundreds or thousands a month, where per-message and per-resolution fees add up;
  • the bot has to check orders, bookings, stock or customer records;
  • Bulgarian has to be as good as English, including product names and the way your customers write;
  • you want the conversations and the knowledge base stored where you decide, which matters under GDPR for clinics, insurers and financial firms.

Here is a worked comparison at 1,000 resolved conversations a month. On per-resolution pricing at $0.99, the AI alone costs about $990 a month, or roughly $23,800 over two years, before seats. A custom build at €6,000 with €300 a month in running costs comes to €13,200 over the same two years. At 200 conversations a month the result reverses: about $4,750 on per-resolution pricing against the same €13,200. With these figures the crossover sits at roughly 500 to 600 conversations a month.

There is a middle path, and it is the one we recommend most often. Start on a cheap subscription for two or three months. Find out what customers ask, how many of them use the bot and which questions it cannot answer. Then, if the numbers support it, build the custom version with that knowledge in hand. Those few months of subscription fees are cheap market research.

Will it pay for itself?

Work it out before you buy anything. You need four numbers from your own business:

  • how many customer enquiries you get a month, across chat, email, phone and social media;
  • how many minutes an average answer takes;
  • what an hour of the person answering costs you, salary and social security included;
  • what share of the enquiries are routine questions with a written answer somewhere.

An example with round numbers. A shop gets 600 enquiries a month and each takes four minutes, which is 40 hours. Half are routine: delivery times, returns, sizes, order status. If the bot handles those, it frees 20 hours a month. At €12 an hour that is €240.

€240 a month does not justify a €6,000 build on its own, and anyone who tells you otherwise is selling something. In Bulgaria, saved staff time alone rarely pays for a chatbot at this size. Three other things usually do:

  • Sales outside working hours. A customer who asks at 22:30 whether a product fits and gets an answer straight away often buys. A customer who waits until morning has often bought elsewhere.
  • The hire you put off. When enquiries grow by half, the bot absorbs the growth and you do not need a second support person. That is a full salary, not 20 hours.
  • Speed. People who get an answer in ten seconds ask more questions, and people who ask questions are closer to buying.

If none of the three applies to you, and your enquiries are few and all different, a chatbot will not earn its cost. A better FAQ page and a clear contact form will do more for less.

The verdict

An AI chatbot costs between $40 a month and €15,000 up front, and the right figure for you depends mostly on volume. Under a few hundred conversations a month, buy a subscription, set it up yourself and treat it as a test. Above that, or as soon as the bot needs to see your orders, bookings or customer records, a custom build with its own running costs is cheaper over two years and gives you something you own.

Be careful with per-resolution pricing. It was designed for markets where a human answer costs several times more than it does here. In Bulgaria it can cost about as much as the person, and it gets more expensive the better the bot performs.

Whatever you choose, pay for the small test first: a month on a subscription or a two-week proof of concept. A chatbot fails when the content behind it is poor, and a test shows you that for very little money. Our pricing page explains how we quote: a fixed scope and a fixed price, in writing.

Frequently asked questions

How much does an AI chatbot cost per month?

Subscription tools start at about $40 a month and reach $500 for higher message volumes. A custom chatbot costs €150 to €500 a month to run after the build, covering model usage, hosting and maintenance. Tools priced per resolution cost about $1 for each conversation the bot closes.

How much does it cost to build a custom AI chatbot?

For a small or mid-sized business in Bulgaria, expect €3,000 to €6,000 for a chatbot that answers from your documents and €6,000 to €15,000 if it has to connect to your order system, booking calendar or CRM. A two-week proof of concept costs €1,000 to €2,500.

Is there a free AI chatbot for a website?

Several tools have free plans, and they are fine for trying the idea. They limit the number of messages, show the vendor's branding and rarely connect to anything. For a business that depends on the answers being right, plan to pay.

Why do AI chatbot prices vary so much?

Because the name covers very different products. A scripted bot with buttons, a subscription bot that reads your website and an assistant connected to your order system are all called chatbots. The price follows the integrations, the volume and how much testing the answers need.

How much does the AI model cost per conversation?

Very little. At current prices for a mid-range model, a conversation of five answers costs about five cents, and a small model costs a fraction of that. On most custom chatbots the model is the smallest line on the monthly bill.

Can an AI chatbot replace a support employee?

It does not replace one. It takes the routine questions, which in most businesses are about half the total, and leaves the complicated ones to people. The saving shows when your enquiries grow and you do not have to hire the next person.

How long does it take to build an AI chatbot?

A subscription tool can be live in a few days. A custom chatbot takes two weeks for the proof of concept and usually four to eight weeks more for the full build, depending on the integrations and on how quickly the content is ready.

Want a real number for your chatbot?

Tell us what your customers ask most often and roughly how many enquiries you get a month. We'll reply within three business days with a written proposal, a fixed price and a plan for a two-week proof of concept.

Start the conversation →