---
title: "Automate routine work with AI"
url: "https://mikkeltschentscher.dk/en/services/ai/ai-automation"
language: "en"
updated: "2026-11-01"
---

# Automate routine work with AI

AI automation removes the routine work where someone moves and corrects data by hand many times a week. I have been building automations since 2010 and make sure the solution runs in the systems you already use.

[Get a quote](https://mikkeltschentscher.dk/en/services/ai/ai-automation#inquiry "Let’s have a no-strings conversation")[+45 22 51 31 79](tel:+4522513179 "Call Mikkel Tschentscher at +45 22 51 31 79")

-   I usually reply the same day
-   Written estimate before work starts

![Mikkel Tschentscher](https://mikkeltschentscher.dk/images/derived/speaking.ae4c8c8ab739.png)

star reviews

[Read every review](https://mikkeltschentscher.dk/en/reviews)

-   Carries the project all the way through
    
    > When you work with Mikkel, you can be certain the project gets carried 100% of the way.
    
    Simon Bille Hjort MartinsenColleague at Bigum&Co(translated from Danish)
    
-   The most solution-oriented developer
    
    > In Mikkel's world there are challenges rather than problems.
    
    Michael Roscoe PedersenEntrepreneur(translated from Danish)
    
-   Competent, precise and a huge support
    
    > He is simply a pro. Highly competent and precise, brings good input, and is a huge support.
    
    Carrsten Johan Thessen(translated from Danish)
    

## AI automation

### What I most often automate

-   Document handling: extraction from invoices, order confirmations, datasheets and contracts arriving as PDFs.
-   Customer service: categorising and routing incoming enquiries so they land with the right person immediately.
-   Product data: descriptions, specifications and translations across many SKUs.
-   Internal tools: reports currently assembled by hand from three systems every Monday.
-   Integration with existing systems: moving data between webshop, accounting and inventory without manual go-betweens.

### When it pays off

The arithmetic is simple: how many times a week does the task happen, how long does each one take, and what does it cost to build. If payback arrives within a reasonable horizon, it is usually worth doing.

Beware the trap of tasks that feel annoying but only happen twice a month. They loom large in the mind and are rarely worth automating, however much everyone agrees they are stupid.

### Errors and control

Automation with AI is not deterministic, so there has to be a plan for the cases where the system is unsure or wrong.

My standard setup is a threshold: if the system is confident enough, the case goes through, and otherwise it lands in a queue a person looks at. The queue also shows how well it is really doing.

### What is AI automation?

AI automation uses language models to automate tasks that classic automation cannot handle. Rule-based tools require structured data and fixed formats, while a language model can read emails, PDFs and free text that never look the same twice. That makes it possible to automate work that previously needed a person to read and judge.

The gain is largest in companies where staff key information from documents into systems daily, sort incoming enquiries, or assemble the same reports week after week. The more often the task happens, the faster the automation earns itself back.

There is a difference between recommending automation and being able to build it. The integrations with your systems, the error handling and the threshold for when a person steps in are development work, and I do that myself. After 16 years as a developer I also know when an off-the-shelf tool is enough and when code needs writing.

I always start by looking at the actual workflow rather than the description of it, then build the smallest automation that removes real work, and measure the effect against the baseline. You own the code, the prompts and the configuration, so you can carry on without me.

### Start small: the first automation

The first AI automation has to succeed above all. So pick a workflow that happens often, where the content varies enough that ordinary rules are not sufficient, and where a mistake can be fixed. Incoming invoices or sorting a shared inbox are typical candidates. Avoid starting with something that goes straight out to customers, or that depends on data nobody quite trusts.

Build that one workflow all the way through, from the moment the document or email arrives until the result sits in the right place in your system. It is tempting to automate the first half of three different workflows, but half automations rarely move work away. They just move it somewhere new, where someone still has to pick up the pieces.

Let the automation run alongside the manual process for a while. Staff work as usual and compare with what the system would have done. That gives you numbers on how often it gets it right before anything is handed over to the machine, and it reassures the people who will work with it. Only once the first workflow runs reliably in day-to-day operations does it make sense to take on the next.

### Shadow AI: when staff find the tools themselves

Shadow AI refers to the AI tools staff use at work without the company having approved them or knowing about it. It might be a colleague pasting customer emails into a free chatbot to get a draft reply, someone who has built their own workflows in an automation service on a private account, or a browser extension that reads everything on the screen. It rarely comes from bad intent. People are trying to get their work done.

