How much does an AI agent or chatbot cost to build?

A chatbot on your website, an assistant that answers WhatsApp or an agent that picks up the phone and books appointments: they all fall under the same question, yet they are very different projects. The price depends on what the system needs to know, what it may do, which channels it works on and how well it has to handle questions that fall outside the script.

This guide contains no project amounts. An amount without a scope is especially tricky with AI, because the work mainly sits in knowledge, rules and testing, and those differ a lot between businesses. What you do get: the difference between a chatbot and an agent, the factors behind the price, the costs that continue every month, the hidden items and a list of questions to compare quotes with.

What is the difference between a chatbot and an AI agent?

A classic chatbot works with buttons and fixed scripts. An AI chatbot answers questions in plain language from your knowledge. An AI agent goes a step further: it holds the conversation and then handles it in your systems, such as booking an appointment, making a reservation or preparing a quote. The more the system does by itself, the more has to be designed and tested.

Want to be clear first on what fits your question? Read AI chatbots for SMEs and AI in customer service.

What determines the price of an AI agent?

These are the factors that set the size of an AI agent project. The more of them sit at the heavy end, the bigger the project.

Which choices make an AI agent more or less expensive?

The cheapest agent is not the one that costs least to build, but the one that starts with what takes the most work off your hands and grows from there.

• Cheaper: starting with one channel and the questions that come up most often, then adding channels and tasks on the same knowledge.
• Cheaper: knowledge that is already written down: current terms, prices, opening hours and examples of good answers.
• Cheaper: answering questions and handing over first, and adding actions only once the answers are right.
• More expensive: many actions across different systems, or systems without a good connection.
• More expensive: the phone as a channel, with a voice that has to respond quickly and naturally.
• More expensive, but often worth it: strict approval rules in the first weeks. More rules mean more work, but also more certainty.

How to set up those approval steps in practice is covered in human oversight in AI workflows.

Which costs continue every month?

An AI agent also costs something every month, and that part depends more on usage than with a website. These are the kinds of costs that continue.

• Language model usage: every conversation takes computing power. How much depends on the number of conversations and how long they last.
• Telephony: a phone number and call minutes, if the agent answers the phone.
• Messaging channels: a business connection to a messaging channel can carry its own cost per conversation, under that channel’s terms.
• Hosting and monitoring: the environment the agent runs in, with a log of every conversation.
• Tuning and updating knowledge: reading along, improving answers and adding new information when your offer or terms change.
• New tasks: expanding to new channels or actions once the agent has proven itself.

At Vorlyo the AI assistant is part of the software per industry. Your software is a one-off investment, fully set up for your business. After that you pay a fixed monthly licence for hosting, updates, support and tuning. The price follows the intake and the call. What the assistant does is on the page about the AI assistant.

Where are the hidden costs of an AI agent?

In AI projects, the unexpected costs rarely sit in the technology. They sit in knowledge, rules and time.

• Knowledge that is not written down: if answers only live in your team’s heads, someone has to write them out first.
• Outdated information: an agent that quotes old prices or terms costs trust. Someone has to keep the knowledge current.
• Exceptions: the question that is just slightly different, the customer with a special arrangement. Every exception needs a rule or a handover.
• Growing usage: more conversations mean more usage costs. Ask how they grow and how you can track them.
• Systems without integration: if your calendar or CRM lacks a good connection, a workaround means extra build work.
• Privacy arrangements: conversations contain customer data. Agreements on processing and retention should be in place up front.
• Your team’s time: in the first weeks your team approves actions and gives feedback on answers. Plan for that.

How do you compare AI agent quotes?

With AI, quotes quickly look alike, because everyone promises “a smart assistant”. The difference is in what happens when the agent is unsure. Ask every party the same questions.

How does the intake work at Vorlyo?

In the intake you tell us which questions keep coming back, on which channels and what the assistant may do. We reply within one working day. After that it follows fixed steps.

• 1. Intake: your questions, channels and the tasks the assistant may carry out.
• 2. Call with a demo: we show how the AI assistant in the software for your industry works, and then the price follows: a one-off investment in the software and a fixed monthly licence.
• 3. Setup: we feed the assistant your documents, terms and examples and test with real questions and awkward exceptions.
• 4. Live and tuning: first with approval on everything, then increasingly independent, while our team reads along and improves.

How the AI assistant works and what it can do is on the page about the AI assistant.