The short answer
There's no honest number you could read here and apply to your case. The real range is so wide — from an assistant that handles one specific task to an automation layer that ties several systems together and runs largely unsupervised — that any single figure ends up being either an empty promise or an anchor that has nothing to do with what you'll actually pay. "AI agent" doesn't describe one thing: it describes an entire family of solutions with very different complexity, scope, and risk.
The honest answer isn't inventing a range that sounds specific — it's explaining what determines where your case falls within that range. That's what follows: the four factors that drive the price, the three typical project scales, and why the cost doesn't end the day the agent goes live.
What drives the price
Four factors drive the price more than anything else. The first is how many systems you need to integrate, and whether they have an API. Connecting to a tool that already exposes a documented API is straightforward; connecting to a closed system, an aging ERP, or a spreadsheet doubling as a database means building a bridge where none exists — and that costs real development time.
The second is data volume and how messy it is. A process built on clean, structured data is cheaper to automate than one carrying years of inconsistencies: field names that changed over time, formats that vary depending on who entered them, information duplicated across systems that never stayed in sync.
The third is how many business rules and exceptions the process has. A workflow with two or three clear rules gets automated fast. A workflow with fifteen exceptions — each depending on who the client is, what stage they're at, or how much money is involved — means mapping out every case before a single line of code gets written.
The fourth is how much human review the process needs. An agent that drafts something for someone to check before it goes out is simpler than one that acts without supervision on sensitive data or dollar amounts — that second case needs more validation, more error handling, and more testing before it ever goes live.
Three typical project scales
With those four factors in mind, here's how the three most common project sizes break down in practice.
| Scope | What it includes | What drives it up |
|---|---|---|
| One process, one agent | Automates one specific task end to end: sorting requests, generating a recurring document, following up on one kind of case. Usually touches a single system. | Messy input data, or a business exception that forces you to validate every case before it acts. |
| A workflow connecting 2 or 3 systems | The agent reads and writes across more than one tool — CRM, ERP, spreadsheets — so information moves on its own instead of someone copying it by hand. | Systems with no documented API, or business rules with many exceptions that have to be mapped one by one. |
| Several processes or departments | Several automations sharing infrastructure and sometimes data with each other, built as an operating layer rather than a one-off project. | Coordination across teams, data governance between departments, and how much human review each decision requires. |
This isn't an exhaustive list: where you land within a given scale depends on the four factors above.
The cost almost nobody mentions
Monthly operation isn't an optional add-on — it's part of the real cost of the project. An agent running in production needs monitoring to catch failures, fixes when an external system changes how it responds, and it consumes infrastructure and AI model tokens on every run. As a rough way to size it without numbers: for most projects, a year of operation ends up costing a meaningful fraction of what it took to build the automation in the first place — not a rounding error you can leave out of the budget.
A project delivered without an operating plan degrades over time: nobody notices when the agent starts failing quietly, and trust in the system erodes faster than it took to build. Budget for monthly operation from day one, not as a surprise three months in — it's the difference between an agent that's still running a year from now and one somebody has to rebuild from scratch.
What happens next
The above is a map, not a quote. The specific range for your case depends on your actual systems, your actual data, and your actual rules — and that only becomes clear once someone looks at it case by case. The fastest way to narrow it down without committing to anything is to talk it through: tell us what process you want to automate, which systems are involved, and how messy your data is, and in the free assessment we'll work out which of the three scales fits your process, with a real price range for your case.
If you already know which process you want to automate, book your free assessment and we'll go through it together.