How DominoIQ helped a customer in the tourism sector to grow

In the spring of this year (2025) we started with the first Beta Implementations of the new Domino IQ based features and functions at a company for packaged coach tours and river cruises in Germany. The company is one of the market leaders in this sector reselling packaged travel arrangements in the whole DACH Region. During the COVID season, the company went through rough times, almost going bankrupt with revenues coming to a total halt during the pandemic. During that time, layoffs had been necessary and a large amount of highly qualified speciallists lost their jobs and went into different professions. Fast forward to today - the market segement is booming and the company is grwoing rapdidly, so fast, that they are not able to find enough skilled personell to meet the growth. As a long term HCL Notes/Domino user, we sat down and came up with an idea, how Domino IQ could come to the rescue - let complex marketing and controlling tasks be automated using LLMs, RAG and MCP servers, allowing the customer to do a lot more with less - and meet their goals to regrow the business in a lot more efficient way than before. In this presentation, I will talk about the use cases and the business impact that using AI and Domino IQ had. Also, I will. give an overview of the technical implementation Join me on a customer journey through making productive use of AI and Domino IQ!

Transcript

1. Copenhagen & Stockholm, October 2025 WELCOME We build bridges between people & technology!
2. Heiko Voigt, 21.10.2025 Basic AI features, complex flows and future steps How DominoIQ helped a customer in the tourism sector to grow
4. About me… • Founder & CEO of HCL Business Partners SIT GmbH in Germany and Harbour Light Software Development Ltd in Canada • More than 30 years in Business with Lotus, IBM, HCL – yellow bleeder • Full-Stack Software Developer • HCL Ambassador since 2019 • OpenNTF Board Member since 2022 • Core Team Member of Cross Canada Collaboration User Group (C3UG) • Blogger, Sailor, Home Brewer, proud father of twin girls.
5. What are we looking at today ... • The customer and it‘s situation • Why the use of AI? • The AI Implementation - HCL DominoIQ as the base • The current AI enabled Workflows • The next big thing – Virtual AI Travel Agent • Outcomes, Lessons learned Agenda
6. The business – a hidden champion • Market leader in pre-packaged Coach- and river cruises in Europe • OEM approach – no direct business but re-selling Voyages through small tour operaters and travel agencies with their own label. • High standard, high quality. The company had excellent ratings from its customers and competitors • „Gold Standard“ in service, reliability and quality ! • The business uses HCL Notes/Domino – apps that run your business – since 2005. The customer
7. How we came to where we are The customer and it‘s situation 0 500 1000 1500 2000 2500 3000 2020 2021 2022 2023 2024 2025 Employees, Projects and Revenue Employees Projects per year Revenue • In 2020 the company was doing just great. Lot‘s of employees, lots of projects, big revenue. • And then – COVID hit. And it hit hard. • After being almost bankrupt in 2023, the company rebounced in 2024
8. Now, we face some more issues! The customer and it‘s situation 0 500 1000 1500 2000 2500 3000 2020 2021 2022 2023 2024 2025 Employees, Projects and Revenue Employees Projects per year Revenue • They can‘t get the people back that they needed to let go during the pandemic. • Especially, the SMEs they have are suffering – too much daily tasks block them from more complex voyages that need to be planned. • The lack of personell became the main blocker for growth.
9. Now, we face some more issues! The customer and it‘s situation • During the pandamic, a lot of the long standing partner businesses like Hotels, Guides, local tour operators etc. went out of business – on a global scale. • The SMEs of the customer (usually for each region they serve the have about 2) need to identify new and reliable partners to work with. • But they also have to work on heads-down in daily business to keep the company afloat. • Our task: bring in AI to reduce the workload and let employees who are no SMEs to be able to run the regular business to let the SMEs go back to do more important work. Supplier Catalog In an NSF ;-)
10. A quick look into how the business operates … The customer and it‘s situation • The Customers are booking coach tours or river cruises ahead of their actual travel dates (I want to go to Sicilly in February for 7 days in four star hotels) • The company does a calculation based on criterias like region, requested ammenities and creates a proposal that includes Places to visit, restaurants, hotels, tour guides etc. • When the customer books the package, the company „books“ all defined deliverables with respective providers like hotels, guides and restaurant and pays them upfront. • All of this is done on a project base and via E-Mail communication linked back with the respective project. Each deliverable of the project is a seperate entity / category of the project and several people are working on the same project. Complex projects are overseen by an SME to insure quality. customer project component component component Contracts Component Suppliers Vouchers
11. The AI strategy – leverage quick wins and invest in the giant leap! What AI has to do… • Help in supplier selection (pre-write emails, translate e-mails, auto-scan attachments and move to project, proceed workflow states based on return data) • Create booking components of a project with the respective supplier (Create request document and email, translate emails, automate parts of the process through Agents) • Quick and easy to implement – huge impact – productivity increase by 80% in comparison of AI-supported vs non-AI supported Projects! => Higher throughput – almost doubles the volume! customer project component component component Contracts Component Suppliers Vouchers Quick wins:
12. The AI strategy – leverage quick wins and invest in the giant leap! What AI has to do … • Build a virtual SME to allow „standard“ tarvel agents to build complex proposals and select appropriate component suppliers • Physical SMEs will only do quality assurance • Reduce the workload of SMEs to focus on business development • Reduce the amount of work for complex proposals by 50% customer project component component component Contracts Component Suppliers Vouchers The goals of the giant leap…
13. What AI to use ? ChatGPT for everything – well, be carefull! The AI implementation customer project component component component Contracts Component Suppliers Vouchers • We have about 2000 E-Mails a day going out to suppliers and about 2000 coming back in • There are about 150 travel documents written each day (not from scratch but we were able to reduce the amount of time to write one letter from 15 to 7 Minutes using AI) • Converting translation transactions, creation of emails and summarizing transactions in needed tokens to pay with services like ChatGPT would simply eat up the benefits. • So local AI with different models was way more attractive from a business perspective. • Renting server capacity with respective Nvidia/AMD HW was a good trade off.
