Guide
How much does a custom AI agent cost?
Updated on
There is no list price for a custom AI agent. Its cost depends on the scope given to the agent, the number of tools to connect, the quality of the data, the volume to handle and the follow-up after go-live. The quote comes out of an audit, the only price we publish, from €800 excl. VAT.
Why is there no list price?
Because a custom agent is not a product you take off a shelf. It is design, connection and tuning work, done for one particular company.
Standard software has a price because it is the same for everyone. A custom agent is built from your tasks, your tools and your documents. Two projects both called “meeting report agent” may have almost nothing in common: one writes to a single, well-documented tool, the other has to feed three applications, one of which was never meant for it.
A price quoted before anyone has seen how you work is likely to be the price of someone else’s project. So this guide contains no amounts, with one exception that you will find further down. What it does explain is what pushes the bill up or down, so that you can read a quote with your eyes open.
What makes the cost of an AI agent vary?
Five factors, roughly in this order of importance.
Scope
This is the first lever. How many tasks does the agent take on, and how far does it go with each?
An agent that prepares draft meeting notes for review takes less work than one that writes them, updates three records and schedules what comes next. The level of autonomy matters too: the more the agent acts on its own, the more exceptions, checks and tests have to be planned for.
The number of tools to connect
Every tool the agent has to read from or write to is a job in itself: email, calendar, CRM, management software, shared files.
They are not all equal. Some applications are designed to be connected and documented accordingly. Others barely are, or not at all. This is one of the points that can only be assessed by looking at your tools, and it is what our tool integration service covers.
Data quality
An agent works with the information it finds. Complete, current client records and consistently filed documents: the project moves quickly. Duplicates, fields each person fills in their own way, a scattered history: things have to be put in order first, and that work is part of the budget.
Volume
Volume counts in two ways. During the build, a large number of different cases to handle means more tuning and more testing. In operation, every document read and every minute of dictation transcribed uses resources. Ten reports a week and several hundred do not weigh the same.
Follow-up
An agent that goes live and is then left alone often ends up on the sidelines. Your tools evolve, your procedures change, and teams run into cases nobody had foreseen. User training and adjustments after go-live have a cost, and adoption depends on them. A budget that leaves them out is an incomplete budget.
What lightens or weighs down the budget?
| Factor | What lightens it | What weighs it down |
|---|---|---|
| Scope | One task, one team | Several chained tasks, several teams |
| Autonomy | The agent prepares, a person approves | The agent acts alone on varied cases |
| Tools | One or two tools built to be connected | Several tools, some of them closed |
| Data | Complete, current, organised | Duplicates, multiple formats, scattered history |
| Volume | Few cases, all alike | Many cases, very different |
| Follow-up | One trained team, one named contact | Many users, procedures that keep changing |
How is a budget put together?
A serious quote separates two things.
What is paid once. Designing the agent with your teams, connecting it to your tools, testing it on your real cases, training and go-live.
What comes back. Running the agent, which depends on the volume handled, and follow-up: fixes, adjustments, and changes when one of your tools is updated.
Always ask for those two parts to be set out separately in writing. An all-in price that blends them lets you neither compare two proposals nor know what the agent will cost you over time.
Ask what is out of scope as well. A good quote says what the agent will not do. That is what prevents misunderstandings.
How can you tell whether the price is justified?
By comparing it with what the task costs today. Not with a promised gain: with a measurement.
If nobody has recorded how long the task takes, how often and for how many people, there is no way of saying whether the agent is worth its price. You are buying an impression. With that starting figure, the reasoning becomes simple: on one side the time the task ties up, on the other the cost of the agent, build and running included.
That is why the quote comes after the measurement, and not before.
What is the only price we publish?
The price of the AI audit: from €800 excl. VAT. The exact price depends on the number of teams to meet, and it is agreed before we begin.
The audit takes place on your premises. It records repetitive tasks one by one, with the time they take and how often they happen, and ranks the ones an agent can take over. It also looks at your tools and your data, which are exactly the factors described above.
The quote for an agent comes out of the audit. You receive the record of time lost, the list of tasks ranked by gain, a roadmap and an estimated return-on-investment plan. You then have an estimate of what each step would cost and what it could give back, before committing anything to development.
Those documents are yours. You can carry on with us, with someone else, or stop there.
How can you keep the cost down?
A few decisions weigh on the cost.
- Start with a single task, the one the audit ranks first. You widen the scope afterwards, building on what works.
- Keep your tools. An agent plugged into what you already have avoids a migration, which is a project in its own right.
- Let a person approve. An agent that suggests and gets reviewed is simpler to build than one that must have foreseen every case.
- Prepare the data. Removing duplicates from a file or agreeing a filing rule can be done in-house, before the project.
- Name a contact person in the team. Feedback arrives faster and fewer adjustments are needed.
- Do not choose an agent when simple automation will do. The guide AI agent, chatbot or automation explains how to tell the difference.
Where to go from here?
If you have not yet identified the task to start with, read Where should an SME start with AI?. If you have one in mind and want to know what an agent would involve in your company, the custom AI agents page describes how an engagement works, and the audit is the first step.