What does it costto introduce AI in an SME?

Why is there no price list?

Every week someone asks us what AI costs. It is a fair question, and the honest answer is that it depends on what you want AI to do. Two companies of the same size in the same sector rarely have the same processes. One wants AI to read incoming orders and book them in the ERP. The other wants a digital colleague that follows up every quote until the client answers. Both are AI projects, but they are not the same amount of work.

What we can do is show what drives the cost, so you can estimate it for your own situation before you talk to anyone. That is more useful than a number on a website that is wrong for most readers.

The model is the cheapest part

Many people expect the AI model to be the big expense. In practice it is usually the smallest line. Using a large language model through an API costs cents per document or per conversation. Even a busy company that lets AI read thousands of mails a month often pays less for the model than for one software licence.

The real work sits around the model: getting the right information to it, connecting it to the software you already use, making sure people can check what it does and keeping everything secure. That is where the hours go, and that is what you are really paying for.

Five things that drive the cost

The first is scope: how many steps of a flow you want to change. Letting AI read one type of document is a small project. Changing a whole flow, from the first mail to the invoice, is a bigger one.

The second is your current software. If your ERP, CRM or property software has a decent API, connecting to it is straightforward. If data lives in an old system without one, or in Excel files on a shared drive, it takes more work to read and write safely.

The third is the state of your data. You do not need perfect data to start, but if the same client exists three times under different names, someone has to decide which one is right. We usually clean up only what the flow needs.

The fourth is control: how strict approvals and access rights need to be. A suggestion that a person approves with one click is simpler than an agent that acts on its own within agreed limits. Both are possible, but the second needs more safeguards and testing.

The fifth is the number of people and teams involved. A flow used by three people in one office is easier to roll out than one used by forty people across sales, planning and the field.

What a project usually looks like

Most projects we do run from a few months to a year. A single AI step inside software you already have sits at the short end. A platform with several agents, approvals and signals sits at the long end. We build in sprints of two weeks, so you see working software early and steer while it is being built, instead of waiting months for a big reveal.

Costs follow the same pattern. A focused step can be a modest investment that pays itself back within a year through saved hours. A full platform is a larger investment, but it replaces several tools and a lot of manual work at once. We always price on your situation, never from a template.

Costs people forget

Running costs: hosting, model usage and monitoring. These are usually small, but they are recurring, so put them in your calculation from day one.

Your own time: someone from your team needs to explain how work really flows, test early versions and give feedback. A project without that input takes longer and fits worse. Count on a few hours a week from the people who know the work best.

Change: people need to learn a new way of working. Training is short when the software is built around how they already work, which is one more reason to start from your process instead of from a tool.

How to think about payback

The simplest way is to count hours. How many hours a week does your team spend retyping, searching and chasing? Multiply by a realistic hourly cost and by the share of that work AI can take over. In most companies we visit, one well chosen flow frees up somewhere between a few hours and a few days a week.

The less visible gains matter too: fewer mistakes, faster answers to clients, nothing slipping through when someone is on holiday. They are harder to put a number on, but clients usually notice them first.

An example in numbers

Take a property manager with twelve people who handle about eighty mails a day from owners, tenants and contractors. Reading each mail, finding the right building and file and passing it to the right person takes a few minutes per mail. Together that is well over a full working day, every day. A mail agent that reads and links those mails, with a person approving the link, takes most of that work away. The investment for such a flow is far below the yearly cost of the hours it saves, and the model usage costs a few euros a day.

How to start small

Do not start with a big AI strategy. Start with one flow that costs your team a lot of time and is easy to measure. Prove it works, then add the next one.

What it costs depends on what we build, because that follows your processes. That is why we always price on your situation. We map how work flows through your company, show where time gets lost and where AI and software pay off, and give you a concrete proposal with a price. After that you decide, with real numbers in hand instead of guesses.