14. The Infrastructure The AI implementation HP Z2 Mini G1a 128 GB Unfified RAM Ryzen AI Max+ PRO 395 4 TB SSD HD HETZNER Cloud Server for Model fine tuning HCL Domino 14.5 Domino IQ MCP Implementation (Node.js) Shimmy / Ollama Shimmy/Ollama Mistral latest 7B Shimmy / Ollama Mistral latest 7B Gemini (fine tuned) Gemini (fine tuned) N8N Automation AI Agents N8N Automation AI Agents Part of App- Server Cluster Drop-in OpenAI API Replacement for Local LLMs
15. Quick wins: outbound Proposal E-Mail Flow using pure DominoIQ The AI implementation HCL Domino 14.5 Domino IQ Shimmy / Ollama Mistral latest 7B • Notes Apps got enhanced – when RFP gets defined, project manager can use AI functions to create the RFP • Core data from the project (where to, when, what gets requested) are passed as context to Mistral to create RFP request. • After final modification, RFP can be translated to target language (English, French, Italian, Spanish, Polish, and others) • After final review by SME, RFP gets sent out via E-Mail to the supplier. • We use multiple instances of Mistral with different language presets for target translations. Mistral latest 7B Mistral latest 7B Mistral latest 7B
16. Screenshots… The AI implementation HCL Domino 14.5 Domino IQ Shimmy / Ollama Mistral latest 7B Mistral latest 7B Mistral latest 7B Mistral latest 7B
17. Inboud Supplier Response E-Mail Flow The AI implementation HCL Domino 14.5 Domino IQ MCP Implementation (Node.js) Shimmy / Ollama Mistral latest 7B Gemini (fine tuned) N8N Automation AI Agents • „When new Mail Arrives“ Agent on N8N-Domino triggers N8N Webhook • N8N reads Message via REST-API from Domino • N8N Calls language recognition and translation to german if needed using one of the MISTRAL Latest 7B models we have. • N8N Calls Content Extraction Agent flow using Fine-tuned Gemini Model to extract provider data, Project data. Uses MCP interface to access Data on HCL Domino project DB and provider DB. Also creates a summary and adds flags, if questions or warnings are recognized. PDFs are scraped and analyzed. • N8N adds E-Mail to manual re-work if no or not enough data can be extracted. Translation will be added as well. • If we gather enough data, N8N adds E-Mail to project folder as response to correct request-for-proposal and adds translation as well. • N8N updates to-do lists for project managers.
18. Domino Inboud Supplier Response E-Mail Flow The AI implementation New E-Mail & Attachments Tokenize Content (Mistral Fine tuned) Tokenized Data N8N Datatable Reasoning (Gemini FT) N8N Datatable MCP Add Data to Project Structure Move to manual rework Translation (Mistral) Estimated accuracy > 0.75 N8N Datatable Clear Datatables Domino N8N Agent checks new Mail, calls Webhook if needed Reads mail and attachments via REST-API
19. Why fine-tuning? The AI implementation HETZNER Cloud Server for Model fine tuning Shimmy / Ollama Gemini (fine tuned) • Basic rules for provider-ids, project-ids for all types of interactions (no re- loading from RAG for every transaction needed) • More precise data extraction in coherent JSON format – easy to process in further steps.
