# About dropd. Source: https://dropdigital.de/en/ueber-uns Language: en Site: DropD., https://dropdigital.de Sections on this page: - #geschichte, Wie dropd entstand: Die Entstehung im laufenden Betrieb eines Mittelständlers. - #gruender, Die Gründer: Lucas Koch und Can Tillmann mit Hintergrund und Kennzahlen. - #officers, Operating Agent Officers: Lena und die Officers: Mandat, Modell, Betrieb, Grenze. Zwei Menschen bleiben verantwortlich. - #mission, Wofür wir stehen: Der Leitsatz: KI, die im Unternehmen läuft, nicht auf der Demo. - #haltung, Haltung: Echtbetrieb, Herstellerneutralität, Datenhoheit zuerst. - #cases, Kennzahlen: Die Zahlen hinter dem Ansatz. - #stack, Stack: Modelle, Plattformen und Werkzeuge, herstellerneutral. - #holding, Muttergesellschaft: DropDigital GmbH, ein Unternehmen der AllSet-Ventures AG. ## About dropd. Lucas built AI in a mid-market firm, Can scaled brands. Based in Ratingen, An der Pönt 44. AI consulting from 20 staff. DropDigital GmbH, a company of AllSet-Ventures AG. Lucas built AI in a mid-market firm, Can scaled brands. Based in Ratingen, An der Pönt 44. AI consulting from 20 staff. DropDigital GmbH, a company of AllSet-Ventures AG. About us Two practitioners. Not slide-deck founders. DropD. did not come out of a seminar. It came out of live operations. Lucas built AI in a mid-market company, Can scaled brands and processes. Together we build AI that runs in the company, not on demos. How DropD. started Not in a seminar room. In the engine room. DropD. is not a consulting startup that read up on AI. The approach grew in the day-to-day of a mid-market company, over years, across every department, with real numbers and real mistakes. It started in a real company. Lucas was head of marketing at a mid-market company with 50+ employees and €10 million in revenue. From 2022 he put AI in place there, not as a pilot for the board slide, but across marketing, sales, logistics and purchasing. /lucas-whiteboard.jpg Projects became a system. What worked stayed. What only shone in the demo went out. From dozens of real implementations came a repeatable system, distilled from operations, not from a course. /lucas-beratung.jpg Then the second practitioner arrived. In parallel, Can had built his own brands from launch to seven-figure revenue: marketing, full funnel, and the AI architecture behind it. Together that became DropDigital: a firm for AI that runs in the company. A company of AllSet-Ventures AG. /founders-duo.png What we stand for We build AI that runs in the company, not on the demo. Vendor-neutral, from live operations, with data sovereignty first. From live operations, not from demos. DropD. comes from a running company with 50+ employees, not from consultant slides. What we recommend, we have running in operations ourselves. Vendor-neutral, with no commission agenda. We do not sell a tool and we do not take a commission. We solve the problem, with the tool that fits the company. Data sovereignty first. Where it matters, we develop in-house AI: adapted open-source models, GDPR-compliant, hosted in Germany or on-premise. Your data stays your data. Part of AllSet-Ventures AG DropDigital GmbH is a company of AllSet-Ventures AG. Managing directors: Lucas Koch and Can Tillmann. Office An der Pönt 44, 40885 Ratingen. The GmbH commercial register number will follow. The AG: District Court Dusseldorf, HRB 112775. allsetventures.de Where we sit Ratingen, Rhineland. Not Silicon Valley. In the middle of the Düsseldorf economic region, where the mid-market actually works. We come on site, work by video, or host on-premise at your company. Link: https://dropdigital.de/en/ueber-uns#geschichte Id: page-ueber-uns ## Lucas Koch, Founder, DropD. Founder, DropD. Lucas Koch Founder, DropD. Former head of marketing at a mid-market company with 50+ employees and €10 million in revenue. Since 2022, strategic AI implementation across every department: marketing, sales, logistics, purchasing. DropD. does not only work with ready-made tools. It develops its own AI tailored to the company, hosted in Germany in a GDPR-compliant way on request. The system is distilled from real projects, not from slide decks. 