AI Traction ExamplesDEBack to the home page

Portfolio

Built and running

Here are some examples of systems we built and what value they brought. Seven are in use, one is in testing.

50new companies every morning, without anyone searching
10agents work at night, one keeps watch
4 sinstead of 79 seconds per voice note
  1. 01New clients from public signals50 new companies every morningrunning
  2. 02Ten agents and one that keeps watchFailures visible the same morningrunning
  3. 03The idea from the drive, at the deskMore published, every word her ownrunning
  4. 04The enquiry that goes quietNo email without approvalin testing
  5. 05Voice note in, task out4 seconds instead of 79 per noterunning
  6. 06Recording in, video outVideos without a video editorrunning
  7. 07Invoicing without a subscriptionE-invoices without a subscriptionrunning
  8. 08The day starts with a planA schedule instead of the inboxrunning

About the pictures. We recreated the views beside the text using sample data instead of real names. We write the rules and instructions behind them afresh for each business, so they fit the way it works.

01 / 08 · New clients

Every night since September 2026

New clients from public signals

What it brought

Every morning a short list of companies where work is piling up right now, with the right contact. Nobody has to search by hand any more.

2,649signals in one night
50reported the next morning
0AI calls for the search
Before

Contacts came from search filters any competitor could use. Those filters didn't tell us whether a company had a problem we could solve right now.

Today

Every night, a radar looks for companies with signs that sales work is piling up, such as an inside-sales position that has been open for months. It scores each company and finds the managing director. The morning report contains only new companies whose scores pass the threshold.

What stays with a personThe conversation. When someone replies, we write back ourselves.

Applied to your business: Every Monday, a wholesaler sees which regular customers haven't ordered for six months.

Radar · 06:3050 new
4/5Inside-sales role open for five months · managing director found
4/5New signal since yesterday · head of sales found
4/5Two signals at once · managing director found
+ 47 more

02 / 08 · Operations and safety

Around the clock since September 2026

Ten agents and one that keeps watch

What it brought

Ten agents work at night without anyone watching. If a job fails, it shows up on the phone the same morning, not days later.

10agents
16scheduled jobs
10 %weekly sample
Before

As more work runs on its own at night, checking it gets harder. A job can stop overnight and go unnoticed for days. And blindly trusting an agent that writes to the outside world is a bad idea.

Today

A coordinator on our own server runs ten agents and sixteen scheduled jobs. Each job reports what it saw and did after every run. A rule guard checks the work against the shared rulebook every night and pauses any area that breaks a rule. It also checks a ten per cent sample of all autonomous actions each week. Alerts go to the phone, and there is an emergency stop.

What stays with a personEvery decision. An agent that needs a decision submits it and waits. Its own recommendation doesn't count.

For you: The same supervision runs on your own server. Nobody can reach it from outside.

Telegram · Alerts00:52
Rule guard: area “Contact requests” paused. Reason: daily limit reached, one more action was planned.Release with /resume · details in the log
Telegram · Status06:30
Quiet night. 15 of 16 jobs green, one paused (see alerts). Two decisions are waiting for you.
how’s it going?

03 / 08 · Ideas and writing

In use since September 2026Coach, soloRuns on the client’s own server

The idea that comes while driving is still there at the desk

What it brought

No idea gets lost any more, and she publishes more, without a single line written by AI.

“What it has helped me do is increase the quality of my output without AI-ing my words.”

The client, two weeks after the start
Before

A coach runs meditations and retreats. Ideas came to her on the move, in a mix of German and English, then disappeared before she reached her desk. Her newsletter went out every few months. Post ideas piled up.

Today

She records an idea as a Telegram voice message. Her own server transcribes it and keeps the recording there. A single message becomes separate idea cards in her collection. When she sits down to write, the system brings up matching ideas and asks her questions about them.

What stays with a personEvery word. The system doesn’t write a single sentence for her. That was her condition. It became a rule for how we build.

Applied to your business: After a customer visit, the field rep speaks three sentences into the phone. The visit report is in the system the next morning, in his own words.

TelegramIdeas
▶1:12
Three ideas filed.Transcribed on her own server · wording unchanged
An evening without screens as an exerciseNewsletter · today
What silence does to decisionsPost · today
A question for the next group sessionGroup · today

04 / 08 · Follow-up

Analysis done · prototype in testingTechnical manufacturer, B2B

The enquiry that goes quiet

What the test shows

The system is still in testing. In the last test run, no email went out without approval and no fact was invented.

0emails sent without approval in the test run
0invented facts
14test situations
Before

Enquiries arrive and get answered quickly. Then the conversation stops. Following up means piecing together what the project was about, so it rarely happens. Many older projects have gone years without contact.

Designed

The system reads the inbox and prepares a follow-up draft when a project goes quiet, picking up the earlier conversation. Each morning, the managing director sees what’s due and either approves the draft with one click or edits it. The work started with a paid process analysis at his own screen. He then got a sketch of the proposed system that he could click through before spending a single euro on the build.

What stays with a personHe approves every email before it goes out. And it sounds like him, not like a newsletter.

How we test before building: A simulated managing director works with the prototype. A second model grades every run. All the test data is invented.

Due todaythree items
Line 2 equipment, quote from 12 SeptQuiet for 21 days · last question: lead time for a special part
Good morning Mr Example, the question about the special part is settled: six weeks from order. Does that still fit your schedule for line 2?Draft · refers to the email from 12 Sept
ApproveEditLater

05 / 08 · Ideas and writing

Since September 2026

Voice note in, task out

4 seconds instead of 79 per note, and no idea stays stuck on the phone.

We send thoughts that come on the move to our server as voice messages. The server transcribes each note, separates the ideas and files them with the wording intact. The coach’s system above grew out of this one.

A short note that names its own topic needs no AI model at all.

06 / 08 · Content

Since July 2026

Recording in, video out

The videos on our channel are made without a video editor.

We put a recording into a folder. The system cuts pauses and slips, adds subtitles and produces an edited video with a choice of thumbnails. That's how we make the videos on our channel.

What stays with a person: making the recording and choosing the thumbnail.

07 / 08 · Admin

Since October 2026

Invoicing without a subscription

E-invoices in the ZUGFeRD format, without a subscription to accounting software.

We use our own software to write quotes and invoices. Invoices go out as e-invoices in the German ZUGFeRD format. We photograph receipts on the phone, and the software calculates the VAT return.

What stays with a person: The AI prepares drafts. Finalising and sending is done by hand.

08 / 08 · Daily planning

Every working day

The day starts with a plan

Each morning starts with a schedule, and nothing done in parallel work sessions gets lost.

Each morning, the system reads yesterday’s journal entry and the open deadlines, adds any new voice notes and suggests a schedule with times. In the evening, it gathers what happened across all open work sessions and writes the journal entry. If that entry is missing at 9 pm, the server writes a draft.

What stays with a person: deciding what matters today.

Which of these would take care of work that gets left undone in your business today? The first thirty minutes are on us.

Book 30 minutes