Understanding the request, drafting the plan
The agent takes a request in plain language by email, form or chat. It turns that into a concrete plan with individual steps. You see the plan before anything happens.
AI agents & automationAI agents that plan,
AI agents that plan,
ask, and only then act.
An agent reads a request, turns it into a plan and carries it out across your systems. Before it changes anything, a person signs off.
Book a free AI opportunity review
45 minutes, no strings attached. You leave with a written shortlist.
What it is about
AI agents & automation
A lot of everyday work is a chain of small steps in different tools: read an email, look something up in the ERP, create a record, send a reply. An AI agent handles that chain. It reads the request, works out which tools it needs and proposes an execution plan.
The model is not the hard part. The frame around it is. Which systems may the agent touch, with which permissions, and where does a person have to agree? We build agents so that every step is logged and anything that writes data waits for approval.
Not every automation needs AI. Where a process runs the same way every time, rules and conventional workflows are cheaper, faster and easier to audit. We will tell you plainly which approach fits your process.
What we build
The building blocks
Human approval
The agent may run read-only steps on its own; anything that writes needs sign-off. You decide who approves and at what value or risk a second person is required. Responsibility stays where it belongs.
Tool access through APIs
The agent works through defined interfaces to your ERP, CRM, ticketing system, mailbox or database. Each tool has a clear description and hard limits. If a system is not connected, the agent cannot reach it.
A log of every step
Input, plan, approval, executed action and result are all recorded. You can always see why the agent did what it did. That helps with errors, with questions from colleagues and with reviews by your data protection officer.
Permissions, limits and fallback
The agent only gets the permissions its task requires. When data is missing or it is unsure, it hands over to a person instead of guessing. Fixed rules and conventional workflows take over wherever a process is unambiguous.
Approach
From the first conversation to operations
- 01
Free AI opportunity review
In a no-obligation first conversation we walk through your processes and check whether an agent makes sense or a simple rule will do. You get one page with two or three starting points.
- 02
Prototype on a real process
We build the agent for one clearly scoped process and test it with real data, masked where needed. You see early on where it works reliably and where it does not.
- 03
Delivery at a fixed price per package
The production version is delivered under a fixed-scope contract with a fixed price and acceptance criteria agreed in advance. Approval step, permission model and logging are part of it.
- 04
Operation and further development
We run the agent, track error rates and add further tools once the first process is stable. The code and documentation belong to you.
By industry
Where this applies in practice
The industry overview lists concrete use cases. Detailed pages are available in German.
- Machinery and plant engineeringUse cases: Quotation and configuration support · Automating spare parts requests
- Logistics and freight forwardingUse cases: Order entry from emails · Answering shipment status requests
- Skilled tradesUse cases: Quotes from site measurements and notes · Qualifying incoming requests · Supplier invoices and scheduling
- Tax advisory and law firmsUse cases: Preparing deadline and tax assessment checks · Supporting client communication
- Medical practices and outpatient centresUse cases: Inbox triage for prescriptions, appointments and requests
- Real estate and property managementUse cases: Tenant requests and ticketing · Service charge receipts and documents
- Retail and e-commerceUse cases: Customer service assistant
- Industry and manufacturingUse cases: Preparing complaints and 8D reports
- Utilities and municipal energy providersUse cases: Customer service assistant for instalments, meter readings and moves · Pre-sorting market communication exceptions
- Insurance brokersUse cases: Capturing claims notifications
- Construction and architectureUse cases: Documenting change orders
- Staffing agenciesUse cases: Preparing candidate communication
- Hotels and hospitalityUse cases: Turning group and conference requests into offers · Multilingual guest communication
- Agriculture and foodUse cases: Order intake from emails and faxes
Related services
What often comes with it
- RAG knowledge assistantYour staff ask questions in plain language and get answers drawn from your own documents. Every answer points to the passage it came from.
- Document AIWe automate your incoming mail: documents are recognised, read, checked, and passed to your systems. Uncertain cases go to a person for approval.
- Custom softwareSome processes do not fit any off-the-shelf product. We build them as a web application your team actually likes using and that you can still maintain in five years.
From a project
From a requirements list to a business model
The client wanted an internal tool to run their ERP instances. What emerged was a white-label platform with a licence model, including an AI agent with zero-data-retention architecture.
FAQ
Questions about AI agents & automation
What is an AI agent in a business context?
An AI agent is software that takes a task in plain language, plans the necessary steps itself and carries them out through connected systems. Unlike a chatbot it does not just answer, it acts, for example by creating a record or checking an order.
When is conventional automation better than an AI agent?
Whenever the process is fixed and the inputs are structured. Rules and workflows are then cheaper, faster and easier to audit. An agent pays off when requests are phrased differently each time, information comes from several sources, or the route to the result varies.
Can an AI agent act in our systems without oversight?
Not the way we build them. The agent may run read-only steps itself; anything that changes data or leaves the company needs human approval first. You decide which actions that covers and who may approve.
Which systems can the agent work with?
Any system that has an API or can be given one, such as ERP, CRM, ticketing, email or a database. For each system we define which actions are allowed. We do not work with Microsoft Dynamics, Power Platform or SAP.
How much does it cost to build an AI agent?
It depends on how many systems are connected and how complex the approvals are. After the free AI opportunity review and a short concept you receive a fixed price per package, so you know the cost before the project starts.
How can we see what the agent actually did?
Every step is logged: request, plan, approval, executed action and result. The log is readable for your business teams and serves as the basis for reviews under the GDPR and the EU AI Act.
AI agents & automation: let's find out in 45 minutes whether it pays off for you.
Book a free AI opportunity review
45 minutes, no strings attached. You leave with a written shortlist.