50+ Employees in the first pilot, birthplace of the DropD. system €10 million+ Revenue environment with real AI implementations up to 80% less time spent on formerly manual processes Link: https://dropdigital.de/en/ueber-uns#gruender Id: person-founder ## Can Tillmann, Co-founder, DropD. Co-founder, DropD. Can Tillmann Co-founder, DropD. Architecture studies, then the switch into digital. First his own shop, then scaling several brands from five-figure to seven-figure revenue. His part at DropD.: marketing, full funnel, and the AI architecture behind it. Processes you do not rebuild every month. Processes that run. 0 → 7 figures Brand building from launch to seven-figure revenue 5 → 7 figures Existing brands scaled to seven-figure revenue Full funnel + AI Marketing, processes and AI architecture from one hand Link: https://dropdigital.de/en/ueber-uns#gruender Id: person-co-founder ## Operating Agent Officers We run dropd. with the same kind of agents we build for customers. Each role has a mandate, a model and a boundary. The portraits are personas, not staff. Signing is Lucas and Can. Lena CXO Chief eXternal Officer Writes in this chat. Books a slot when you want one. Answers what the page actually says. Does not invent prices. Around the clock. No pause, no holiday. The voice facing out. The chat on this site is her surface. Lena answers what dropdigital.de actually says: offers, process, data sovereignty, who is liable. She leads the conversation and opens the matching surface, a booking, an enquiry. You submit, she does not. She knows the public site corpus, not your files. She does not invent prices. Around the clock, no pause, no holiday, and no authority to trigger anything binding. Site chat: questions from the public corpus, in the language of the page. Propose the next step: a consultation, an enquiry, a concrete slot. Open surfaces where you submit yourself. Mail, booking, consent. Hand off cleanly to Lucas when the question leaves the public corpus. Hybrid Not hard-wired. Behind Lena sits our own OpenAI-compatible interface: local via Ollama (open-weight, Qwen-class) or a cloud API, depending on the environment. Switching is configuration, not a rebuild, and not a vendor SDK. The public site corpus only. No customer data, no internal files, no training on the conversation. Does not submit anything that costs money or goes outside. Does not invent prices, promises or legal advice. No access to DATEV, contracts or internal handoffs. That is Mira, Nora, Vera. Mira COO Chief Operating Officer Lena's counterpart, inside the house. Keeps what was decided: handoffs, briefings, the current state. So the knowledge does not live in one head. Runs 24/7, including when nobody is in the office. Lena's counterpart inside the house. The state, the handoff, the briefing. Mira does not speak to visitors. She keeps what was decided internally, so the knowledge does not live in one head when Lucas or Can are not in the room. Handoffs, briefings, the current state of a matter. She runs when nobody is in the office. She does not decide, she writes it down. Releases stay with the two who are liable. Write down handoffs and briefings as soon as something is decided. Keep the state of a matter current, not the story around it. Place internal knowledge so it stays findable if someone is out. Put the next internal step in front of Lena or a human. Do not execute it. Local Open-weight via Ollama on our own infrastructure, Qwen-class. Mira sees internal notes, so no US cloud and no third-party training. Handoffs, briefings, the current state. Internal, not public, not in the site chat. No customer contact. Facing out is Lena, or a human. No release, no send, no booking in dropd's name. Does not store visitor chats. Lena's log stays Lena's log. Nora CCO Chief Controlling Officer Reads DATEV and the ERP before anyone has to ask. Marks the deviation. Not the essay about it. One page on the table. No dashboard nobody opens. Does not approve. The number waits for a human. Reads DATEV and the ERP before anyone has to ask. Marks the deviation. Nora does not do arithmetic inside the language model. The number comes from DATEV or the ERP, the model writes the deviation onto one page a human can read. No dashboard nobody opens. No approval: an invoice, a budget move, a payment waits for a human. That is why she runs locally or on a server in Germany, not in a US cloud where financial data has no business being. Read extracts from DATEV and the ERP once they are connected. Mark the deviation: actual against plan, without the essay. Put one page on the table that a human can use in a meeting. Prepare the question, do not decide. The number waits. Local Open-weight locally (Qwen or Llama class via Ollama). The model writes the page. DATEV does the arithmetic. A cloud API does not see these numbers. Financial extracts and internal control figures. In-house or on a server in Germany. No training, no outflow. No approval, no payment, no booking in the ERP. Does not invent numbers when the source is missing. She says so. No tax advice and no disclosure to third parties. Vera CLO Chief Legal Officer Reads