Ask three suppliers what an AI agent costs and you may get three very different answers. That is not always because someone is overcharging. It is usually because they are quoting for different things. This article explains what you actually pay for, why quotes vary, and how to keep the budget under control.
The three parts of the cost
1. Build cost
This is the one-off work of making the tool: understanding your process, connecting to your systems, writing the agent's instructions, testing it on real cases and getting your team started. It is usually the largest single item.
What drives it up:
- Number of systems involved. Reading one inbox is simpler than reading an inbox, updating an accounting system and writing to a CRM.
- How messy the input is. Invoices from three suppliers are easier than invoices from three hundred.
- How much the agent is allowed to do on its own. Drafting for a person to approve is simpler than acting automatically, because automatic actions need more safeguards.
- How clear your process is. If the rules have to be worked out during the project, that takes time.
2. Running cost
Once the agent is live, it costs something to run:
- AI model usage. Most agents use a language model from a provider, billed by how much text is processed. It grows with volume: more emails or documents, higher cost.
- Hosting. The agent needs somewhere to run, usually a cloud service.
- Other services. For example, a document-reading service or a database for searching your files.
For a small company with moderate volumes, running costs are often modest compared to the build. But they depend entirely on your volume and the tools chosen, so ask any supplier for an estimate based on your actual numbers before you start.
3. Upkeep
Your business changes: new products, new prices, new suppliers, a new version of your accounting software. The agent needs to keep up. Upkeep covers monitoring that it still works, updating its instructions and documents, and fixing things when a connected system changes. Some companies handle this in-house; others prefer a monthly agreement.
Why quotes vary so much
When two quotes differ a lot, check whether they include the same things:
- Scope. Is it one task or several? Does it cover connecting to your systems, or only the AI part?
- Testing. Is the agent tested on real cases from your business before handover, or only on examples?
- Training and handover. Does your team learn how to use it and what to do when it is wrong?
- Running costs. Are they included, estimated, or not mentioned?
- Upkeep. What happens after handover? Is support included for a period, or billed separately?
- Ownership. Who owns the code and the instructions? Can you move to another supplier later?
A low quote that leaves out testing and handover can end up more expensive than a higher one that includes them.
Fixed price versus hourly
Both models are legitimate. They put the risk in different places.
Hourly billing means you pay for the time spent. It is flexible: if you change your mind mid-project, the work simply continues. The downside is that the total is unknown at the start, and unclear scope can make a small project grow. You carry the risk of overruns.
Fixed price means you agree on what will be delivered and what it costs before work starts. You know the total in advance. The supplier carries the risk of underestimating. The trade-off is that scope must be written down clearly, and changes mid-project need to be agreed separately.
For a first AI project, a fixed price with a tight scope is often easier to manage. You can say yes or no to a clear number, and you learn what works before committing to more.
Questions to ask any supplier
- What exactly will the finished tool do, and what is out of scope?
- What will it cost to run per month at our volume, and who pays those bills?
- How will you test it on our real cases before handover?
- What happens when the agent gets something wrong, and how will we notice?
- Where is our data processed and stored, and is it used to train AI models?
- What does upkeep cost after handover, and is it optional?
- Who owns the result, and can we take it to another supplier?
- If we want changes during the project, how are they priced?
How AgentSprint prices this
AgentSprint works on fixed prices, with the final price and scope agreed in writing before work begins:
- AI Audit, from €1,500 (1 week). A review of your workflows and a prioritised list of what is worth automating, with an estimate of effort and benefit for each item.
- AI Agent Sprint, from €6,000 (2–3 weeks). One working AI tool, in production use at the end of the sprint.
- Care Plan, monthly. Hosting, monitoring and improvements after handover, if you want them.
The price does not change during the work unless the scope is changed together. You can read more about what happens inside a sprint.
Keeping the budget under control
- Start with one task. Prove value on something small before expanding.
- Write down what success means before anything is built, so you can tell whether it worked.
- Do the maths with your own numbers. A worked example: say a team spends 10 hours a week sorting invoices and emails. Multiply that by what an hour of their time costs you, and you have a rough weekly figure to compare against the build and running cost. If the numbers do not add up, the project is not worth doing yet.
- Agree on running cost estimates up front, based on your real volume.
- Keep a person in the loop at first. It is cheaper to catch errors early than to fix their consequences.
A clear scope, a known price and an honest estimate of running costs will tell you more than any sales pitch.