Guide
How much does an AI agent cost for a business?
By Ruslan Rio · Updated · 4 min read
The short answer: a custom AI agent has an implementation cost, paid once, and a monthly cost that depends on how much work it does and how much review it needs. At Providence Crafts implementation starts at €799 excluding VAT, and the diagnosis that comes first is free. This guide explains what sits behind each part and what makes a project cost more or less.
The three parts of the cost
Any serious quote for an AI agent should separate three things. If a supplier gives you a single figure, ask what it covers.
- Implementation: understanding the process, building the agent, connecting it to your systems, testing it on your data and training your team. Paid once.
- Usage: what it costs each time the agent calls a language model, through an API or on your own servers, plus hosting. It depends on volume: how many invoices, emails or calls go through it.
- Running: reviewing every month where it fails, adjusting its instructions and adapting it when your systems change. This is what stops the agent degrading over time.
What drives the implementation price up
- How many systems need connecting. Reading an inbox is not the same as reading the inbox, querying the ERP and writing to the accounts.
- How closed those systems are. Software with an API connects in days; software without one means finding another route.
- The exceptions in the process. The more rules and special cases, the more there is to build and test.
- Where the model runs. A model on your own infrastructure gives more control over the data but takes more setting up than one used through an API.
- Whether there is voice. A phone agent has more moving parts — telephony, speech recognition and speech synthesis — than one that works with text.
That is why our figure says “from”: €799 is the starting point for an agent scoped to one process with few integrations. The exact price is agreed in the proposal, before work starts, and does not move along the way.
The monthly cost
Language-model prices have fallen sharply in recent years and, at the volume of a small or mid-sized business, usage is usually the small part of the bill. What weighs most each month is the running: someone watching what the agent does and correcting it.
We do not publish a monthly figure because it depends on volume and on how much review the process needs. It is itemised in the proposal, usage on one line and running on another, so you know what you pay for and why.
How to tell whether it pays off
The sum is simple: the hours a month your team spends on the process, times what an hour costs, against what the agent costs. But it has to be done with data, not by eye.
That is exactly what our diagnosis measures, and it is free: we walk through the process task by task, measure how long it takes and tell you whether an agent pays off. If it does not, we tell you that too.
Some savings do not show up as hours: invoices no longer overpaid, calls no longer missed, quotes sent the same day. They are worth counting, without inflating them.
What to ask before you sign
- Does the price include connecting the agent to my systems, or is that extra?
- What does it cost per month, with usage and maintenance shown separately?
- Is it tried on part of the process first, before it is opened up fully?
- Where is my data processed, and by which model?
- What does the agent do when it does not know something?
- What happens to the agent if we stop working together?
Agents that often come up in these quotes
Shall we look at your process?
The diagnosis is free: we measure where your team’s time goes and tell you what is worth automating.
Start the conversation