Pricing · Building an AI agent
What does building an AI agent cost? The honest numbers
As an indication: a simple workflow takes 20 to 40 hours to build, an agent connected to your ERP 60 to 150 hours. At € 150 per hour, that's € 3,000 to € 22,500. Then come hosting, maintenance and model usage. Here you'll see where that money goes, and how to keep it affordable.
- Indications, not a quote
- € 150 per hour, one rate
- Always a human in the loop
NinA Agent · quote request
livereads the email and attachment (PDF, 3 pages)
looks up the customer in Exact: existing customer, volume discount
fetches prices and stock from the ERP
model usage for this task: ±34,000 tokens ≈ $ 0.10
01 / The short answer
What does an AI agent cost?
As an indication, having an AI agent built by NinA takes between 20 and 150 hours. At our rate of € 150 per hour, that's € 3,000 to € 22,500. The more systems, exceptions and autonomy, the more hours. An agent that works across several departments can come in above that.
After that, you pay every month to keep it running: hosting, maintenance and model usage. At NinA, that's covered by a maintenance contract of € 245, € 445 or € 895 per month, including AI usage.
One thing up front: we only build agents within an AI Partnership. So the hours below come out of your partnership hours. You'll read why further down. And these figures are indications, not a quote. We work out the real hours in the AI Quick Scan, based on your own processes and systems.
01Simple workflow
20 to 40 hours, € 3,000 to € 6,000.
02Workflow with AI and several systems
40 to 80 hours, € 6,000 to € 12,000.
03Agent connected to your ERP
60 to 150 hours, € 9,000 to € 22,500.
04Agent across several departments
From 150 hours, from € 22,500.
05Then, per month
Maintenance € 245, € 445 or € 895, including AI usage. Silver and Gold also include hosting in the EU.
02 / Prices by type
From a simple workflow to an agent in your ERP.
Four types, from simple to complex. The difference isn't in the AI itself but in everything around it: how many systems, how many exceptions and how much the agent is allowed to decide on its own.
| Simple workflow | Workflow with an AI step | Agent connected to the ERP | Agent across departments | |
|---|---|---|---|---|
| Example | Form to CRM, with a confirmation email and a notification in Teams | Reading invoices from email and PDF and preparing them in the accounting system | Reading quote requests, looking up customer and prices in the ERP, preparing a draft | Customer questions, orders and returns across several systems and teams |
| Systems | 1 or 2 | 2 or 3 | 3 or more, including the ERP | Several, per department |
| What the AI does | Little: mostly fixed steps | Reading, recognizing, filling in | Planning, looking up, drafting | Planning, choosing, routing to the right place |
| Build layer | 3: workflow automation | 3: workflow automation | 4: custom build | 4: custom build |
| Indicative build hours | 20 to 40 hours | 40 to 80 hours | 60 to 150 hours | From 150 hours |
| Indicative build cost (× € 150) | € 3,000 to € 6,000 | € 6,000 to € 12,000 | € 9,000 to € 22,500 | From € 22,500 |
| Human in the loop | Spot checks | A person approves | A person approves | A person approves, per team |
Swipe sideways to see all columns →
Indications from NinA AI, October 2026, based on the hours this kind of work takes us. Not a quote: we work out the real hours in the Quick Scan. All amounts exclude VAT. Running costs are listed below.
03 / Where the hours go
Why one agent takes 20 hours and another 150.
You can build an agent demo in an afternoon. An agent that works in your own systems every day, with real customers and real exceptions, is something else. This is where the hours go.
Usually the biggest item
Integrations
Every system is an integration: Exact, AFAS, SAP, Dynamics, HubSpot, Salesforce, your WMS. If it has a good API, it goes quickly. If not, it takes hours.
Determines the quality
Data and examples
An agent learns your standards from examples: good quotes, good answers. If there aren't any, or the data is messy, that's where the time goes first.
The sting in the tail
Exceptions
The customer with the unusual discount, the order sent as an Excel file instead of a PDF. Every exception the agent has to recognize costs build and test time.
Always needed
The place where a person approves
A screen, an email or a Teams message where an employee sees the proposal and approves or adjusts it. Keeping it simple saves hours.
Can't be skipped
Security and logging
Permissions, passwords stored in the right place, a record of what the agent did and why. So you can trace it afterwards.
