Katarzyna Bobrowska

UX/UI Designer | Product Designer

Katarzyna Bobrowska

UX/UI Designer | Product Designer

Designing a financial copilot that keeps people in control

Product Design · Conversational UX · AI-assisted Prototyping

Product Design · Conversational UX ·
AI-assisted Prototyping

Kestrel is a concept for a personal finance app with an integrated financial copilot. It helps people explore savings goals, understand financial risks, and adjust their plans before making a decision.

The project focuses on a simple question:

How might we make financial planning more conversational without compromising clarity, safety, or user control?

My role: Product concept, UX/UI design, interaction design, design evaluation and iterative prototyping.

Tools: Claude Code, Next.js, TypeScript, Tailwind CSS.

My Design Process

I started with the challenge of planning savings when users have limited funds, variable expenses, and upcoming payments.

I designed a three-stage journey: Analyse, Adjust, and Review.

First, the Copilot analyses the user's finances and identifies potential risks. Then, users can adjust their savings plan through a conversation. Finally, they review the proposed actions and decide what to approve.

Rather than starting with a traditional set of static wireframes, I developed the experience iteratively as a working prototype, using Claude Code to support implementation and automated testing.

This allowed me to evaluate actual interactions, identify problems, and refine the experience throughout the process.

I started with the challenge of planning savings when users have limited funds, variable expenses, and upcoming payments.

I designed a three-stage journey: Analyse, Adjust, and Review.

First, the Copilot analyses the user's finances and identifies potential risks. Then, users can adjust their savings plan through a conversation. Finally, they review the proposed actions and decide what to approve.

Rather than starting with a traditional set of static wireframes, I developed the experience iteratively as a working prototype, using Claude Code to support implementation and automated testing.

This allowed me to evaluate actual interactions, identify problems, and refine the experience throughout the process.

Designing the AI Experience

I chose a conversational interface because financial planning often involves questions, uncertainties, and changing priorities.

Instead of asking users to complete a fixed form, I wanted them to be able to express their intentions naturally and explore different scenarios.

However, the current prototype does not use a live large language model.

It uses rule-based logic to interpret supported inputs and simulate financial scenarios. Financial calculations are kept separate from the conversational interface.

This approach allowed me to focus on interaction design, transparency, financial safeguards, and user control without building a production AI system.

I chose a conversational interface because financial planning often involves questions, uncertainties, and changing priorities.

Instead of asking users to complete a fixed form, I wanted them to be able to express their intentions naturally and explore different scenarios.

However, the current prototype does not use a live large language model.

It uses rule-based logic to interpret supported inputs and simulate financial scenarios. Financial calculations are kept separate from the conversational interface.

This approach allowed me to focus on interaction design, transparency, financial safeguards, and user control without building a production AI system.

Evaluating and Improving the Design

I reviewed the prototype iteratively using Nielsen's usability heuristics, Gestalt principles, and WCAG 2.2 accessibility guidance, supported by AI-assisted evaluations and automated checks.

During my review, I identified a critical usability issue: users couldn't proceed to Review when their plan was unsafe or they hadn't confirmed it was realistic.

The automated tests passed because they reflected the original interaction requirements, but those requirements created a dead end.

I revised the flow so that users could always review their plan, regardless of its financial safety status.

The ability to review a plan should not depend on being ready to approve it.

The change preserved financial safeguards: users could inspect an unsafe plan, but transfers that violated the safety rules remained blocked.

The updated interaction was then implemented and covered by regression tests.

What I Learned

The project reinforced an important principle: a technically functioning interface isn't necessarily a usable one.

AI-assisted prototyping helped me move quickly, but it also required critical evaluation of the generated solutions.

My role was not just to define what the application should do, but to question whether its behaviour made sense for the user.

The goal was to design a Copilot that supports financial decisions without taking control away from the person making them.

Kestrel is a fictional concept prototype using simulated financial data. It has not been validated through formal usability testing with participants or connected to a real banking system.

Santander

Zen.com

BMW

Ferrero

Wedel

Sector 3.0

Samsung

WWF

The 3E System

Procter & Gamble

Credit Agricole