contracts before anything goes live. DPA, TOMs, AI Act dossier. Prepares, does not sign. Knows the line between engineering and legal counsel. No clocking off, because deadlines do not either. Reads contracts before anything goes live. Prepares, does not sign. Vera is not a lawyer and does not give legal advice. She reads what is about to go live: DPA, TOMs, the AI Act dossier covering purpose, data types, model, autonomy and intervention points. She puts the gaps on the table so a human or counsel can decide. She knows the line between engineering and legal advice, and she does not cross it. Deadlines do not clock off, neither does she. Signing is Lucas and Can, or your counsel. Read contracts and annexes before a system goes into operation. Prepare DPA, TOMs and the AI Act dossier: purpose, data, model, logs, retention. Mark the gap, do not close it. Classification stays with counsel. Hold the line: prepare the engineering, do not replace legal advice. Local Open-weight locally via Ollama. Contracts and dossiers are not prompt fodder for a US API. Structure and check, not a signature. Draft contracts, DPA, TOMs, AI Act files. Local or Germany. No training on your contracts. No legal advice, no signature, no representation before authorities. No classification under the EU AI Act. That is your counsel. No go-live without a human release, even if the dossier looks complete. Jule CKO Chief Knowledge Officer Writes the knowledge section and the AI radar. Researches sources, rulings, numbers. Every line with a citation. Publishes nothing without Lucas signing off. Runs 24/7 so the picture is never older than the week. Writes the knowledge section and the AI radar. Every line with a citation, none without sign-off. Jule writes the articles under /wissen and the AI radar: how-tos, documented use cases, the weekly state of AI and the law. She researches sources, rulings and numbers and puts the citation next to the sentence. Vera reads the legal pieces. Nothing is published that Lucas has not signed off; he is accountable for the content under section 18 (2) MStV. She replaces no advice and writes no references that do not exist. Draft articles for /wissen and the AI radar, with a source on every number. Keep existing articles current when the law or the models change. Turn use cases and project learnings into readable steps. Submit every draft for sign-off, never publish herself. Hybrid Research and drafting through a frontier model via API, with no customer data in the prompt. Polish and alignment with the knowledge base locally. Public sources, anonymised project notes, the site's knowledge base. No customer data, no form enquiries. No publishing without human sign-off. No invented customers, numbers or quotes. What has no source stays out. No legal advice. The legal reading comes from Vera and from counsel. Kai CISO Chief Information Security Officer Access, logs, what may leave. Checks before anyone connects. Does not sleep. Attacks do not either. Reports. Does not reach into your operation on its own. Access, logs, what may leave. Reports. Does not reach in on its own. Kai checks before something is connected: who may access what, what may leave the house, what stays in the log. The actual access logic is code and policy, not just a prompt. The model helps read signals, it does not decide whether your operation gets rebuilt. He does not sleep, attacks do not either. He reports internally. He does not reach into your operation, and he does not open a foreign cloud to judge whether something may leave. Hold access and permissions against policy before anything connects. Read the logs: what tried to leave, what looks off, what must stay in-house. Report, internally and in a way that can be reconstructed. No silent pass. Keep least privilege and logs so a human can reconstruct the incident. Local A smaller open-weight model locally, plus rules in code. Judgement on access is policy, not model mood. No foreign cloud for logs. Access and log data in-house. No outflow, no training on your logs. No intervention in your operation, no lock-out, no patch without a human. No pentests against third parties, no surveillance of people. Does not hand logs to a US vendor so they can be analysed there. Ada CFE Chief Frontend Engineer Writes the frontend. Components, states, what someone actually touches. Ships surfaces, does not speculate about your business. On the code 24/7, without sprint theatre. Hands off to a human when taste has to be decided. Writes the frontend. Components, states, what someone actually touches. Ada ships surfaces, not a business strategy. Components, states, what a human actually touches on