The risk lies in data security and in operations. Customer data and confidential information end up with vendors you have no data processing agreement with, and nobody has an overview of what has been sent where. At the same time, small automations appear that day-to-day operations start to depend on, without anyone knowing they exist. When the employee changes jobs or closes their private account, they stop working, and only then are they discovered.

A ban rarely works, because the need does not go away. Instead, use shadow AI as a map of where the need is. Ask openly what people use the tools for, and you have a list of the tasks most obvious to automate properly. Give them an approved alternative that does the same job, and a few short rules for the rest. Then people no longer have a reason to go around it.

### AI automation for companies across Denmark

An automated workflow runs in your systems, not at my address, so geography matters very little. I work from Glostrup in the Copenhagen area and have solved tasks for companies in Aarhus, Odense and Aalborg as well as outside Denmark, often without meeting in person.

The process is the same regardless of postcode: a video call about the workflow, access to the relevant systems, a test period where you can follow the queue, and then production. If you would rather meet face to face at the start, we will arrange it.

The tasks I most often solve remotely are invoice processing, sorting incoming post and email, and synchronising product data between systems that have never talked to each other.

Being a freelancer means short decision paths: you write directly to the person who built the solution and get an answer without going through a ticket system. For jobs of this size, that is often the difference between a fix the same week and a fix next sprint.

### Errors, accountability and personal data in automated workflows

A language model can read the same invoice correctly nineteen times and wrongly the twentieth. That is a property of the tool rather than a bug to be fixed away, and the answer is to build around it with thresholds, queues and spot checks.

Accountability stays with you. An automated process does not relieve the company of being in control of what it sends out, which is why I keep the approval step on anything going to a customer or moving money, however well the system scores in the measurements.

Personal data deserves its own consideration. Many workflows can be built so personal data is masked or left out entirely before anything reaches the model, and where that is not possible we choose EU hosting and data processing agreements that match your requirements. I document the setup so you can present it in an audit.

## What you get

Concrete deliverables, not a report of recommendations someone else has to find time for.

-   An automated workflow
    
    A solution running in the systems people already use, not in a new tab someone has to remember to open.
    
-   A threshold and a queue
    
    A setup where confident cases go through and doubtful ones land with a person straight away.
    
-   Before and after numbers
    
    An account of how long the task took before, how long it takes now, and how many cases still need a hand.
    
-   An off switch
    
    A way to switch everything off without calling me. It sounds trivial and is what people most often ask for afterwards.
    

## How it works

Four steps. You know what happens when, and you can stop after any of them.

1.  01 / 04
    
    ### Pick the process
    
    Together we find the piece of work that repeats often enough for automation to pay off. The more contained the task, the faster the first version reaches production.
    
2.  02 / 04
    
    ### Set a baseline
    
    We record how long the task takes today, how often it occurs, and how many errors slip through. Without those numbers, nobody can tell afterwards whether the solution was worth the effort.
    
3.  03 / 04
    
    ### Build against a test set
    
    I collect real examples from your day-to-day and test every prompt and model change against them. That keeps quality stable when the solution is tuned later on.
    
4.  04 / 04
    
    ### Production with logs and a kill switch
    
    The solution is wired into your existing systems with full logging and a switch that can turn it off. I document the setup and train someone on your side to maintain it.
    

## Brands I've worked with

-   ![MT Højgaard Danmark](https://mikkeltschentscher.dk/logos/mt-hoejgaard.svg)
    
-   ![Egmont](https://mikkeltschentscher.dk/logos/egmont.svg)
    
-   ![NNIT](https://mikkeltschentscher.dk/logos/nnit.svg)
    
-   ![Visma](https://mikkeltschentscher.dk/logos/visma.svg)
    
-   ![Lomax](https://mikkeltschentscher.dk/logos/lomax.svg)
    
-   ![Pascal](https://mikkeltschentscher.dk/logos/pascal.svg)
    
-   ![able.](https://mikkeltschentscher.dk/logos/able.svg)
    
-   ![OOONO](https://mikkeltschentscher.dk/logos/ooono.svg)
    
-   ![Novo Nordisk Fonden](https://mikkeltschentscher.dk/logos/novo-nordisk-fonden.svg)
    
-   ![Energii](https://mikkeltschentscher.dk/logos/energii.svg)
    
-   ![Maersk Tankers](https://mikkeltschentscher.dk/logos/maersk-tankers.svg)
    
-   ![Nordkysten Entreprenørfirmaet](https://mikkeltschentscher.dk/logos/nordkysten.svg)
    

## Frequently asked questions

How does AI automation differ from ordinary automation?

