Designing for
Scalable AI Interactions
Designing for
Scalable AI Interactions
Designing for
Scalable AI Interactions
A modular design system and governance framework to operationalize human-centered AI in wealth management workflows.
A modular design system and governance framework to operationalize human-centered AI in wealth management workflows.
A modular design system and governance framework to operationalize human-centered AI in wealth management workflows.






role
role
role
Lead UX Researcher & Designer
Lead UX Researcher & Designer
Lead UX Researcher & Designer
Timeline
Timeline
Timeline
Feb 2025 - Jul 2025
Feb 2025 - Jul 2025
Feb 2025 - Jul 2025
methods
methods
methods
Exploratory research, Thematic analysis, Atomic design
Exploratory research, Thematic analysis, Atomic design
Exploratory research, Thematic analysis, Atomic design
outcome
outcome
outcome
Design principle, Design system
Design principle, Design system
Design principle, Design system
The Challenge
The Challenge
The Challenge
Van Lanschot Kempen, like many financial institutions, was moving quickly to adopt AI capabilities. However, it led to a fragmented internal experience. Different product teams were developing AI features in isolation, without a shared design language or consistent interaction model.
Van Lanschot Kempen, like many financial institutions, was moving quickly to adopt AI capabilities. However, it led to a fragmented internal experience. Different product teams were developing AI features in isolation, without a shared design language or consistent interaction model.
"Yeah, I know the basic things AI is good at. But I struggle translating it to my daily work."
"Yeah, I know the basic things AI is good at. But I struggle translating it to my daily work."
— Private Banker (Interviewee)
— Private Banker (Interviewee)
As a result, the organization experienced operational fragmentation. Bankers were forced to constantly switch contexts, while trust in AI tools declined due to inconsistent and unpredictable interaction patterns.
As a result, the organization experienced operational fragmentation. Bankers were forced to constantly switch contexts, while trust in AI tools declined due to inconsistent and unpredictable interaction patterns.



