🚀Introducing Morpheus UI v1- A modern ui system for ambitious products. Try on figma

Ifeanyi Achi

Menu
Medium Linkedin Envelope

Ifeanyi Achi

Menu

Designing Claudius – A Conversational AI for smarter, faster Payments

What if your bank account could have a conversation with you? In Nigeria’s bustling economy, users are ready for financial tools that are as dynamic as they are. 

Busy Nigerian professionals find traditional fintech apps repetitive and impersonal. Core tasks like paying multiple people or understanding their spending history are high-friction, manual processes that create anxiety and waste valuable time. The current interaction model of forms and clicks fails to meet the expectations of users getting accustomed to the evolving smart, personalized digital experiences

This case study explores a new fintech paradigm with AI assistant, Claudius.

Claudius exists as a powerful, conversational layer on top of a clean and familiar dashboard, capable of executing complex multi-recipient and multi-currency payments from a single voice or text command. 

This project was about redesigning the relationship between a user and their money from a purely transactional one to an intelligent, conversational partnership.

  • Role: Product designer, Interaction design, User researcher
  • The team: Solo
  • Year: 2025

💡 Background

As the sole creator of this project, I was responsible for the entire product lifecycle, from initial concept to a fully functional prototype. 

Product Strategy & Research: I independently conducted market analysis and user research to identify the core pain points of transactional friction, which formed the strategic foundation for an AI-inclusive approach.

Full-Stack Design & Architecture: I designed every aspect of the user experience, from the high-level system architecture down to the conversational micro-interactions. This included architecting the database schema for multi-currency wallets, recipients, payroll-style groups, and scheduled payments, and other key functionalities

AI-Powered Development: In lieu of a traditional engineering team, I leveraged AI development tools. I used Claude as a “backend consultant” to design and write the SQL for the entire Supabase database, including tables, RLS policies, and complex PostgreSQL functions. I then used Figma Make to build out the application logic.

AI Persona & Voice Design: I crafted the persona for Claudius, defining its personality and conversational rules. A key part of this was researching and planning for the integration of authentic voice models, such as Nigerian English (en-NG), to create a culturally resonant and trustworthy user experience.

Test the experience

Use the information below to test this interface

URL – https://weight-cat-89155266.figma.site
Email address –  ifeanyi@demo.com
Password – Password

What were the painpoints?

To move beyond surface-level assumptions, I conducted a series of user interviews targeting Nigerian professionals. My goal was to uncover the emotional friction in their financial lives, not just the functional gaps. The insights were profound and revealed a deep disconnect between how people live and how their banking apps operate

The "black tax" & the burden of family support

Many Nigerians who serve as the financial backbone of their families, experience “Black Tax” not just as a financial strain but as emotional labor. What should be an act of love- sending money to family- often becomes a stressful, manual process with no way to organize or automate recurring obligations

The burden of repetition

For some users, sending multiple payments- whether for payroll or multiple recipients- feels like a repititive, draining routine. What should be a simple act of responsibility becomes a time-consuming, stressful process that technology could easily streamline

The Everyday Struggle to Track Spending

Users feel trapped in data-rich but insight-poor financial apps. Despite having detailed transaction histories, they struggle to locate specific payments or understand spending patterns

When life gets in the way of payments

Users who often multitask (juggling work, travel, and daily responsibilities) struggle to keep up with payments. With limited time and constant distractions, they frequently forget or delay important transactions, leading to missed obligations and stress

Uncovering the unmet needs: The target personas

My research crystallized into a clear market insight: the Nigerian professional class is underserved by one-size-fits-all fintech solutions. To capture this opportunity, I developed two distinct user personas that embody the most critical and unmet needs in this segment. These archetypes allowed me to move beyond generic features and design a product tailored to the nuanced financial realities of our target audience – from managing complex family obligations to scaling a small business

Focusing the Problem, Framing the Opportunity

With a clear and empathetic understanding of the user needs, the next step was to synthesize these complex human needs into a focused, actionable design challenge. I used the “How Might We” (HMW) framework to reframe the problems as opportunities. This transformed the pain points from a list of user complaints into a clear, optimistic mission statement that would guide the entire ideation process

  • How might we make sending to one or many feel instant and effortless?
  • How might we make payments automated but still flexible and human?
  • How might we enable users to uncover patterns and meaning within their transactions so they feel more in control of their money?

Translating Insight into Product Direction

The insights uncovered pointed toward three core pillars for improving the user experience – automation, simplicity, and clarity. Each design direction aimed to reduce cognitive load, make money movement feel seamless, and give users a stronger sense of control over their finances.

Building on the opportunity areas, I began brainstorming multiple ways to address users’ recurring frustrations with effort, time, and visibility.

I used a mix of rapid sketches, user journey mapping, and feasibility discussions with the engineering team to ground the ideas in reality.

Test the experience

Use the information below to test this interface

URL – https://weight-cat-89155266.figma.site
Email address –  ifeanyi@demo.com
Password – Password

Solution 1: Enhancing a familiar interface

I designed an interface to allow users send money to multiple users at the same time. The users could either be saved recipients or new recipients.

