AI agent costs: what do you pay and when does it pay for itself?
For many business owners, AI agent costs are the deciding question. You hear about agents that take over half an employee's workload, but also about projects that cost more than they deliver. So what is the truth? And how do you know in advance whether an agent will pay off in your situation?
The honest answer is that costs vary widely per application. A standard agent in software you already use costs relatively little. An agent with integrations to several systems and custom processes requires a larger investment. But in almost every case you can estimate in advance what you will pay and what it will deliver.
All amounts on this page are indicative, exclude VAT and are calculated on an annual basis. The actual costs depend on your platform, your licences and the complexity of your process.
What does an AI agent cost? The cost items at a glance
The costs of an AI agent consist of one-off and recurring items. If you only look at the licence price, you miss a large part of the picture.
| Cost item | Type | What it involves | Indication for a first agent |
|---|---|---|---|
| Licences | Recurring | The software the agent runs in, such as your CRM or helpdesk | Often already in place |
| AI usage | Recurring | The language model the agent uses, paid via credits or per use | Tens to a few hundred euros per month |
| Set-up | One-off | Working out the use case, configuring the agent, integrations and testing | A few thousand euros |
| Getting data in order | One-off | Cleaning up CRM data, filling gaps in the knowledge base | Mainly internal hours |
| Training | One-off | Teaching employees to work with the agent, including AI literacy | A few half-days |
| Management | Recurring | Monitoring, fine-tuning and maintaining the agent | A few hours per month |
Costs in Zoho, Creatio and your own AI environment
The platform you choose affects the way you pay.
Software vendors change their prices and terms regularly. So always request an up-to-date, tailored calculation.
Hidden costs to take into account
The costs that are most often forgotten are not in your vendor's quote. They are the internal hours and the costs that only become visible over time.
- Internal time. Your team spends time working out the use case, testing and giving feedback. Factor that in from the start.
- Cleaning up data. A messy CRM or an outdated knowledge base needs to be put in order first. That often takes more time than expected.
- Growing AI usage. If the agent works well, you will use it more often. AI usage grows with it. Keep an eye on that.
- Maintenance when things change. If your products, prices or processes change, the agent needs to be updated.
- Correcting mistakes. In the early phase an agent makes mistakes that you need to correct. That is why you start in assistant mode.
- Uncontrolled growth. Too many separate agents without coherence lead to higher costs and poorer results. Build up in a targeted way.
If you include these items in advance, you won't face any surprises. They don't make a good business case unachievable, just more realistic.
What does an AI agent deliver?
Set against the costs is the return. That is more than just hours saved, although those are usually the basis of the calculation.
Time savings
The most direct return is time. An agent handles tasks completely or prepares them so that an employee can finish them much faster. You can value that time against the wage costs of the employees who currently do the work.
An important nuance: time saved is not automatically money saved. Usually nobody leaves. The gain lies in capacity: your team can handle more work, can grow without extra staff and has time for higher-value work.
Returns beyond time savings
These returns are harder to express in euros, but they do matter. Include them in your business case as additional value, even if you cannot quantify them precisely.
Calculating the ROI of an AI agent
You calculate the ROI of an AI agent with a simple formula:
ROI = (return − costs) ÷ costs × 100%
The challenge is not the formula, but estimating the return and costs realistically. A worked example makes that concrete.
Worked example: an agent in customer service
Take a fictional SME with a customer service team that processes 1,000 tickets a month. The company deploys an AI agent on its own service platform.
The assumptions:
- The agent handles 30 per cent of tickets completely on its own. That saves 10 minutes per ticket.
- For the remaining 70 per cent, the agent prepares the answer. That saves 3 minutes per ticket.
- Labour costs are €45 per hour worked. That is the average for the Netherlands according to figures from Statistics Netherlands (CBS) for 2024, including employer contributions.
The time savings:
| Component | Calculation | Hours per month |
|---|---|---|
| Fully handled | 300 tickets × 10 minutes | 50 |
| Prepared | 700 tickets × 3 minutes | 35 |
| Total | 85 hours per month, 1,020 hours per year |
At €45 per hour, that represents a value of €45,900 per year.
The costs in year 1:
| Cost item | Amount |
|---|---|
| Set-up (one-off) | €8,000 |
| Training (one-off) | €1,000 |
| AI usage (€150 per month) | €1,800 |
| Management (4 hours per month × €45) | €2,160 |
| Total year 1 | €12,960 |
The existing licence for the service platform has been left out of the calculation, because the company already had it.