the page. She writes around the clock, without sprint theatre. Taste, brand and the last merge stay with a human. Routine work runs on a local open-weight model. When the surface is awkward, a frontier model may help. Nothing ships until someone has read the diff. Frontend of the site and internal surfaces: components, states, accessibility. Ship what someone touches. Not what looks good on a slide. Hand off to a human when taste, brand or a breaking change has to be decided. No speculation about your business. Ada knows the code, not your margin. Hybrid Code work: Qwen or DeepSeek locally via Ollama for routine. Frontier (Claude or GPT class) only when the surface warrants it. Merge and deploy by a human. Interface code, no customer data, no DATEV extracts. The diff stays in-house until someone releases it. No merge, no deploy, no go-live without review. No decision on brand, tone or price. No access to financial or contract data. That is Nora and Vera. Beck CBE Chief Backend Engineer Writes the backend. APIs, data, the server in Germany. Holds up what Ada shows. No service that falls over at night because nobody is there. Changes nothing in your system until a human releases it. Writes the backend. APIs, data, the server in Germany. Beck holds up what Ada shows: APIs, data, the service that does not fall over at night because nobody is there. The server sits in Germany, not in some arbitrary US-cloud region. He changes nothing in your system until a human releases it. Like Ada, he writes with a model, local or frontier depending on the task. Executing, merging, putting it on the server is a human. Logs and data stay where the contract put them. Backend, APIs, storage, the service on the server in Germany. Hold up what the surface promises. No endpoint that lies still at night. Build connections (calendar, mail, CRM) so a human has the last click. Change nothing in a customer system until someone releases it. Germany Same as Ada for the code: local open-weight via Ollama, frontier only where the task demands it. Where the data runs is the server in Germany or on-premise, not the model. APIs, operations, logs. In-house or in Germany. No training on your stores, no silent outflow. No schema change, no deploy, no deletion without a release. No reach into your ERP, DATEV or mail just because a prompt suggested it. No hosting in a US region just because it would be cheaper. OAO · Operating Agent Officers Eight roles, no clocking off. We run dropd. with the same kind of agents we build for customers. Each role has a mandate, a model and a boundary. The portraits are personas, not staff. Signing is Lucas and Can. Lucas Koch and Can Tillmann remain liable. The officers execute. The two are on the hook. The two who are liable Responsible for Model and runtime Where it runs Model Data Does not Speaks on the site Internal Talk to Lena Link: https://dropdigital.de/en/ueber-uns#officers Id: page-ueber-officers ## Stack and vendor neutrality We are vendor-neutral and model-agnostic: the best tool for each task. From local open-source models (GDPR-compliant in Germany) to the strongest frontier models. Plus the platforms mid-market companies actually run on: Microsoft, Google, AWS, and end-to-end automation. Our stack Vendor-neutral: from Llama to Claude, on your server or in Germany. Which models and tools No favorite tool, the whole toolbox. We are vendor-neutral and model-agnostic: the best tool for each task. From local open-source models (GDPR-compliant in Germany) to the strongest frontier models. Plus the platforms mid-market companies actually run on: Microsoft, Google, AWS, and end-to-end automation. In-house development Our core: tailored AI. Custom agents, open-source models adapted to your company, and RAG on your own data, deployable in the cloud, hosted in Germany, or fully on-premise. Open source & local models Llama Mistral Qwen Gemma DeepSeek GLM Kimi MiniMax Hunyuan Command R Frontier models GPT Claude Gemini Grok Runtimes & clouds Ollama Hugging Face AWS Azure Google Cloud Microsoft & automation Microsoft Microsoft 365 Power BI Power Automate Power Platform Copilot n8n Make Development & orchestration Next.js React TypeScript Tailwind Python Node.js LangChain Data & operations Supabase PostgreSQL BigQuery Docker Hetzner Link: https://dropdigital.de/en/ueber-uns#stack Id: section-stack ## Key figures The numbers behind the reference project. 50 + Custom AI solutions in live operation 80 up to % less time spent on manual processes 12 + Mid-market companies with real AI implementations 11 Partner companies from SME to enterprise Link: https://dropdigital.de/en/ueber-uns#cases Id: fact-kennzahlen