This project investigates how to design a scalable, human-centered AI interaction framework for internal wealth-management workflows at Van Lanschot Kempen. It responds to operational fragmentation caused by multiple independently developed AI features, inconsistent interaction logic across tools, and varying levels of trust among staff. The research phase combined an exploratory field study (11 semi-structured interviews with private bankers, relationship managers, and investment advisors) with a thematic analysis of workflow pain points and scenario mapping of core tasks such as text writing, information lookup, and advice generation.
The iterative design process produced three interlocking outcomes. Firstly, a set of design principles that emphasize human control, transparency, clarity, and actionable feedback. Secondly, an atomic component library of reusable UI elements and panel formats to ensure consistent interaction patterns across features. Thirdly, an AI Feature and Component Selection Wizard that guides designers and product owners through interaction style, input/output configuration, and container selection, producing a concrete specification to support cross-functional handoff.
A formative evaluation with three designers and two product owners in moderated sessions revealed strong willingness to adopt the framework, increased confidence in design decisions, and clear practical value for aligning product and design stakeholders. Recommendations for refinement include clearer terminology, richer visual previews, and better mapping to the Figma component library.
The thesis contributes a method for operationalizing human-centered AI principles into a domain-tuned design system for regulated environments, along with a practical tooling approach that shortens the path from concept to implementation.