The ROI:
In the first year, the agent doesn't run at full strength straight away. If you calculate with 80 per cent of the full return, the return in year 1 is just over €36,700. The ROI in the first year then comes to around 180 per cent. From year 2 onwards, the one-off costs drop away and the return rises further.
A word of caution
This example has been deliberately kept simple. In your situation, the number of tickets, the time saved per task and the costs may be quite different. On top of that, the time saved translates into capacity, not automatically into lower costs. So use the example as a way of thinking, not as a promise.
What is a realistic payback period?
You calculate the payback period of AI by dividing the one-off investment by the net return per month.
In the worked example, the one-off investment is €9,000. The net return at full capacity is about €3,500 per month: 85 hours saved times €45, minus AI usage and management. The agent then pays for itself in less than three months. If you allow for a start-up period in which the agent is still learning, you end up at around four to six months.
What affects the payback period
For most well-chosen first use cases at SMEs, a payback period of a few months to a year is within reach. If it takes longer, take a critical look at whether you have chosen the right use case.
-
Volume
The more often the task occurs, the faster you recoup the investment.
-
Complexity
A standard agent pays for itself faster than an agent with a lot of customisation.
-
Data quality
With clean data, the agent performs well sooner.
-
Adoption
A team that embraces the agent sees a return faster than a team that works around it.
How to build a business case for AI
A good business case for AI doesn't just help you justify the investment. It also gives you a yardstick to assess objectively after the pilot whether the agent works. A business case doesn't need to be a thick report. These five components are enough.
-
The problem
Describe which process you want to improve and why. For example: on average, leads only get a response after two days.
-
The current situation in figures
How often does the task occur, how much time does it take and who carries it out? Use real figures from your systems.
-
The expected return
Estimate how much time the agent saves and which other benefits you expect. Be careful: it is better to work with a conservative estimate.
-
The costs
Include all one-off and recurring items, including internal hours.
-
The metrics
Record which key figures you will track to measure success, such as lead time, response time or the number of tasks handled automatically.
Keeping the costs of AI usage under control
After going live, you also want to keep a grip on costs. These measures help you avoid surprises.
- Set budgets and limits. Many platforms let you set a monthly maximum for AI usage.
- Monitor usage. Check every month how many credits or tokens the agent uses and whether that is in proportion to what it delivers.
- Choose the right model. Not every task needs the heaviest language model. For simple tasks, a lighter and cheaper model is often enough.
- Keep instructions and knowledge compact. The shorter and clearer the instructions, the lower the usage per task.
- Evaluate periodically. Check every quarter whether the agent is still doing what it should and whether costs and return are still in balance.
That way your AI investment stays profitable, even as you expand its use.
We wanted to know up front what it would cost and what it would deliver. Mobile XL's business case let us make a well-founded decision.
Calculating AI agent costs for your situation
An AI agent doesn't have to be an expensive gamble. With a clear use case, a realistic estimate of costs and return and a few good metrics, you know in advance where you stand. Mobile XL helps SMEs in the Netherlands and Flanders build a well-founded business case and then set up the agent responsibly in Zoho, Creatio or a private AI environment.
Frequently asked questions: AI agent costs and ROI
-
That depends on the application. A first agent on a platform you already use typically costs a few thousand euros to set up, plus tens to a few hundred euros per month for AI usage and management. All amounts are indicative and exclude VAT.
-
Licences, AI usage, set-up, getting your data in order, training and management. Set-up and training are one-off, the rest is recurring.
-
Subtract the costs from the return, divide the result by the costs and multiply by 100 per cent. You usually calculate the return based on hours saved times the wage costs per hour, supplemented by other benefits.
-
For a well-chosen first use case at an SME, the payback period is usually between a few months and a year, depending on volume, complexity and how well your team adopts it.
-
Largely, yes. Zoho's Zia Agents platform is free to use, you pay for the language model. At Creatio the licence includes a bundle of AI usage, with extra capacity for intensive use.
-
Internal time for working out and testing, cleaning up data, growing AI usage, maintenance when your processes change and correcting mistakes in the early phase.
-
The one-off investment is higher, and you pay on an ongoing basis for infrastructure and usage. In return you get full control over your data, and the environment can be used widely for several applications.
-
Describe the problem, map out the current situation in figures, make a conservative estimate of the return, include all costs and record which key figures you will measure.
-
More revenue through faster follow-up, higher customer satisfaction, fewer mistakes, shorter lead times and the ability to grow without a matching increase in staff.
-
Set budgets and limits, monitor usage every month, choose the right model for each task, keep instructions compact and evaluate every quarter whether costs and return are in balance.
Curious what an AI agent would cost and deliver in your business?
Book a no-obligation call and together we will work out the numbers for your first use case.