Classic automation follows fixed rules and requires structured data. AI automation can also handle text and documents that never look alike, such as emails and PDFs in every possible format. That opens up tasks that previously needed a person to read and judge.

What is shadow AI?

Shadow AI refers to AI tools staff use at work without the company's approval or knowledge, such as a free chatbot or an automation service on a private account. The problem is that data can end up with vendors without a data processing agreement, and that nobody has an overview. The answer is rarely a ban, but an approved alternative and clear rules.

Which tasks can you automate with AI?

Tasks that happen often and follow a pattern, even when the content varies. Typical examples are sorting incoming email, extracting data from invoices or forms, first drafts of replies, summarising cases and meetings, and moving data between systems. That kind of work exists in most small and medium-sized businesses. Where it matters, a person approves the result before it moves on.

Which AI program is best?

For chat, ChatGPT, Claude, Gemini and Microsoft Copilot are all strong, and the right choice often depends on what you already use. For automation, the model itself matters less. What decides it is how it is connected to your systems and how the result gets checked.

What does AI automation cost?

The price follows the complexity. The number of workflows, the access to your systems and the volume of documents going through are what typically move it most. Describe the workflow in an email and you will get a reply the same day, and a fixed quote once a few details are in place.

Do we need an IT department to get started?

No. What it takes is a person who knows the workflow well, access to the systems the automation needs to work in, and someone who owns the solution afterwards. I handle the integrations and the technical setup.

Do you use tools like Zapier or Make?

For simple chains, yes, it is faster and cheaper. As soon as errors need proper handling or volume gets large, writing it as code becomes cheaper.

How are personal data and GDPR handled?

The first step is deciding whether personal data needs to pass through the model at all, because masking can often avoid it. Where it cannot, we choose hosting and data processing agreements that match your requirements, and everything is documented so you can answer if anyone asks.

Do you automate for companies outside Copenhagen?

Easily. The solution runs in your systems, so it makes no difference whether you are in Aarhus, Odense, Aalborg or outside the country. Most of it is handled over video, and I am based in Glostrup if you want to meet.

Does this mean someone loses their job?

In the projects I have done, the answer is usually no. What disappears is the part of the day nobody liked. The decision is yours, though, and I believe it should be taken openly rather than sprung on people.

Can we start small?

Yes, and it is the best way to begin. Pick one workflow that happens often and where mistakes can be fixed, build it all the way through, and let it run alongside the manual process until the numbers show it gets things right. Then you can take on the next one with the experience from the first.

How quickly does an effect show?

On a bounded workflow there is usually something running within a month. The effect can be assessed from day one, provided the baseline was measured beforehand.

Can we adjust it ourselves afterwards?

The parts about rules and phrasing, yes. I put those somewhere they can be changed without touching the code, while the integrations themselves need a developer.

Your next step

## Get a quote for AI automation

Tell me about your project. I’ll get back to you with questions, options and a concrete next step.

![Mikkel Tschentscher](https://mikkeltschentscher.dk/images/derived/speaking.ae4c8c8ab739.png)

### Mikkel Tschentscher

Developer, consultant and instructor

-   16+ years of experience
-   Work directly with me
-   From advice to implementation

[+45 22 51 31 79](tel:+4522513179)[hello@mikkeltschentscher.dk](mailto:hello@mikkeltschentscher.dk)

Mikkel Tschentscher ApS · CVR 46886712

star reviews

## Related

-   [AI agents](https://mikkeltschentscher.dk/en/services/ai/ai-agents)
-   [AI implementation](https://mikkeltschentscher.dk/en/services/ai/ai-implementation)
-   [AI development](https://mikkeltschentscher.dk/en/services/ai/ai-development)
-   [AI workshop](https://mikkeltschentscher.dk/en/services/workshops/ai-workshop)

Original: https://mikkeltschentscher.dk/en/services/ai/ai-automation