Company AI usage
Company AI usage
Project Goal
Project Goal
Project Goal
Design a scalable, human-centered framework that embed Generative AI into existing workflows, ensuring consistent, transparent, and effective AI interactions.
Design a scalable, human-centered framework that embed Generative AI into existing workflows, ensuring consistent, transparent, and effective AI interactions.
Design a scalable, human-centered framework that embed Generative AI into existing workflows, ensuring consistent, transparent, and effective AI interactions.
01
01
Discovery & Research
Discovery & Research
Discovery & Research
Identifying Operational Constraints in Regulated Workflows
Identifying Operational Constraints in Regulated Workflows
Identifying Operational Constraints in Regulated Workflows
11
In-depth interviews
In-depth interviews
In-depth interviews
5 Private bankers, 4 Relationship managers, 2 Investment advisors
5 Private bankers, 4 Relationship managers, 2 Investment advisors
3
core workflows
core workflows
core workflows
Text writing, Look up information, Get advice
Text writing, Look up information, Get advice
4
critical constraints
critical constraints
critical constraints
Identified for regulated contexts
Identified for regulated contexts
key insight 01
key insight 01
key insight 01
Prompting Barrier
Prompting Barrier
Prompting Barrier
The cognitive load of "prompt engineering" often outweighed the benefits of AI. Users explicitly requested structured inputs (templates, presets) to bypass the "blank canvas" paralysis.
We assumed users wanted conversation. What they actually wanted was efficiency.
The cognitive load of "prompt engineering" often outweighed the benefits of AI. Users explicitly requested structured inputs (templates, presets) to bypass the "blank canvas" paralysis.
We assumed users wanted conversation. What they actually wanted was efficiency.
The cognitive load of "prompt engineering" often outweighed the benefits of AI. Users explicitly requested structured inputs (templates, presets) to bypass the "blank canvas" paralysis.
We assumed users wanted conversation. What they actually wanted was efficiency.
"If you just say 'write this', it doesn't work well. It often takes longer to write a good prompt than to just write the email myself."
"If you just say 'write this', it doesn't work well. It often takes longer to write a good prompt than to just write the email myself."
— Private Banker 5
— Private Banker 5
— Private Banker 5
key insight 02
key insight 02
key insight 02
Trust Variance Spectrum
Trust Variance Spectrum
Trust Variance Spectrum
We found that trust in AI is risk-dependent. Users treat AI as a "Junior Analyst": helpful for drafting and exploratory work where they maintain control, yet potentially dangerous when used for advice that carries legal or professional liability.
The cognitive load of "prompt engineering" often outweighed the benefits of AI. Users explicitly requested structured inputs (templates, presets) to bypass the "blank canvas" paralysis.
We assumed users wanted conversation. What they actually wanted was efficiency.
We found that trust in AI is risk-dependent. Users treat AI as a "Junior Analyst": helpful for drafting and exploratory work where they maintain control, yet potentially dangerous when used for advice that carries legal or professional liability.
Text Writing
Text Writing
Text Writing
Users feel safe because they can verify and edit the output immediately.
Users feel safe because they can verify and edit the output immediately.
Users feel safe because they can verify and edit the output immediately.
Information Lookup
Information Lookup
Information Lookup
"Goal known, input unknown." Highly desired, but users fear hallucinations.
"Goal known, input unknown." Highly desired, but users fear hallucinations.
"Goal known, input unknown." Highly desired, but users fear hallucinations.
Advice & Planning
Advice & Planning
Advice & Planning
High skepticism. AI advice lacks proven reliability and users bear responsibility
High skepticism. AI advice lacks proven reliability and users bear responsibility
High skepticism. AI advice lacks proven reliability and users bear responsibility
Research Synthesis
Research Synthesis
Research Synthesis
Defining the design gap
Defining the design gap
A single “Chat” interface oversimplifies financial workflows. Depending on the situation, users may prioritize speed and convenience or demand transparency and accountability. These contextual shifts require flexible interaction patterns rather than a one-size-fits-all chat model.
A single “Chat” interface oversimplifies financial workflows. Depending on the situation, users may prioritize speed and convenience or demand transparency and accountability. These contextual shifts require flexible interaction patterns rather than a one-size-fits-all chat model.
A single “Chat” interface oversimplifies financial workflows. Depending on the situation, users may prioritize speed and convenience or demand transparency and accountability. These contextual shifts require flexible interaction patterns rather than a one-size-fits-all chat model.
Minimise friction
Minimise friction
Minimise friction
Shift from "Prompt Engineering" to "Display → Select". Users need smart presets, not blank text boxes.
Shift from "Prompt Engineering" to "Display → Select". Users need smart presets, not blank text boxes.
Embed Context
Embed Context
Embed Context
AI must operate inside the document or portfolio view, inheriting the client context to give relevant response.
AI must operate inside the document or portfolio view, inheriting the client context to give relevant response.
Easily undone
Easily undone
Easily undone
Humans retain control over AI, with the ability to quickly delete or modify AI actions.
Humans retain control over AI, with the ability to quickly delete or modify AI actions.
These conflicting needs required us to move from building "features" to designing a Modular Interaction System.
These conflicting needs required us to move from building "features" to designing a Modular Interaction System.
These conflicting needs required us to move from building "features" to designing a Modular Interaction System.
02
02
Framework Design Outcome
Framework Design Outcome
Framework Design Outcome
A Modular AI Design System
A Modular AI Design System
A Modular AI Design System
Based on the research, I developed three interlocking deliverables to ensure AI features are trustworthy, usable, and well-integrated. This allows product teams to assemble AI features that are consistent in design, and tailored to the specific requirement of the task.
Based on the research, I developed three interlocking deliverables to ensure AI features are trustworthy, usable, and well-integrated. This allows product teams to assemble AI features that are consistent in design, and tailored to the specific requirement of the task.
Design Principles
Design Principles
Design Principles
The ethical and practical foundation for all future AI development.
The ethical and practical foundation for all future AI development.
Human Control
Human Control
Provide obvious stop or undo options. Users must explicitly confirm critical actions like record updates.
Provide obvious stop or undo options. Users must explicitly confirm critical actions like record updates.
Transparency
Transparency
Use labels for generated content. Explain how suggestions were formed and cite sources.
Use labels for generated content. Explain how suggestions were formed and cite sources.
Clarity
Clarity
Use familiar layouts and prioritize direct information over complexity.
Use familiar layouts and prioritize direct information over complexity.
Actionable
Actionable
Offer "Regenerate" or "Edit" options. Embed feedback mechanisms to flag mistakes instantly.
Offer "Regenerate" or "Edit" options. Embed feedback mechanisms to flag mistakes instantly.


Component Library
Component Library
Organized via Atomic Design logic to ensure consistency and speed.
Organized via Atomic Design logic to ensure consistency and speed.
Templates (Panel Formats)
Templates (Panel Formats)
Templates (Panel Formats)
Defined 4 core formats: Embedded Inline, Floating Widget, Side Panel, and Full-page Workspace.
Defined 4 core formats: Embedded Inline, Floating Widget, Side Panel, and Full-page Workspace.
Organisms
Organisms
Organisms
Complex modules like Prompt inputs and configurable Chat bubbles.
Complex modules like Prompt inputs and configurable Chat bubbles.
Atoms
Atoms
Atoms
Foundational elements (Icons, Tags, Buttons…) adopted from company's existing design system to lower adoption friction.
Foundational elements (Icons, Tags, Buttons…) adopted from company's existing design system to lower adoption friction.



