I set up AI that sorts customer enquiries, summarises them and writes a draft reply — a person reads it and sends it.

No bot answers the customer directly. You receive the documentation, and the assets are in your name.

The most common picture of AI in customer service is a chatbot that answers on the website, and that is almost never where I start. A chatbot is the most visible point in the entire customer journey — every mistake becomes a bad encounter, in front of a customer who often knows you personally.

What actually saves time in a small business is invisible to the customer: enquiries get sorted and summarised, and a reply is already written when someone sits down to answer. Read more about what actually works. It is one of several narrowly defined AI services I offer — see the full range on the AI consulting page.

I also run my own systems in live use, among them Nyheter AI, where an editorial team of AI agents monitors 27 sources around the clock. That is how I know what breaks after three months, which is a different thing from what works in a demo.

What can actually be built

Sorting what comes in

Enquiries are tagged and land with the right person, with a summary of a few lines at the top. Nothing goes to the customer at this stage, and it is often the single biggest time saving in an operation where several people read the same inbox.

Draft replies

Once the enquiry is sorted, a suggested reply is ready, built on how you actually reply. The employee reads it, changes it where needed and sends it — the heavy part is rarely writing the reply, it is getting started.

Background for first-line staff

A summary of the case history, previous purchases or similar enquiries, so that whoever replies does not have to search for the context themselves.

Direct replies to customers — only in a narrow category

Opening hours, delivery status, where to find an invoice. Unambiguous, harmless questions with the same answer every time. Never questions about money, contracts or complaints.

AI chatbot for customer service – when I build one

If you want an AI chatbot for customer service, I can build one, but a narrow one. It answers the unambiguous questions – opening hours, delivery status, where the invoice is – and everything else goes on to a person.

Two conditions always apply. The way to a person must be visible in the very first reply, and the bot must never answer anything to do with money, contracts or complaints.

Sorting and draft replies often give back more time than the bot, and then that is where we start. If you do not know which suits you: describe the volume and what the questions are about in the first call, and I will tell you – even when the answer is that a bot does not pay off for you.

How a build works

A free first call. You describe how many enquiries you receive and what they usually concern. I tell you whether it can be lifted off you, roughly what it would come to, and sometimes that it is not worth the money.

A fixed price. Quoted according to scope — the difference between sorting an inbox and writing drafts for several types of enquiry is too large for a list price. You know the cost before the work begins.

The material is gathered. The first piece of work is almost always gathering what you actually reply — earlier emails, an FAQ, what is in the head of the person who has been answering for years. That work is worthwhile even if you later decide against automation.

A person at the last step. The system sorts, summarises and drafts — someone reads and sends. It removes most of the time without moving the responsibility.

Handover and a follow-up after a month. Both are included. Only then do you see how much the employees actually change in the drafts, which tells you more than any other measure.

When AI should not answer the customer

When the enquiry is sensitive. Complaints, claims and anything that calls for empathy rather than a correct answer belong with a person from the start, not as an exception discovered after three messages.

When a wrong answer costs a customer. The technology is sometimes wrong, and it is wrong with great confidence in its voice. In a small business where customers often know you personally, that risk is seldom worth the saving.

When the questions are not repetitive. If every enquiry is unique there is no pattern to build the drafts on, and the preparatory work becomes longer than the work you wanted to save.

More on the reasoning — and on when a bot is actually the right solution — in the article on what actually works. If the aim is instead to automate other recurring tasks, such as quotes or reports, there is AI automation; if you are setting up a larger system, there is AI implementation.

Who you are hiring

I have a Master's degree in psychology from Örebro University, work as a speaker and coach, and have published a non-fiction book. In parallel I build and run a number of digital services — among them My AI team, which builds bespoke AI employees for small and medium-sized businesses, and Nyheter AI, where an editorial team of AI agents monitors 27 sources around the clock.

Glänne & Söner is a one-person business. That it still manages publishing, coaching, talks and development is because I have built an AI team that handles parts of the work — the same kind of system I build for others, and because I run my own in live use, I know what breaks after three months.

The combination of psychology and technology is unusual for an AI consultant, and in customer service it is the point rather than just the technology: whether a customer reply lands well depends seldom on how well it is written, but on whether it meets the customer where they actually are. That is harder for a model to judge than for someone who understands people.

More about me

Common questions about AI customer service

What does an AI customer service solution cost?

Builds are quoted according to scope, since the difference between sorting an inbox and writing drafts for several types of enquiry is too large for a list price to mean anything.

Roughly: a limited build that sorts and drafts replies for one type of enquiry usually comes to a few half-days, not a few months. You receive a fixed price after a first free meeting, before the work starts, not an ongoing invoice. Benchmarks are in the article on what an AI consultant costs.

Does the AI reply directly to our customers?

Almost never, and that is a deliberate limitation. The system sorts enquiries, summarises them and writes a suggested reply — a person reads it, changes it if necessary and sends it.

The exception is a narrow category of unambiguous questions such as opening hours or delivery status, and only if you yourselves want that.

Can you build an AI chatbot for our customer service?

Yes, but it is rarely where I start. A chatbot answering on the website is the most visible point in the whole customer journey, and every mistake meets the customer directly.

Where a bot fits, I keep it to the unambiguous questions — opening hours, delivery status, where the invoice is — and let everything else go on to a person.

Do we need to change our customer service system?

Usually not. The aim is for the sorting and the drafts to end up where you already work, whether that is a shared inbox or a customer service system. I do not sell a customer service system of my own – the sorting and the drafts are connected to what you already have.

What your system allows is something we go through in the first conversation. Sometimes the answer is that nothing sensible can be connected, and then I say so.

What is needed for a draft to be good?

The answers need to exist somewhere. Most businesses have them in earlier email conversations, in an FAQ or in the head of the person who has been answering for years.

The first piece of work is therefore seldom technical, but gathering what you actually usually reply, and that work is worthwhile even if you later decide against automation.

What happens to personal data in customer enquiries?

Customer enquiries almost always contain personal data, sometimes sensitive.

Before anything is automated we go through where the data ends up, whether there is a data processing agreement with the supplier, and how long the material is kept.

How many customer enquiries do you receive in a week?

Describe the approximate volume and what the most common questions concern, and I will tell you whether it can be lifted off you and roughly what it would come to. Free of charge.

Book a first meeting