Text Link
1/5

Desired Service

PROJECT DETAILS

2/5

BUDGET

3/5

TIMEFRAME

4/5

CONTACT DETAILS

We use your details solely to process your enquiry. Details in our privacy policy.

5/5

THANK YOU

Our AI agent was quick: the reply to your initial enquiry is already in your inbox.

ERROR

THANK YOU!

We will get back to you shortly.

AI agents · Development and operation

AI agent development for whole workflows

Receive, transcribe, match, extract, write back. We build agents that run a defined workflow through to the end, with approvals wherever a person should decide.

Five stagesApproval before writingOn your own hardwareEvery step logged
94 %
Less effort
23 h to 84 min per event
800+
Meetings processed
in the running project
3
Languages
German, English, Japanese
0 EUR
For the first call
even when we advise against it
AI agent
An AI agent does not wait for the next instruction. It runs a defined workflow through to the end on its own: take the input, process it, write the result back into the systems. A chatbot answers questions. An agent completes steps. This is what people mean by agentic AI.
01 Starting point

The meeting ends. The work does not.

The recording sits in a folder. Someone writes the minutes in the evening. What was decided arrives as accurately as that person remembers it.

Add several sites and you add time zones and languages. Just finding out who currently owns a topic can take until next week.

We build the part in between. Decisions stay with people, the handovers go to the machine. Wherever a check is needed, a person sees the result before it is written. That is how the meeting project for a client with sites in Germany and Japan has run for months.

88 % use it, a third scale it
88 % of surveyed organisations use AI regularly in at least one business function. Nearly two thirds have not started rolling it out across the enterprise, and only 39 % report any effect on operating profit at all. An agent that takes over a whole workflow is the shortest route from trying to running.
Source: McKinsey, The state of AI in 2025, Global Survey, 5 November 2025, 1,993 respondents across 105 nations. mckinsey.com
02 Structure

Five stages, each on its own

Each stage runs independently. If one fails, the workflow resumes there, not at the start.

Stage 1

Intake

The recording and calendar data come in and are matched to participants and agenda.

CalendarRecordingMatching
Stage 2

Transcription

Speech becomes text. If the recording must not leave the building, speech recognition runs on your hardware.

WhisperOwn hardwareNothing sent out
Stage 3

Name matching

The same person appears as Müller, Mueller and M., plus whatever speech recognition misheard. Matching against the staff directory merges them. Skip this stage and the Owner column is worthless.

DirectorySpellingsCheck
Stage 4

Extraction

Decisions, tasks, deadlines, owners. Before anything is committed, it all sits in one list that a person reviews.

DecisionsDeadlinesApproval
Stage 5

Write back

Into the knowledge system, into Microsoft 365, into the line of business application. From then on the content is searchable.

Microsoft 365DatabaseSearch
03 Components

What we put together

01

Speech recognition

Runs entirely on your side if you want it to. In the meeting project it runs on the client's hardware for exactly that reason.

WhisperLocalNothing sent out
02

Matching against master data

The staff directory is the truth. Spelling variants and misheard names are mapped onto it. If nothing matches, the agent asks instead of guessing.

Spelling variantsMishearingsQuery
03

Extraction rules

Decision, task, deadline, owner. A fixed pattern, so the results can be counted and searched later.

DecisionsDeadlinesOwners
04

Human approval

A list to review before anything is committed. Corrections feed into the next runs.

Before writingCorrectionLearning
05

Integration

Through APIs where they exist. Where they do not, through file handover or screen automation. We pick connection standards such as MCP only once we know which systems are involved.

Microsoft 365DatabaseRPA
06

Log

Which stage did what and when stays on record. Without it, nobody can trace a cause afterwards.

LogRestartTraceable
04 Where it fits

When an agent fits and when it does not

Lots of meetings alone are not a reason.

SituationBetter choiceWhyWhat we recommend
Recurring meetings several times a weekAgentMinutes and distribution cost hoursThe typical case
A few meetings a monthWaitBuild and operation do not pay backStandardise the template first
Clear decision rulesRule based automationA language model blurs decisionsImplement without a model
Recordings must not leave the buildingOperation on your own hardwareSpeech recognition runs in house tooLocal configuration
Several sites and languagesMultilingual setupSpellings and language switches lower accuracyMatching as a stage of its own

Scroll sideways to see all columns.

05 Approach

Four steps to parallel operation

We start with one type of meeting. That is where the assumptions prove right or wrong, and everything else depends on it.

Before that we measure the current state. Without a measurement there is no effect to show later. Technical documentation under the EU AI Act is part of the handover.

1 Baseline
Which meetings cost how many hours a month?
2 Define the pattern
What gets extracted, where a person checks, where it is written.
3 Build
Stage by stage, run through in a test environment.
4 Parallel operation
Old and new ways run side by side. The switch happens after the accuracy has been checked.
06 Common questions

AI agent development: common questions

What is the difference between an AI agent and a chatbot?+
A chatbot answers a question. An agent works through a defined workflow: take the recording, transcribe it, match names, extract decisions and owners, write to the minutes system. Nobody has to trigger the individual steps.
Our recordings must not leave the building. Is that still possible?+
Yes. Speech recognition then runs on your hardware. The running meeting project is built exactly that way, for the same reason. At most a summary goes outside, and even that only to servers in the EU or not at all.
What happens when the agent makes a mistake?+
Every stage has an abort condition, and before anything is written a person checks the list of decisions and names. All steps are logged. Afterwards you can see in which stage what happened and restart the workflow from there.
How long does it take to get into production?+
A contained workflow for one type of meeting is in production within a few weeks. Writing back into several business systems or loading old minutes takes longer. Because we build stage by stage, you use the first parts early.
From what size does it pay off?+
If recurring meetings happen several times a week and someone spends hours on minutes and distribution, it almost always pays. In the meeting project the effort dropped from 23 hours to 84 minutes per event. With a few meetings a month we advise against it.
Does the agent connect to Microsoft 365 or our database?+
Wherever an interface exists, yes, directly. Microsoft 365 and SQL databases are the normal case. Without an interface we go through file handover or automated operation of the user interface.
Do you use several agents or MCP?+
When it is needed. We advise against starting with several agents, though. Only once one workflow runs through completely does splitting make sense, otherwise nobody knows where it got stuck. We choose connection standards such as MCP once the systems to be connected are known.
07 Read on

Related topics

Next step

Tell us which meeting costs the most follow up work.

Thirty minutes are enough to say whether an agent would carry there.

DEJPEN