The final stretch
Testing, documentation and handover
Testing with real data, a period in which a person checks everything, and documentation so that the code and the knowledge are yours.
04 / What it looks like
What you're actually having built.
An email with an attachment comes in, the agent reads it, extracts the data and prepares it in your own systems. Every step that touches one of your systems is an integration, and every integration costs hours.
- 01
An email comes in
Integration: mailbox
A quote request lands in the shared inbox. The agent picks it up, attachment included.
- 02
The agent reads it
AI step
Customer, items and quantities, taken from the email and the PDF.
- 03
Looks it up in the ERP
Integration: ERP
Customer details, prices, discounts and stock, straight from your own systems.
- 04
Prepares the draft
Integration: ERP or CRM
A draft quote, ready in the system your team already works in.
- 05
A person approves
Approval screen
An employee checks it, adjusts where needed and approves. Only then does it go out.
05 / Running costs
What does an AI agent cost per month?
After the build, you pay for three things. The build is a one-off; these costs continue for as long as the agent is running.
€ 245+
Hosting and maintenance
The maintenance contract
Bronze € 245, Silver € 445 or Gold € 895 per month, invoiced annually in advance. Always with updates and AI usage. Silver and Gold also include EU hosting, round-the-clock monitoring and backups; with Bronze the agent runs on your own servers.
cents
Model usage
Per task, in tokens
The AI model charges per piece of text. With a mid-sized model, preparing a quote costs about 10 US cents. At NinA, this is included in the maintenance contract.
€ 150
Further development
Per hour, from your package
A new system, a new exception, a second department. Those are hours from your partnership, at the same rate.
06 / Worked example
A quote agent, from build to payback.
An example, not a client case. An account manager produces 20 quotes a month and spends 4 hours on each one: looking up details, checking prices, copying and pasting text. With an agent that prepares the draft, that drops to 1 hour. The account manager still decides on the price and the relationship.
The saving
3 hours × 20 quotes = 60 hours a month. At € 45 per hour, that's € 2,700 a month, or € 32,400 a year.
The costs
- Build: an agent connected to the ERP, say 100 hours × € 150 = € 15,000, from your partnership hours
- Maintenance: Silver contract, € 445 a month, including hosting and AI usage
- Model usage per quote: about 30,000 input tokens and 4,000 output tokens. With Claude Sonnet 5.5 ($ 2 and $ 10 per million tokens, Anthropic, October 2026; check the current price with Anthropic), that's about $ 0.10 per quote, or $ 2 a month
The outcome
After maintenance, you're left with € 2,255 a month. That pays back the build in about seven months. After that, the agent delivers around € 27,000 a year, for one account manager. If five account managers use the same agent, the numbers change considerably, while the build stays almost the same.
Note: the guarantee of at least 2x back applies to the partnership as a whole, not to a single agent. In the Quick Scan we add up all the processes on the roadmap.
01Saving
60 hours a month, € 2,700.
02Build
Indication 100 hours, € 15,000.
03Running
€ 445 a month for maintenance, model usage included.
04Paid back
In about seven months.
05Scale
More users, almost the same build. That's where the real gain is.
07 / Why within a partnership
Why we don't build standalone agents.
An agent isn't a website you deliver and forget. Models get updated, your ERP gets an update, a new type of customer question comes in. An agent without an owner works worse after six months, or stops working altogether. That's where one-off builds often run aground: the builder is gone, nobody knows how it works, and the team goes back to retyping.
That's why we only build within the AI Partnership. You get a dedicated team that builds the agent, trains your people, keeps it running and adjusts it. With a roadmap, so the second agent follows the first.
In terms of hours, that means an agent of 60 to 150 hours fits into package S (30 hours a month) in two to five months of pure build time. Besides building, hours also go into advice, training and meetings. In package M (50 hours) or L (80 hours) it goes faster, and we build two or four processes at a time.
01Package S
€ 4,500 a month, 30 hours, one process at a time.
02Package M
€ 7,500 a month, 50 hours, two processes at a time.
03Package L
€ 12,000 a month, 80 hours, four processes at a time.
04Always first
The AI Quick Scan, € 6,000. Then a minimum of 12 months, invoiced quarterly in advance.
05No lock-in
Code, data and documentation are yours, and you take them with you within 30 days.
08 / How it works
From idea to an agent that runs.