Page example: Inline AI Usage
Page example: Inline AI Usage



Wizard prototype
Wizard prototype
The Selection Wizard
The Selection Wizard
A decision-support tool that bridges the gap between principles and pixels. It guides Product Owners through a structured questionnaire to generate a tailored design specification.
A decision-support tool that bridges the gap between principles and pixels. It guides Product Owners through a structured questionnaire to generate a tailored design specification.
1
Define AI Role (Contextual vs. General)
Define AI Role (Contextual vs. General)
2
Select Interaction Style & Input Methods
Select Interaction Style & Input Methods
3
Output: Recommended Components
Output: Recommended Components
Application Flow
Application Flow
How this framework help in product development process?
How this framework help in product development process?
How this framework help in product development process?
STEP 1
STEP 1
STEP 1
Define AI Features
Define AI Features
Define AI Features
Use the AI Selection Wizard to scope and decide which AI features fit business goals.
Use the AI Selection Wizard to scope and decide which AI features fit business goals.
Use the AI Selection Wizard to scope and decide which AI features fit business goals.
Focus:
Focus:
Focus:
Aligns functionality with value and compliance
Aligns functionality with value and compliance
Aligns functionality with value and compliance
STEP 2
STEP 2
STEP 2
Design with Components
Design with Components
Design with Components
Use the Component Library to prototype new AI tools and ensure a smooth design handoff for development.
Use the Component Library to prototype new AI tools and ensure a smooth design handoff for development.
Use the Component Library to prototype new AI tools and ensure a smooth design handoff for development.
Focus:
Focus:
Focus:
Ensures consistency, scalability, and transparent interaction
Ensures consistency, scalability, and transparent interaction
Ensures consistency, scalability, and transparent interaction
APPLICATION
APPLICATION
APPLICATION
Use in daily work
Use in daily work
Use in daily work
Use the Component Library to prototype new AI tools and ensure a smooth design handoff for development.
Use the Component Library to prototype new AI tools and ensure a smooth design handoff for development.
Use the Component Library to prototype new AI tools and ensure a smooth design handoff for development.
Focus:
Focus:
Focus:
Use the new AI-powered tools for daily work
Use the new AI-powered tools for daily work
Use the new AI-powered tools for daily work
03
03
Validation & Iteration
Validation & Iteration
Validation & Iteration
Evaluation with 5 internal experts revealed a clear split in how different roles using the tool, leading to critical refinements.
Evaluation with 5 internal experts revealed a clear split in how different roles using the tool, leading to critical refinements.
Evaluation with 5 internal experts revealed a clear split in how different roles using the tool, leading to critical refinements.
Designer's View
Designer's View
Designer's View
Views the wizard as valuable for ideation or scoping.
Experienced designers will likely move directly to Figma for hands-on work.
Appreciates following design guidelines and using company style.
Views the wizard as valuable for ideation or scoping.
Experienced designers will likely move directly to Figma for hands-on work.
Appreciates following design guidelines and using company style.
Views the wizard as valuable for ideation or scoping.
Experienced designers will likely move directly to Figma for hands-on work.
Appreciates following design guidelines and using company style.
"Especially I like the input part, because sometimes we think about what can be the input, but it's never clear until we are doing the design... that actually is the essential part when you are creating the design."
"Especially I like the input part, because sometimes we think about what can be the input, but it's never clear until we are doing the design... that actually is the essential part when you are creating the design."
— Designer A
— Designer A
— Designer A
Product owner's View
Product owner's View
Product owner's View
Focus on the practical business scenario.
Realistic about product complexity.
Tool helpful for sparking requirements, brainstorming, and mapping capabilities.
Technical constraints, UI design tools alone cannot solve technical and operational complexities, it lays the groundwork for more detailed work to follow.
Focus on the practical business scenario.
Realistic about product complexity.
Tool helpful for sparking requirements, brainstorming, and mapping capabilities.
Technical constraints, UI design tools alone cannot solve technical and operational complexities, it lays the groundwork for more detailed work to follow.
Focus on the practical business scenario.
Realistic about product complexity.
Tool helpful for sparking requirements, brainstorming, and mapping capabilities.
Technical constraints, UI design tools alone cannot solve technical and operational complexities, it lays the groundwork for more detailed work to follow.
"I think this is really good starting point for all the teams that are exploring and and thinking about how to integrate in a friendly way for our bankers."
"I think this is really good starting point for all the teams that are exploring and and thinking about how to integrate in a friendly way for our bankers."
— Product owner C
— Product owner C
— Product owner C
