Craft
Services · AI automation
AI automation with approval: hand over routine work, keep the decisions.
We automate routine work for small and mid-sized businesses and practices, including with AI: sorting enquiries, reading documents, writing drafts. Every workflow has a checkpoint: anything unclear goes to a person, and anything that goes out is approved by a person first. We agree with you in advance which data an AI service receives.
What you get
- An honest assessment of which routine work AI can reliably prepare, and where it would be simpler without AI.
- Automated workflows for incoming requests, documents, drafts or internal knowledge, connected to the systems you already use.
- A checkpoint in every workflow: anything unclear stops, and a person checks and approves whatever goes to your customers.
- Agreed in advance: which data an AI service receives and where it processes it.
- Before-and-after measurements, for example of time per case, manual steps and follow-up questions.
Automation with judgement
Jidoka
automation with a human touch
AI with a checkpoint
The term comes from the Toyota Production System: the machine detects for itself when something is wrong, stops and calls a person. That is exactly how we build automation with AI.
Who this is for
Overflowing inboxes
Enquiries arrive by email and form, and are sorted and retyped by hand. AI assigns them, extracts the details and flags what is missing.
Repetitive writing
Replies, quotes or summaries are much the same from case to case. AI writes drafts; a person checks and approves them.
Scattered knowledge
Answers are buried in manuals, old emails and the heads of a few colleagues. AI makes internal knowledge easy to find, so not every question lands with the same person.
Sound familiar? Tell us about it
The way
How we introduce AI
Look at the routines
Using real cases, we look at which work repeats itself and where it costs time, on site or via screen sharing.
Set the rules
Together we define what AI takes on, what stays with people, which data a service receives and where it processes it. You receive a proposal in which every item is easy to follow.
Start small, with a checkpoint
We automate one workflow first, with approval by a person. You try it out in day-to-day work before the next one is added.
Measure and improve
Before and after, we measure the same things, such as how long a case takes and how often someone has to follow up. What works becomes the new standard.
Try it: AI with a checkpoint
Example with fictitious dataThree sample enquiries come in. The AI assigns them, extracts the details and drafts replies. One enquiry is unclear: it stops and waits for you. Nothing goes out until you approve it.
What this means for you: Routine work such as sorting and drafting runs by itself, but nothing leaves your business before a person has approved it.
Example with fictitious data, recreated in the browser. No AI service is connected and nothing is sent.
What this sample shows
- Three made-up enquiries arrive: “Reschedule appointment”, “Attachment, no text” and “Question about a quote”.
- The AI sorts each enquiry, extracts the details and writes a draft reply.
- At the checkpoint the flow stops. Only when a person presses “Approve” does the draft count as approved; then the next enquiry follows.
- “Attachment, no text” is unclear: it stops at sorting and goes to a person, without a draft.
- Simulated in the browser: no AI service is connected and nothing is sent.
Could this fit one of your processes? Request a free intro call
Our rules for AI
These rules apply to every automation we build.
What AI can take on
- Sorting and preparing incoming requests
- Extracting details from documents
- Drafting replies and quotes
- Making internal knowledge easy to find
What stays with people
- Checking and approving what AI writes
- Decisions on individual cases
- Responsibility for the result
What we clarify first
- Which data an AI service really needs
- Where that data is processed
- Whether it would be simpler without AI
Technology and data
- n8n for automated workflows between your systems
- TypeScript and Node.js for custom logic and validation rules
- We choose the AI service together with you, taking into account where it processes data
- Only the data a step really needs; with particularly sensitive data, we check especially carefully whether AI is a good fit at all
- If we operate the workflows for you, they run in data centres in Germany or elsewhere in the EU; where an AI service processes data is agreed in advance
Frequently asked questions
Does the AI make decisions on its own?
No. AI does the groundwork: it sorts, extracts and writes drafts. Decisions on individual cases and approval of whatever goes out stay with people, as does responsibility for the result.
Which data does the AI service receive?
Only what a step really needs. We agree this with you in advance, along with where the service processes that data.
What happens if the AI gets it wrong?
That is what the checkpoint is for. Unclear cases go to a person, and nothing goes to your customers before someone has checked and approved it. We use whatever the checks reveal to improve the workflow.
Is AI worth it for us at all?
That is the first thing we clarify. Sometimes a clear template, a better form or a different order of steps solves the problem more simply. If so, we tell you.
What does AI automation cost?
That depends on the workflow and scope. We start with one workflow. After the intro call, you receive a written proposal in which every item is easy to follow.
Can we start with consulting?
Yes. We work out with you which routine work AI can reliably prepare and rank the options by effort and benefit. Whether anything is built afterwards, and what, is up to you.
Read on: Where AI has to stop in a process
Our article explains at which points an AI-assisted process has to stop and how to define them, with a simplified example.
Read the articleConnection
Which routine work would you like to hand over?
Describe a process that takes too much time today. In the free, no-obligation intro call, we work out whether AI reliably helps there or whether there is a simpler way. We usually reply within two working days.
Arigatō – in advance.