What an AI agent costs a Czech company
20 Jul 2026 · David Sak · 6 min read
"How much does an AI agent cost?" is the first question on every intro call, and the answer "it depends" deserves unpacking. Here's an honest breakdown: what really determines the price, what operations cost and when the investment pays back. The numbers are indicative ranges from my practice, not a price list; you'll get an exact price only after I assess your brief.
Three levels of complexity (and three price leagues)
1. An agent that answers. A chatbot over your documentation, FAQs, policies, product sheets. Answers with source citations, escalation to a human. The smallest meaningful deployment; with me from CZK 25,000, typically in the lower tens of thousands depending on the scope of documentation and channels.
2. An agent that acts. Besides answering, it performs actions in your systems: opens a claim in the ERP, sends a summary to Slack, creates a ticket. The price grows with the number of integrations, each connection means authentication, permissions and testing. Typically upper tens of thousands.
3. An agent in a regulated environment. Finance, healthcare, law: private infrastructure, document-level access control, an audit trail, possibly an open-source model running entirely on your side. Here we're talking upper tens to hundreds of thousands, proportionate to what's at stake.
What really drives the price up
- The state of your data. A tidy wiki = a fast start. Knowledge scattered across e-mails and colleagues' heads = consolidation first, agent second.
- Number of integrations. Every extra system (CRM, ERP, helpdesk) adds work, and systems without an API multiply it.
- Security requirements. An API with no data retention is cheaper than a model in your own cloud. Compliance requirements can double the budget.
- Channels. A web widget is the baseline; Slack, e-mail and telephony each add their own layer.
Operations: what gets forgotten
An agent isn't a one-off purchase. Budget for monthly costs on API tokens (for typical company traffic, hundreds to low thousands of crowns), hosting and monitoring. Above all, maintenance of the knowledge base, the agent answers only as well as the documents it reads are current. That's why I offer ongoing care: capacity for changes, a monthly answer-quality report and development based on what production shows.
When it pays back
A simple calculation I do on every intro call: how many queries a day the team handles × minutes per query × hourly rate. If support handles 50 repetitive queries a day at 5 minutes each, that's over 20 hours a week, and an agent resolving 70-80% of them on its own pays for itself within months. If you handle 5 queries a day, I won't sell you an agent, because it won't pay back; more in the article When an AI agent pays off.
How to get an exact number
Write me two sentences about what the agent should do and where that knowledge lives today. I'll get back to you within 24 hours, and after a 30-minute call you'll receive a written proposal: scope, price and timeline. The price then holds, no extra charges without your approval.