Saved recipients

I also added a schedule feature where users can set one time or recurring transfers

Group payments was also added. This solved for the mini payroll use case so users can pay a specific group of recipients at once or automatically

Initial testing

After a quick usability testing, I discovered users were happy but nor satisfied as they still had to go through a set of interfaces and clicks. Yeah, it solved the problem but was not enough

The experience still felt like a data entry and retrieval flow – lot of clicks and just newly added features to polish the UX.

Introducing Claudius: A conversational operating system for money

This feedback led to the next iteration: to build a financial experience where the interface is natural language

To begin the exploration, I laid down ground principles for Claudius:

  1. Delegate, Don’t Operate: Users should delegate tasks, not operate a machine.
  2. Clarity Before Commitment: Every financial action must be confirmed in simple, human-readable terms.
  3. Trust Through Competence: The AI should be so reliable and intelligent that it earns the user’s trust over time

The Command Center: The Chat Interface

I explored multiple LLMs – including OpenAI, Gemini 2.5, and the Grok model – to evaluate their ability to understand natural user intent. While each performed impressively, I ultimately chose OpenAI, mainly because I already had available credits. I then designed a chat interface that allows users to perform any financial action or query within the app using natural language

Intent: The conversational flow

For Claudius to be trusted with financial actions, the process needed to be transparent, deliberate, and anchored in trust – a trust that was carefully designed to align with users’ existing mental models.

The entire experience is built around intent:

  • Users type a message 
  • OpenAI interprets the intent
  • Claudius responds accordingly.
 

For instance, if a user wants to send money to multiple recipients, the system identifies this as a bulk payment intent. If they schedule a payment (whether one-time or recurring), it recognizes a scheduled payment intent. When users ask about their transactions, OpenAI extracts the transaction inquiry intent and presents relevant, contextual financial data.

Every step of the transaction- from intent detection to confirmation – occurs within the chat interface. Before completion, a transaction summary is displayed, and users must still confirm with their PIN for security and clarity.

Sending money to a single recipient

Sending money to multiple recipients

Voice Command: The Ultimate Frictionless Experience

The design was built for voice-first interaction, allowing users to execute complex tasks like “Schedule a payment of 1.5 million Naira to my supplier next Friday” with a single utterance.

I wanted to localize Claudius voice so it has a Nigerian voice but it required extra technical efforts and complexity. So I opted to use regular English voices from Open AI TTS.

I also explored other provisions like deepgram and webspeech. While they had strong suits in other areas like transcription, it didn’t really fit the vision I had for Claudius.

Using voice to schedule a transfer

Engineering deliberate confirmation

For Claudius to be trusted with financial actions, the confirmation step needed to be crystal clear, intentional, and grounded in user confidence.

The Problem:
How do we confirm a transaction initiated by AI in a way that feels both secure and conversationally natural? A simple “Are you sure?” felt too weak, while a standard pop-up felt impersonal and out of place within the chat flow.

The Initial Design:
The first iteration introduced a confirmation card directly within the chat, summarizing the transaction details with clear “Confirm” and “Cancel” buttons. This approach gave users a visual checkpoint before proceeding. Upon tapping “Confirm,” the PIN modal would appear for final authorization.

The User Insight & Pivot:
While functional, I hypothesized that for an action as sensitive as sending money, users needed a more deliberate and conscious act of consent. I explored a text-based confirmation flow, where users typed “Yes” to proceed – a moment that demanded reflection before commitment.

However, after testing both approaches, I settled on the button-based confirmation card as the final design. It struck the right balance – maintaining the natural flow of conversation while offering a structured, physical action (a tap) to signify intent. This ensured that no transaction could proceed without explicit, confident user consent, seamlessly leading into the final PIN verification step for added security

Key lessons

AI as a force multiplier

As the solo builder on this project, I needed to move beyond static Figma screens to test real, data-driven interactions.
By using AI as an engineering partner, I was able to quickly design the database (supabase) backend and app logic.
This approach has been eye opening as I feel confident in moving from 0 to 1 quickly when designing mvps and prototypes

Designing Personality into LLMs

Large Language Models need more than intelligence - they need character. Designing an AI like Claudius required deliberate decisions about its tone, voice, manner of understanding, and response quality. Every reply had to sound human, contextual, and distinctly Claudius. This experience reinforced that personality in AI isn’t decoration - it’s a design system in itself, essential for trust, connection, and meaningful engagement.

Designing Personality into LLMs

Iterations included using a text based confirmation which was fast but too effortless for something as sensitive as moving money. It risked accidental approvals and lacked the sense of gravity users expect in financial interactions.
Not all friction is negative. In high-stakes moments like payment confirmation, introducing intentional pauses builds confidence and prevents costly mistakes.
This small act of “positive friction” turned a quick tap into a moment of trust and control - the right trade-off between speed and safety in finance.