Industry

Fintech

Client

Quipu

EduBot – A Free Conversational CFO for Microbusinesses

Designing an AI-powered WhatsApp chatbot to support financial decision-making for underserved entrepreneurs in Colombia.

Year

2025

Role

Product Designer (UX/UI)

Main Project Image
Main Project Image
Main Project Image

From 0 to 1,000: Scaling an AI Chatbot for Financial Inclusion

Quipu was selected by Turn.io to participate in a four-month global cohort focused on building social impact chatbots using AI and WhatsApp. The proposal: design a chatbot that would act as a conversational CFO for microentrepreneurs—helping them understand their business finances and make better decisions, using OpenAI technology.

Results & Impact

  • +3,000 interactions during the first test phase

  • +1,000 active users reached during the pilot

  • Positioned WhatsApp as a viable, accessible channel for financial education

  • Demonstrated the feasibility of deploying a lean AI solution in a low-resource, high-impact context

Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1

Before & After – Conversation Flow Redesign
The previous screen (left) shows a closed, form-like interaction that ended in a partial financial diagnosis. Usability tests revealed a completion rate of just 30%.
The redesigned screen (right) introduces a more open-ended chat experience, where users receive tailored advice and suggestions across six interactions—covering various aspects of their business in a more natural and engaging way.
This iteration led to a significant increase in the number of users who successfully completed the full flow.

Problem & Opportunity

The main challenge was leading the development and launch of EduBot’s beta version while meeting the program’s ambitious milestones:

  • Build a functional demo

  • Scale to 100 users

  • Scale to 1,000 users

We faced technical limitations including the lack of backend integration, short-term memory (the bot lost context after a few messages), and limited cross-team availability due to internal workload.

Goal

Design a chatbot that:

  • Helps users understand and make decisions about their business finances

  • Is fully usable through WhatsApp

  • Operates safely, ethically, and effectively—even in low-tech-literacy environments

Role & Contributions

As Product Designer, I was responsible for:

  • Designing conversation flows focused on accessibility and clarity

  • Refining prompts to avoid irrelevant or harmful outputs

  • Managing token consumption to ensure efficiency

  • Iterating the chatbot through several stages: rigid menus → guided prompts → open dialogue with guardrails

  • Applying ethical principles in conversational UX and prompt design

  • Coordinating ongoing monitoring, learning, and improvements based on real user interactions

Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1

Partial view of EduBot’s conversational flow and prompt structure, including onboarding, consent handling, and custom instructions for LLM responses.

Process

EduBot’s design required not only building a functional chatbot, but also applying strategic UX practices to ensure clarity, relevance, and accessibility—especially for users with limited digital experience. The process combined conversational design, prompt optimization, and collaborative discovery. Key components included:

  • Prompt Engineering: Refined prompts to improve clarity, guide interactions, and prevent hallucinated or ambiguous responses.

  • Safety by Design (Gender-Equitable): Ensured inclusive tone and use cases that reflected users’ real contexts.

  • Accessibility: Structured messages with clear examples and low cognitive load, tailored for users with low literacy.

  • Ethical Design: Framed interactions respectfully, avoiding manipulation or bias.

  • Go-to-Market Strategy: Integrated the bot into Quipu’s WhatsApp channels with a phased rollout plan.

Several UX artifacts supported the product strategy and execution:

  • Stakeholder Ideation Sessions: Co-created early hypotheses and aligned expectations around user impact.

  • User Persona Development: Based on real Quipu users to guide content, tone, and flow logic.

  • Tone & Voice MVP Manual: Defined conversational patterns for delays, follow-ups, clarification, and empathy.

  • Usability Testing: Designed and conducted tests to validate clarity, comprehension, and flow effectiveness.

  • Effort–Feasibility Matrices: Prioritized MVP features based on resource, time, and tech constraints.

These inputs shaped an agile design process:

  • Exploration: Compared rigid menus, guided prompts, and open-ended flows.

  • Prototyping: Built lean, testable interactions for education, diagnostics, and business guidance.

  • Validation: Conducted real-user testing for accessibility and tone alignment.

  • Documentation & Handoff: Captured flows, prompt logic, and edge cases to support future iterations.

  • Scaling: Deployed gradually, reaching over 3,000 interactions in the pilot phase.

Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #2
Project Gallery Image for 50% width of the screen #2
Project Gallery Image for 50% width of the screen #2

Reflections & Trade-offs

Designing with constraints fostered creativity and sharpened focus. Due to limited internal capacity, we adopted no-code tools and low-friction solutions to meet the program’s milestones and deliver a functional beta. EduBot was built using only what was available: Open AI GPT 4-o model, WhatsApp, no-code platforms, and a real user base for testing.

I focused on delivering a usable MVP—clear, efficient, and aligned with the grant’s goals. Prompts were refined to manage tone, guide the conversation, and optimize token usage, while documenting key insights to support future iterations. These limitations became the foundation for a solution with long-term potential.

A key challenge was the cognitive load that prompt creation placed on users with low literacy or digital fluency. When inputs were too open or vague, responses often felt generic—lowering the perceived value of the interaction. This led to several areas for future improvement, including:

  • Exploring speech-to-text functionality to reduce interaction barriers and improve accessibility

  • Revising legal copy at the start of the flow to avoid confusion or early drop-off

  • Introducing more guided or templated inputs to help users engage meaningfully without losing flexibility

EduBot is currently under review for further development and scaling. We’re analyzing interactions, flows, and prompt performance in collaboration with Data and AI teams to shape the next version. With the rapid evolution of LLM platforms and new research around WhatsApp-based conversational interfaces, voice input is again being explored, especially for its potential to support users with limited reading or writing skills.

EduBot will continue to offer free assistance to microbusinesses at the base of the pyramid in Colombia. The insights generated through this initiative will directly inform and enhance our specialized B2B product offerings.