20. Why Shimmy ? The AI implementation HP Z2 Mini G1a 128 GB Unfified RAM Ryzen AI Max+ PRO 395 4 TB SSD HD HETZNER Cloud Server for Model fine tuning Shimmy Drop-in OpenAI API Replacement for Local LLMs • Small footptint runtime for local LLMs (4.8 MB) • Drop-in compatible with OpenAI API – all tools that work with OpenAI APIs work instantly • AMD GPU support ! (and all others) • Systematic workflow: /specify → /plan → /tasks → implement • AI-assistant compatible (Claude Code, GitHub Copilot, DominoIQ) • Built-in multi-instance and load balancing support • https://github.com/Michael-A-Kuykendall/shimmy Mistral latest 7B Gemini (fine tuned)
21. Why N8N ? The AI implementation HCL Domino 14.5 Domino IQ MCP Implementation (Node.js) Shimmy / Ollama Mistral latest 7B Gemini (fine tuned) N8N Automation AI Agents • The Jack-of-all-trades for automation! • Open Source Workflow Automation Solution • Runs on premises ! • Visual building of workflows – drag&drop, easy maintenance, integrated backup and versioning • More than 500 pre-built Integration components to existing solutions • Basic Hooks to REST-APIs, OpenAI APIs, E-Mail, Webhooks plus Code execution nodes for JavaScript or Python to modify data. Use of LangChain Blocks for text/table extraction from PDFs. • Prefers JSON for data transport between nodes. • Supports Memory solutions for LLMs to retain context!
22. The Timetable for 2025 HCL Domino 14.5 Domino IQ MCP Implementation (Node.js) Shimmy / Ollama Mistral latest 7B Gemini (fine tuned) N8N Automation AI Agents Jan Feb March April May June July & August What can we do with AI? Socping 1st Tests with Ollama, 14.5 EA1 Tests with various Models to identify best choice Build automation infrastructure – REST API and N8N Demo runs, switch to Shimmy, Testing and Go-Live Built the use cases, start fine tuining, MCP coding. Fine tuning Model and selecting final models, building flows.
23. What did we achieve? The outcome… • Do more ! • The two implemented AI enhanced workflows reduced the amount of work in the RFP process by 50% per project. • Currently, 50% of the outbound communication uses AI • 100% inbound communication is using AI with 65% fully automated processing. • This allowed for 35% more projects beeing worked on in parallel and an increase in revenue by 23% in the three months of AI usage compared YoY.
24. • Everything around AI is moving damned fast! Be prepared for change and embrace it! • Fine tuning is a new tool that seems to bring more quality and a lot less haluzionizung to LLMs. A problem for the big cloud providers – better for local or own deployments! • Don‘t hard code if you don‘t have to! Build loosely coupled systems – you will iterate over your architecture every couple of months. • Do a serious calculation of costs if you tend to go with OpenAI or other cloud providers. Your benefits might end up being their benefit ;-) • Test, test, test – use different local models and try them for individual use cases. There‘s not one fits all. That‘s the power of tools like Ollama – easy testing, quick adoption. • DominoIQ is ready for prime time – use it and please HCL – extend it ! What to remember and keep in mind… Lessons learned …
25. What’s Next • The virtual travel planning SME • Fine tune specific models for information for a specific country/region • Combine fine-tuning, LangChain Memory, RAG and MCP calls to let LLM „remember“ conversations for each project on re-occurence. • Requests like: „We want to do a coach tour through Sicilly in the spring of next year for one week (7 days / 6 nights) in three star hotels and want to send the customers to the most impoartant historic sites with individual tour guides. Breakfast, lunch and supper should be included with a three course meal in the evening at the respective hotel. Flights to and from germany should be included into the calculation. Sketch out a travel plan and a first calculation for a minimum of 30 people participating“ • Goal: Agent selects POIs, travel routes, providers and a base calculation. No SME needed to build the frist proposal or only as quality control instance. Proposals can be built by regular employees.
26. Contact and Links Thank you ! • My contact details: - hvoigt@sit.de or heiko.voigt@harbour-light.com - https://heikos-blog.ghost.io • Some helpful links: - n8n: https://n8n.io/ - Shimmy: https://github.com/Michael-A-Kuykendall/shimmy - Serdar‘s Blog on how to Connect DominoIQ to external proivders: https://lotusnotus.com/2025/07/connecting-domino-iq-external-llm-providers/ - LangChain4j for Domino: https://lotusnotus.com/2025/06/langchain4j-domino-v1-0-0-released/ - How to add Memory to chat LLM models: https://medium.com/@penkow/how-to-add-memory-to-a-chat-llm-model-34e024b63e0c
28. Copenhagen & Stockholm, October 2025 WELCOME We build bridges between people & technology!