AI, customer service, cost

What does an AI agent cost a business?

Less than people fear for the model, and more than they expect for everything around it. At small-business volumes the model calls cost little each month. What you pay for is an agent that answers only from what you actually know, hands over to a person when it should, and gets better from its own conversations. A narrow agent that answers customers from your own content, with an inbox where a person can step in, is a matter of weeks rather than months.

Published: September 27, 2026

What you are actually paying for

An AI agent is sold as one thing and is usually six:

The model calls: the part everyone asks about, and at small-business volumes the smallest line.
The knowledge base: what the agent is allowed to say, written down and kept current. Most of the lasting effort goes here, and some of it is yours, because nobody else knows your business.
The limits: what it must never answer, how it says it does not know, and when it stops and hands over.
The handover: an inbox where a person sees the conversations and can step in, so no customer is left stuck with a machine.
The actions: checking availability, booking, looking up an order. Each one is its own integration, with its own permissions and its own way of failing.
Operations: monitoring, a ceiling on what it can spend, and updates when the provider changes something.

When two quotes are far apart, they almost always count a different number of these.

Three sizes of agent

One that answers. It knows your content (opening hours, prices, rules, how things work) and answers from it around the clock, with a person behind it. This is the one most businesses need first, and the one that pays back fastest.

One that also acts. It can check whether a slot is free, make the booking or find an order. That is worth a lot, and it is also where the work grows, because every action needs its own permission and a plan for when the system on the other side does not answer.

One inside the workflow. It reads incoming email, sorts it and drafts a reply for a person to approve. It saves the most time and needs the most care, because it works where your customers cannot see it.

What it looks like at Vestfjord Sauna

On vestfjordsauna.no an agent answers guests around the clock from a knowledge base of what the business actually knows. An operator has their own inbox, sees the conversations and can take over, and the agent learns from the conversations it has.

That last part is the one that matters for cost. The more an agent learns from what it could not answer, the less often it needs a person.

Where the cost runs away

An agent allowed to answer everything. It will answer things it does not know, and every wrong answer lands on somebody’s desk.
No way out to a person. A customer going round in circles with a bot costs more than the conversation was worth.
A knowledge base nobody owns. Answers go stale quietly, and the agent keeps giving them with full confidence.
A price per conversation. Fine at the start. Worth doing the sum before the volume grows.
Where the conversations are processed. It is a choice, and it is cheaper to make on purpose at the start than to change later.

How to find out in an hour

We do not put a price on an agent before knowing what it has to do: answering questions about opening hours and handling customer orders are not the same job.

Describe it in the Living Brief and you get an architecture sketch, phases and a range in weeks straight back, without anyone phoning you. If you are not sure where an agent would help, the free AI review looks at the whole business and points at what is worth automating first.

Field notes

Questions we get

The same questions, every time. These are the answers we give across a table.

Can we not just use an off-the-shelf chatbot?

Often you can, and we say so when that is the right answer. A ready-made chatbot is fine for answering questions from a help page. It stops fitting when the agent has to do things in your own systems, when you want a say in where the conversations are processed, or when a price per conversation starts to grow with your volume.

What does the model itself cost per month?

At the volumes a small business sees, very little: a couple of thousand conversations a month is small change next to building and maintaining the agent. The providers set their own rates and change them, so take the current price from them rather than from an article.

How long until it is live?

A narrow first version that answers from your own content and hands over to a person is typically a few weeks of work. You get a demo you can try every week along the way, so you see what it says before your customers do.

What happens when it does not know the answer?

It should say so and hand the conversation to a person. The question it could not answer then becomes new knowledge in the knowledge base, which is how an agent gets better. An agent that guesses costs more than one that says it does not know, because every wrong answer has to be cleaned up by somebody.

Do we own the agent?

Yes. The code, the instructions and the knowledge base are yours, written so another team could take them over. We do not build anything you cannot take somewhere else.

Field notes