# AI workshop for fiduciary offices Do you already use AI and want to apply it to your mandate work? Bring a process from your office, such as checking documents, preparing a mandate review or drafting a client question. Learn to set data rules and approvals, describe the process in an interview and build a useful Skill. Finally, work in teams on a fiduciary app interface in Lovable. ## 1. Use AI safely 09:00 to 10:00 A short introduction: which data may you use, where is it processed, and which decisions stay with you? ## 2. Describe your fiduciary process 10:00 to 11:45 Interview each other using four questions: the current process, the problem, the desired process and a good result. The questions come from the workshop with Remo Guntli. ## 3. Create and use fiduciary Skills 11:45 to 14:45 Use the Skill Generator to turn your process description into a reusable instruction. Run it, then check it against a second case. ## 4. Build a fiduciary app? Try it together 14:45 to 17:00 In teams, use Lovable to build a first interface, such as a mandate board with an AI dialogue and client messages. Then discuss its value, data needs and the next step. ## Interview sheet Process / owner / frequency / current time: ### 1 · How does it work today? - Who does what, in which order and with which files? - How often does it happen and how long does it take? - Which decisions change the next step? Answer: ### 2 · Where does it get stuck? - What takes time or has to be done more than once? - Where do errors occur or information go missing? - What is regularly left unfinished? Answer: ### 3 · How should it work? - Which steps will the Skill perform? - When must it ask a question or stop? - What is the final result and who uses it? Answer: ### 4 · How do we recognise quality? - What must be correct in a good result? - What must never happen? - Which two results show the difference? Answer: ## Copy-ready prompts ### Build with the Skill Generator Use the skill-generator Skill. Build a fiduciary Skill from our agreed process interview. Use the four answers, our template, a good result and an average result. Ask for missing information. Define the trigger, allowed sources, steps, output format, quality rules and human decisions. Create a SKILL.md with the required references. Test the draft on our case and revise it after our review. ### Use a Skill: mandate review Use the treuhand-mandatspruefung Skill. Review the supplied checklist and documents for the Bergblick sample mandate. Separate supported information, missing evidence and conflicts. Give the file and source location for each finding. Create a task list and a draft client question. Keep the new address as a proposal until professional approval. Present the result for review and do not send a message. ### Draft in teams with Lovable Create a clickable interface for a small fiduciary office in Liechtenstein. Use invented sample data only. Our process is: [from the interview]. Users: [roles]. Build three views: a Kanban board with inbox, in progress, waiting for client and awaiting approval; a mandate dossier with documents and missing information; an AI dialogue with a draft client reply and visible human approval. Keep mandates and messages connected. Use a calm office design, MMIND plum and orange colours, and plain British English. Simulate all actions in the browser. Do not connect archives, mailboxes or AI services and do not add login or a database. Show that this is a draft with sample data. ## Facilitation and schedule ### 09:00 · 15 min · Four steps for your mandate work Turn your own process into a Skill. Then work in teams on a first app interface. - Keep introductions brief, then collect expectations. Use each team’s own process; offer the sample case if needed. ### 09:15 · 20 min · Which data may AI see? A local folder does not tell you where AI processes its content. ### 09:35 · 15 min · Where does a person decide? AI prepares the work. Set who checks the result and which action needs approval. ### 09:50 · 10 min · Review one example together A Skill supports a mandate review: it compares documents and prepares a client question. - Open the sample material. Explain the task before building the complete Skill. ### 10:00 · 20 min · 1 · How does it work today? From the trigger to the finished result - Question and three follow-up questions adapted from slide 19 of the Remo Guntli presentation. - Record only with consent; otherwise use the process sheet. Keep confidential content in the approved workspace. ### 10:20 · 20 min · 2 · Where does it get stuck? Use a specific case - Question and three follow-up questions adapted from slide 20 of the Remo Guntli presentation. - Record only with consent; otherwise use the process sheet. Keep confidential content in the approved workspace. ### 10:40 · 20 min · 3 · How should it work? Name the AI tasks and human decisions - Question and three follow-up questions adapted from slide 21 of the Remo Guntli presentation. - Record only with consent; otherwise use the process sheet. Keep confidential content in the approved workspace. ### 11:00 · 15 min · Break The interview continues at 11:15. ### 11:15 · 20 min · 4 · How do we recognise quality? Your professional judgement becomes a working rule - Question and three follow-up questions adapted from slide 22 of the Remo Guntli presentation. - Record only with consent; otherwise use the process sheet. Keep confidential content in the approved workspace. ### 11:35 · 10 min · Turn four answers into a task First agree that the description is correct. Then build the Skill. ### 11:45 · 30 min · Build with the Skill Generator Turn your process into an instruction you can reuse. - Install the Skill Generator in the chosen account in advance. If installation is not supported, use the instructions as text. ### 12:15 · 45 min · Lunch break We will use the Skills at 13:00. ### 13:00 · 45 min · Use a Skill: mandate review Run your Skill. The sample shows evidence, gaps, tasks and a client question. ### 13:45 · 30 min · Second case: does the Skill hold up? A different case shows whether the instruction can be reused. ### 14:15 · 30 min · Reuse it in your office Set who maintains and reviews the Skill and supports its first use. ### 14:45 · 15 min · Build a fiduciary app? When several people need to see the same mandates and tasks, they need a shared record. - As in Dockyard: one person describes, one builds and one reviews. Swap roles. Set the goal as an interface draft. ### 15:00 · 45 min · Draft in teams with Lovable A clear prompt creates the first interface. Then improve one process. - Check the Lovable account and build credits in advance. Each group builds its own project. Keep this workshop to a simulated interface. ### 15:45 · 15 min · Break We will review the drafts at 16:00. ### 16:00 · 30 min · From a screen to a working process Ask another person to check whether the interface is clear. ### 16:30 · 30 min · Closing discussion: what should we build next? Show your Skill and app draft. Use their value to decide which next step is worthwhile. - Plan demonstrations and questions in five-minute team slots; account for group size in advance. - The Mandatwerk interface can be a comparison. Explain that it is an interface draft. ## Materials and preparation - Bring a laptop and an approved ChatGPT or Claude account. Check available features and Skill installation in advance. - Bring a process, a template, a good result and an average result. Only use approved or anonymised content for shared work. A sample case is available. - Prepare one Lovable account per team and check the available build credits. Complete account setup and any purchase before the workshop. - Breaks: 11:00 to 11:15, lunch 12:15 to 13:00, 15:45 to 16:00. Date, price and registration to follow. Each group keeps confidential material in its approved workspace. Shared app drafts use invented data only. Record with consent; distinguish notes from a transcript. The four interview slides are adapted from slides 19 to 22 of the Remo Guntli presentation. Source method: Dockyard process analysis. The team app exercise follows the Dockyard workshop approach. - [Four process questions: Remo Guntli workshop](https://workshops.mmind.ai/de/w/rna-2026-10-06) - [Dockyard KI-Werft](https://workshops.mmind.ai/de/w/dockyard-ki-werft-2026) - [Lovable](https://lovable.dev)