- 01
Business case
Quick Scan
What does the process cost today, what does an agent deliver and how many hours will the build take? We only continue at 2x back.
- 02
Design
After the Quick Scan
The process in steps, the role of person and AI at each step, the systems and the exceptions.
- 03
Build and connect
Depends on the hours
First a working version with real data, then the integrations, the logging and the approval screen.
- 04
A person approves everything
Test period
An employee sees every proposal. That's how you learn where the agent does well and where it doesn't.
- 05
Live and maintained
Ongoing
Monitoring, updates and adjustments. Every month, a report on what the agent delivered.
09 / Ways to save
How to keep an AI agent affordable.
Check whether you need an agent at all
Sometimes a good prompt or your own assistant is enough. That takes no build hours, just a workshop for your team.
Start with one process
One process, one team, the most common variant. Exceptions come in the next round. Start small.
Have examples ready
Twenty good quotes or a hundred good answers save build and test time. And they make the agent better.
Choose the model that's good enough
The most expensive model is rarely needed. We test models on everyday office work in our benchmarks and choose one per step.
The guarantee
At least 2x back in the first year. Otherwise we don't build it.
If the business case from the Quick Scan doesn't show a saving of at least twice the investment, we advise a smaller package, a different process or no agent at all. An anonymized example from practice: € 77,000 saved against € 28,000 in build costs, paid back in about 4.5 months.
10 / Agents at work
What it delivers.
11 / Questions
Frequently asked questions about the cost of an AI agent.
What does it cost to build an AI agent?
As an indication, a simple workflow takes 20 to 40 hours to build (€ 3,000 to € 6,000) and an agent connected to your ERP 60 to 150 hours (€ 9,000 to € 22,500), at a rate of € 150 per hour. An agent across several departments costs more. Hosting, maintenance and model usage come on top. All amounts exclude VAT.
What does an AI agent cost per month?
At NinA, you pay a maintenance contract for whatever runs: Bronze € 245, Silver € 445 or Gold € 895 per month, invoiced annually in advance, excluding VAT. It always includes updates and AI usage. With Silver and Gold we host the agent in the EU and monitor it around the clock; with Bronze it runs on your own servers. Further development comes out of your partnership hours, at € 150 per hour.
Can NinA build a standalone AI agent for me?
No. We only build within the AI Partnership, from € 4,500 per month for a minimum of twelve months. An agent needs to be managed and adjusted after delivery; without a dedicated team it grinds to a halt within a few months. What you can buy on its own: the AI Quick Scan, talks and workshops.
What does an agent's model usage cost?
Usually cents per task. According to Anthropic (October 2026), Claude Sonnet 5.5 costs $ 2 per million input tokens and $ 10 per million output tokens; check the current price with Anthropic. Preparing a quote with about 34,000 tokens then costs roughly $ 0.10. At NinA, AI usage is included in the maintenance contract.
Why do prices for AI agents vary so much?
Because the AI itself is rarely the expensive part. The hours go into integrations with your systems, the quality of your data, the number of exceptions, the screen where a person approves, security and testing. An agent that reads from one system is a different job from an agent that writes to your ERP.
How long does it take to build an AI agent?
That follows from the hours. In package S (30 hours a month), an agent of 60 to 150 hours takes two to five months of pure build time, alongside advice and training. In package M or L it goes faster. At NinA, the first process goes into production in the first quarter of the partnership.
What's the difference between an AI agent and a workflow?
A workflow follows fixed steps: when this comes in, do that. An AI agent is given a goal, plans the steps itself, uses your systems and presents the result for approval. Workflows are cheaper and more predictable; agents are needed when the work has too many variations for fixed rules.
Who owns the agent once it's finished?
You do. Code, data and documentation belong to the client, and you take them with you within 30 days if you stop. Data stays in the EU, on your own infrastructure or on servers with a European provider.
Want to know what your agent would cost? Book an intro call
Tell us which process you want to automate. In fifteen minutes we'll give you a first estimate, and tell you whether an agent is worth it.
01
Intro call
Fifteen minutes, no strings. Where does work get stuck and what do you want to achieve?
02
AI Quick Scan
One day on site. Within a week, a business case and a roadmap.
03
AI Partnership
A dedicated team that builds, trains and shows what it delivers every month.
160+ organizations · Sessions rated 9.3 · Based in Amsterdam