Four client processes and one product of our own. One client prefers not to be named. The numbers are real regardless.
Fair Treat, staffing agency, Germany
One job profile in. A campaign 40 seconds later.
Job_profile_roofer.pdf0.0 s
Question setFive questions for the applicant chat
Ad copyThree variants, informal tone
CreativeImage in the Meta campaign format, with logo
Campaign readyThe client approves, the chat takes applicants
Real 40 seconds, shown here in fast forward.
Before
For every position, ad copy, image and applicant questions were written by hand. Applications arrived unsorted by email, and the end client had no idea what was happening with their ad.
Built
An applicant funnel with chat, where candidates pre-qualify themselves in a few questions, a login for the end client who sees only their applications, and a generator: upload or briefly describe a job profile, and out come the question set, ad copy and creative for the Meta campaign. Data protection with a processing agreement, retention periods and EU hosting.
Results
Two campaigns out of the funnel, three weeks each on Meta. An IT manager role brought in 23 applications at €13.56 apiece, a metal production worker role 36 at €7.30. Every one of them had answered five questions in the chat first. The client only sees people who fit the job.
IT manager, 23 applications in 3 weeks€13.56 per application
Metal production, 36 applications in 3 weeks€7.30 per application
Pre-qualified5 questions in chat
Campaign packagein ~40 seconds
SupabaseClaudeReactMeta AdsNetlify
KI-Lian, our own product: AI phone assistant for practices
Kilian answers the practice phone. Books, reschedules, recognises the patient.
Tuesday, every therapist with a patient08:00
08:00Call comes in. Kilian picks up and introduces himself as an AI.
08:01Number recognised: regular patient, shoulder prescription still open.
08:02Free slot with her therapist found, Thursday 9:30.
08:03Appointment in the calendar, summary in the practice area. Nobody picked up a handset.
Demo call, the way it appears in the practice area.
Before
In a practice the phone rings straight into treatment. Whoever has a patient on the table does not answer, and according to Doctolib three out of four calls to practices go unanswered. We know the problem from the family: Jan's brother is a physiotherapist.
Built
KI-Lian, our own product. Kilian takes every call, books, moves and cancels appointments in the practice calendar (Google or Microsoft 365), recognises callers by their number, records requests and calls the waiting list when someone cancels. Plus the practice area: patient records, call overview with summaries, evening report by email, subscription via Stripe, two-factor login, data processing agreement. On the phone Kilian introduces himself as an AI, as the AI Act requires. Voice and data stay in the EU.
Status
The app is running, and practices can test Kilian for 30 days free of charge, no payment method needed. More at ki-lian.ai.
Build time to launch9 days
Call to appointment~2 minutes
Voice and dataEU, Frankfurt
Pricefrom €149 a month
Next.jsSupabaseFamulorClaudeStripeBrevoNetlify
Persomatch, HR tech vendor for recruiting ads, Germany
Every Monday at 8 am, the machine goes looking for new customers.
Two people, a spreadsheet and Google. Who is hiring apprentices right now? Who is responsible for HR there? What is the email address? The searching ate more time than the calling, and the CRM was never up to date.
Built
A lead engine in n8n that starts by itself every Monday at 8 am. It scans career pages, the federal employment agency, Google Maps and photographed job ads, sorts companies by fit and identifies the right contact in five stages, with a verified email. By half past eight everything is in the CRM, every run with a cost sheet in the dashboard. Sales starts calling instead of researching.
Outlook
Next, the engine takes over the outreach itself: it writes the first email to match the advertised position, follows up after two working days and hands over to the CRM as soon as someone replies. Sales then only sees appointments.
Cost per contact$0.09 to $0.12
Photo to lead70 seconds
Sources4
Workflow> 170 steps
n8nAirtableApifyApolloClaudeSupabaseNext.js
Ducati Münsterland, motorcycle dealer
Photograph the ID. Contract done.
Copy the ID, type the contract, sign, file. Ten minutes per test ride.
Photo, signature, key. Two minutes.
Before
Every test ride meant: copy the ID, copy the driving licence, fill in the contract, get it signed, file it. With over 150 test rides in a season that was a job of its own, on exactly the days the shop was full.
Built
A web app at the sales counter: ID and licence onto the tablet, text recognition reads the data and fills in the contract, the customer signs on screen, the PDF lands legally valid in the archive and via API in the existing systems. The salesperson hands over the key.
"Jan turned our manual contract chaos into a digital system we have used every day since. Stable, secure, and something we can operate ourselves."
VossTec GmbH120 staffBRight industry, role still unclear
Nordlicht Logistik240 staffAWhistleblower law applies, no system in place
Ebert Consulting12 staffCConsultant, not an end customer
Lindner Pharma410 staffACompetitor contract is expiring
Hoffmann Bau35 staffCBelow threshold, revisit in 2027
Krämer & Söhne85 staffAAttended the webinar, compliance role
Before
New contacts arrived as a raw list in the inbox. Someone researched company, size, industry and contact person by hand and copied it into HubSpot. Around 40 hours a month, every month, and sales still did not know whom to call first.
Built
An n8n workflow pulls new contacts from HubSpot, enriches them from several sources, scores them against the sales team's criteria and writes everything back, including the reasons for the score. As a second stage, a drafting assistant that prepares reply suggestions for open emails in the evening.
Manual work~40 h a month less
Accuracy~99% correctly matched
Contacts250 a month
HostingEU, own server
n8nHubSpotWeb search enrichmentClaudeDPA and TOMs
What else we have built
When it has to be fast.
Not automations, but the same standard: short paths, clear numbers, finished instead of planned.
Sander Automotive, automotive24hours to a 4K spotA commercial for a car that was not finished yetFrom four 3D renders, a cinematic 4K spot set in Rome, made entirely with AI, without a film crew.
"24 hours later a 4K clip was in the inbox that looks like we sent a film crew to Rome."Michael Sander, owner, Sander Automotive
Ducati Münsterland, automotive+320% ROASGoogle Ads for a dealership openingOver 30,000 people reached, 150+ test rides, live after 7 days.
"Precisely targeted, quickly implemented and measurably effective."Elena Mikolai, Ducati Münsterland
SanderStrothmann Asia, consulting48hours, zero revisionsInvestor pitch, wonFrom strategy to a finished presentation for C-level in Asia. One point of contact, Keynote and Figma.
"He immediately understood our strategy and transformed it into a compelling narrative and design within just 48 hours."Frederick Fong, Managing Partner
D2C skincare, e-commerce< 90days to launchShop, fulfilment and marketing from zeroShopify, payment, warehouse integration, Klaviyo flows, paid engine and analytics, launch-ready in under 90 days.
"Technically highly skilled, reliable in execution, proactive in project management and creative in finding solutions."René Strothmann, entrepreneur & investor
Next step
Which process is costing you the most right now?
60 minutes, free, remote or on site. You leave with a prioritised list.
