The same questions, every day
Opening hours, delivery status, invoice copies, rescheduling. Important to the customer, but not work that fulfills anyone.
Service 05 — AI Agents
An assistant that knows your systems and takes on real work.
An AI agent isn't a chatbot reciting canned text. It has access to your real data, may perform defined actions — and hands over to a person the moment it reaches its limit. Drawing that limit properly is the actual work.
Who it's forFor companies with recurring text- or document-based inquiries where the answer can be derived from knowledge you already hold.
Starting point
Where it usually goes wrong
Off-the-shelf tools don't know your business
A chatbot without access to your data can only answer in generalities. Customers notice immediately — and it annoys them.
Worry about wrong answers
Justified. Without clear limits and logging, an agent is a risk rather than a relief.
Demonstration
A question answered — handover included
The agent answers from your own material and says where it got it from. Where it cannot go further, it hands over to a person.
- 01
The question comes in
Through the website, the inbox or the internal portal — in whatever language it is asked.
- 02
It searches your material
Your manuals, price lists and records. Not half the internet.
- 03
The answer names its source
Every statement has a reference. Anyone who wants to look it up can.
- 04
Handover to a person
With anything unclear, any complaint or contract question the agent stops — and passes the whole conversation along.
Scope
What you get
- Sharpening the use case
- We define exactly what the agent may and may not do. A narrow, dependable agent beats a broad, unreliable one.
- Connection to your data
- Product data, prices, availability, documents. The agent answers from your records, not from training knowledge.
- Defined actions
- Book an appointment, open a ticket, schedule a callback, send a document — each action approved individually.
- Handover to people
- On uncertainty, complaints or anything outside the frame, the agent hands over — with the full conversation.
- Complete logging
- Every answer and action is traceable. You can check at any time what the agent based its reply on.
- Test phase with real inquiries
- Before launch the agent answers in shadow mode. You read along and correct before anyone else sees it.
- Data protection settled up front
- Which data leaves your building and which doesn't. On request with models running on your own infrastructure.
- Training for your team
- Your people learn how to correct the agent and when they should take over.
Process
How it runs
- 01
Define the use case
Which inquiries, which data sources, which actions. And explicitly: which not.
- 02
Build & connect
Build the agent, wire it into your systems, set the limits and escalation paths.
- 03
Shadow mode
The agent answers, visible internally only. You check quality against real inquiries.
- 04
Launch & care
Switch on, observe, refine. The first weeks decide the quality.
Walkable demos
Don’t describe it — look at it
Invented businesses, real interaction. Click your way through before you ask what something like this costs.
The businesses in the demos are invented. They are not client references.
Price
What it costs
An assistant wired into your systems that takes on real work.
Fixed price
live in roughly 14 days
- Sharpening and bounding the use case
- Build using your data and rules
- Connection to email, CRM or calendar
- Test phase with real inquiries
- Escalation to a human where needed
- Logging, so what happens stays traceable
- Training for your team
Questions
What gets asked about this most
How do I know whether AI makes sense for me at all?
A good indicator: is there a task someone performs the same way several times a week, based on text or documents? Then it's worth a look. If there isn't, we'll say so — an AI agent solving a problem you don't have is just an expense.
How do you handle GDPR and tracking?
We use no third-party tracking services. This website is the example: it sets no consent-requiring cookies. That is why the notice on your first visit asks nothing about statistics or advertising — it only records your own decision, and declining there is exactly as easy as accepting. If you need a specific tool that works differently, we tell you the legal implications up front instead of just wiring it in.
How long does a project take?
A company website typically runs three to five weeks, an app six to ten weeks plus the store review, a single automated workflow one to two weeks, an AI agent two to four weeks. The biggest variable isn't on our side — it's how quickly copy, images and access credentials arrive. The proposal states exactly what we need from you and when.
What happens after launch?
Your choice. Either you take over — code, content and all credentials are fully yours, with no dependency on us. Or you take the retainer: updates, monitoring, security patches and ongoing development at a fixed monthly price, cancellable monthly. No annual contract, no notice period.
Other services
A handful of building blocks. One system.
You'll hear back within 24 hours