Upgrade to Pro — share decks privately, control downloads, hide ads and more …

How KNB is using AI without giving away control...

Avatar for Marketing OGZ Marketing OGZ PRO
September 18, 2026
0

How KNB is using AI without giving away control of their data

Avatar for Marketing OGZ

Marketing OGZ PRO

September 18, 2026

More Decks by Marketing OGZ

Transcript

  1. Using AI without giving away control of your data Data

    Expo 2026 · 9-10 September · Jaarbeurs Utrecht Wouter Siepman (KNB) · Codrin Kruijne (Bravinci) [email protected]
  2. 2 Two questions. Hands up! 1 Who here is actively

    looking at reducing dependency on large public cloud providers? 2 Keep your hand up if the biggest blocker is… cost compliance Using AI without giving away control of your data · Data Expo 2026 capability
  3. 2 Please give a warm welcome to … Wouter Siepman

    Codrin Kruijne Koninklijke Notariële Bravinci Beroepsorganisatie [email protected] [email protected] Michael Doves Bravinci [email protected] Using AI without giving away control of your data · Data Expo 2026
  4. 3 Koninklijke Notariële Beroepsorganisatie • The professional body for all

    civil-law notaries in the Netherlands • A statutory role, established in law — not a trade association • What flows through the notariaat: wills, deeds, property, estates, family data We set professional standards and run collective services for our members Using AI without giving away control of your data · Data Expo 2026
  5. 4 Passing a deed requires answering many questions 2,005,071 notarial

    deeds passed in 2025 Each one needs facts checked, registers consulted, rules applied, parties identified. a record — and up 7% on the year before Questions are growing in volume and in complexity — and there are fewer notaries to answer them Using AI without giving away control of your data · Data Expo 2026
  6. 5 Three generations of answering questions WHAT IT ASKS OF

    YOUR DATA Reports and dashboards 1 You ask. Someone models the data and builds a report. Combined and modelled (Automated) analysis and alerting 2 You engineer systems that spot and present the change. Processed repeatedly Conversational AI and agents 3 You ask in plain language and get an answer back. Exposed to a non-deterministic LLM Using AI without giving away control of your data · Data Expo 2026
  7. 6 Challenges for KNB • Confidentiality is a statutory duty

    • Organizing and assessing data security takes real work • Members are experimenting with AI • The market is moving quickly Given our role, what should we develop to offer our members? Using AI without giving away control of your data · Data Expo 2026
  8. 8 What does “control” actually mean? 1 Where does our

    information live? Storage, hosting, and whose law the provider answers to 2 Where does it go when we use it? Every step, every copy, every place it gets processed 3 Who can see it — and in what form? The original data, derivatives and audit trails Control means you can secure it, audit it, adapt it quickly and move it Using AI without giving away control of your data · Data Expo 2026
  9. 9 One conclusion … …we have to control the stack.

    TWO PRINCIPLES UNDERNEATH EVERYTHING THAT FOLLOWS Bring the AI to the data... And if it must move… Every copy is another place it can leak from, and another thing to govern. …be able to quickly and securely. Using AI without giving away control of your data · Data Expo 2026
  10. 10 What does that stack have to do? 1 Answers

    in a form people can act on — a report is only one of those forms ↓ 2 Interfaces to find and format data — and a way to manage who gets in ↓ 3 Process and check information — rule-based or agentic ↓ 4 Saving data and processes — so we can check their quality ↓ 5 Accessing source systems securely — where the data is actually born ↓ And if none of it may go to public services, all of it runs somewhere we control. Using AI without giving away control of your data · Data Expo 2026
  11. 11 Building it from the foundation up KNB private cloud

    · Dutch jurisdiction Data warehouse Exasol — everything in one place, with its own MCP server for AI agents Integration platform (iPaaS) MuleSoft — where data crosses the line Source systems · registers, office systems, administration Using AI without giving away control of your data · Data Expo 2026
  12. 12 Two ways to turn data into answers KNB private

    cloud · Dutch jurisdiction BI and traditional AI Agentic AI off the shelf assembled by KNB? Analytics and machine learning KNB MCP server? built in — no separate model to host Our knowledge, in a form AI can use Data warehouse Exasol — everything in one place, with its own MCP server for AI agents Integration platform (iPaaS) MuleSoft — where data crosses the line Source systems · registers, office systems, administration Using AI without giving away control of your data · Data Expo 2026
  13. 13 For internal and external users KNB employees · KNB

    members KNB private cloud · Dutch jurisdiction BI and traditional AI Agentic AI off the shelf built by KNB BI interface Chat interface, agent coordination and LLMs Reports, dashboards, access Analytics and machine learning KNB MCP server built in — no separate model to host Our knowledge, in a form AI can use Data warehouse Exasol — everything in one place, with its own MCP server for AI agents Integration platform (iPaaS) MuleSoft — where data crosses the line Source systems · registers, office systems, administration Using AI without giving away control of your data · Data Expo 2026
  14. 14 Why these two as the foundation of our stack

    Sovereign Agentic Database Analytics Platform Information Systems Reference Architecture the BI platform • One place to combine and model data • Reports and dashboards • Runs (ML) AI inside the database • Signals: spots unusual changes • Has its own MCP server for Agentic access • Guided Natural Language Querying Neither of them needs our data to leave in order to 'do its AI'. Using AI without giving away control of your data · Data Expo 2026
  15. 16 Lessons learned 1 An architecture built across several suppliers

    needs constant alignment. Independence is worth having. 2 It is tempting to start with the stack. Keep starting with the use case. The technical work is more fun maybe, the human work is harder. 3 LLM capability will keep climbing. Work out what stays uniquely yours. The knowledge only we hold makes the difference. Not one of these turned out to be a technical problem. Using AI without giving away control of your data · Data Expo 2026
  16. 17 Three things to ask yourself 1 How do you

    keep data and AI safe? Know where things are stored, processed and accessed — and make sure you can move any of it. 2 LLMs are improving fast. What will you build that is uniquely yours? Don't build the thing the next model release will simply absorb. Build the thing only you have. 3 Where is the value you can take today? Most of it is in the unglamorous, statistical work that never leaves your walls. Bring the AI to your data. Don't send your data to the AI. Using AI without giving away control of your data · Data Expo 2026
  17. 17 Let’s talk … … happy to meet you with

    our partners at stand 175 – Bravinci + Exasol + Yellowfin Using AI without giving away control of your data · Data Expo 2026
  18. 7 And we serve two very different populations KNB employees

    KNB members internal notary offices • Our own bureau and our own operations • Notary offices across the Netherlands • That data is ours — and ours to answer for • KNB does not hold their client files — the notary does • We can answer the control questions directly • They inherit whatever answers we can give For members we are not only exercising control. We are supplying it. One stack. Two front doors. Using AI without giving away control of your data · Data Expo 2026
  19. 15 Numbers are one thing. Text is another. Numbers Counts,

    amounts, dates. Structured, checkable, and the same answer every time you ask. Text Deeds, clauses, correspondence. Meaning depends on context — and the words carry legal effect. Ask a language model the same question twice and you may get two different answers. The law cannot work that way. So generative AI needs strict, repeated testing — against a “golden dataset” built with the experts. A fixed set of questions where a notary knows the right answer. Re-run on every change. “AI mag helpen denken, maar nooit beslissen” — Vereniging IT-Notaris, opleidingsdag AI Using AI without giving away control of your data · Data Expo 2026