AI for businesses in Hamburg,
without the magic show.

Most people searching for an AI agency in Hamburg are not really searching for AI. They are looking for an answer to a question that came first: why has someone in the office been retyping the same numbers from an email into another program for three years, and why is that the person who never has time.

This page answers the questions that come before a project. It explains what the word AI actually covers in daily business, which of the three shapes fits which problem, roughly what it costs and how you can tell that the effort will not pay off. If you go on to build the first one yourself, the page has still done its job.

My cards on the table: Kasoria is based in Neu Wulmstorf, in the district of Harburg, a few minutes south of the Hamburg city limits. I am one person rather than an agency with floors, and you talk to the same person from the first call to the running system.

What "AI in a business" actually means

In conversation the word AI almost always means one of three very different things, and mixing them up gets expensive, because they differ a lot in price, build time and risk.

• An automation connects programs that never spoke to each other. It follows fixed rules, it decides nothing, and that is exactly why it is reliable. There is more on the page about automation.
• An AI agent reads text, understands intent and decides for itself within a narrow frame what happens next. It is strong at language and weak at anything that has to be exactly right. How such an AI agent is built has its own page.
• Custom software is the case where neither is enough, because your process needs a surface people work in. That is when the tool becomes a small application.

Most businesses need the first one now and the second one later. Which process comes first is a question of its own, and it matters more than the choice of tool, because the order in which you automate decides whether the project continues after step one or quietly stops.

The difference between an automation and an AI agent

An automation does the same thing on every run. An order arrives, it creates the record, writes the confirmation and attaches the invoice. Run it a thousand times and it runs the same way a thousand times, and when it does go wrong, you can point at the step.

An agent decides again on every run. That is its advantage exactly when the input is messy, so with emails, call notes, photos of a delivery slip, anything a human wrote. It is its weakness exactly when the result has to be correct, because in case of doubt it will invent a plausible number instead of staying quiet.

Whether a process pays off comes down to one multiplication

The same calculation sits in front of every quote, and you can do it without me. You need three numbers: how long the process takes once, how often it happens and what an hour costs in your business.

• One run takes twelve minutes and happens forty times a month, which is eight hours a month and around ninety-six a year.
• At fifty euros an hour that is roughly four thousand eight hundred euros a year sitting inside this one process.
• If two of the twelve minutes remain afterwards because somebody checks the result, you count on four fifths of it, so roughly four thousand euros a year.

One part is missing from that calculation, and quotes like to leave it out: an automation is not a piece of furniture, it is something that runs. Interfaces change, passwords expire, a vendor renames a field. Budget for running and support from the start, or month seven arrives with a process standing still that nobody is watching.

And the honest end of that calculation: below roughly two hours a month, almost nothing is worth building and then maintaining. Processes that small are better collected until three of them together make one project.

Where your data sits when a model reads along

This is the question that comes up most often in Hamburg, and it comes up for good reason, because a language model normally means that text leaves your house. There are three ways to handle it, and they differ less in price than in effort.

• A vendor with a data processing agreement and a European region. The usual route, quick to set up and good enough for most processes, as long as the agreement really exists rather than a checkbox in the account.
• An open model on a server you control. More work and weaker in answer quality than the large models, but not a sentence leaves your infrastructure. The right route for patient records, personnel files and anything under professional confidentiality.
• No model at all. The underrated route: a great many processes that start out as an AI project turn out to need no model, only a clean connection between two programs.

Whichever route you take, one line gets drawn in every project: credentials, complete personnel records and anything covered by professional confidentiality do not go into somebody else's model. That is not caution, it is the condition for the whole thing holding up when it is questioned.

What AI cannot do in a business

The part missing from agency pages, because it sells nothing. It does save you the two most common bad purchases.

• Clean up an unclear process. If three people do the same job three ways, you automate the mess along with it. First decide how it runs, then build.
• Invent numbers that are correct. A model does not calculate, it phrases. Anything that has to add up belongs in the automation next to it, not in the prompt.
• Replace the people who know your customers. What agents are good at is preparation: sorting, summarizing, writing the draft. The conversation itself is still held by somebody on your team.

Anyone telling you otherwise is selling you the project rather than the result. The most honest sentence on the subject is this one: a good automation makes a task smaller, and the task almost never disappears completely.

Why the city is in the title anyway

One thing up front, so you do not hear it first in a call: the work is fully remote, over video with a shared screen. I do not come to your office, and that is the way of working rather than a fallback. The processes this is about run inside programs anyway, and a shared screen shows them more closely than a seat at your meeting table ever would.

If it matters to you that a provider comes to the building for meetings, I am the wrong one, and that is entirely fine. There are plenty of providers in Hamburg who handle it differently.

What the city does change is the part where nearness actually counts, which is knowing the processes. What comes up at a freight company on the Elbe, what a drywall contractor from Harburg still has to write in the evening and how a tender in the north gets checked is something you know from projects here rather than from a template. The businesses further down this page are exactly that.

On top of that comes the unglamorous part: the same time zone, the same language and an address a few minutes south of the city limits that you can look up in the commercial register. For most businesses that counts for more than the question of who sits in the meeting room.