Banco Falabella — Lía Chatbot
Designing a conversational assistant that turned complex financial services into guided, scalable interactions.
I helped Banco Falabella design and train Lía, a conversational assistant created to support users across banking, CMR, travel, and insurance services. My work focused on building clear conversational flows, structured intents, guided templates, and a scalable information architecture that reduced ambiguity for users and improved the chatbot's ability to respond autonomously.
Company
Banco Falabella Colombia
Industry
Finance / Retail banking
Role
UX Specialist / UX Designer
Product type
Conversational assistant / Chatbot
Timeline
Mar 2017 – Aug 2019
Team
UX, Product, Business Stakeholders, Development, Chatbot/NLP, Content
Methods
Conversational UX, IA, Intent Design, Flow Mapping, Templates, Training, Validation
Tools
Sketch, Zeplin, Balsamiq, DialogFlow, Adobe XD
Status
Launched
Some parts of this project are under NDA, so I'm showing a curated version of the process, artifacts, and outcomes. Sensitive conversation flows, business rules, and internal logic have been simplified while keeping the design decisions, methodology, and my role clear. I'm happy to walk through more detail privately in an interview or portfolio review.
Sensitive product details have been blurred or simplified to protect confidential business information.
Context
Banco Falabella needed a conversational channel capable of supporting users across multiple financial businesses: Banco, CMR, Viajes, and Seguros. Each business line had different needs, services, and user expectations, but customers needed one assistant that felt simple, friendly, and easy to use.
Writing good chatbot responses wasn't the hard part. The hard part was building a structure that guided users, kept the conversation from going fully open-ended, and helped the assistant answer accurately across very different kinds of requests.
Why it mattered
Financial services can become confusing very quickly when users are asked to explain everything in their own words. For Lía to work well, the experience needed strong information architecture, controlled conversation paths, and a friendly tone that helped users move forward without feeling lost.
My role
I worked as UX Specialist / UX Designer, with a strong focus on Conversational UX. My responsibilities went beyond writing chatbot messages: I helped structure the assistant's logic, define guided flows, train intents, and create a scalable methodology for improving the chatbot experience over time.
- Designed conversational flows for banking, CMR, travel, and insurance use cases.
- Structured intents, templates, and guided interaction patterns.
- Helped train Lía by identifying what information the chatbot needed and what could create confusion.
- Improved the chatbot experience by reducing open-ended questions and using more guided templates.
- Worked with business stakeholders to understand services, priorities, and expected user requests.
- Connected content structure, user needs, and technical chatbot behavior into a clearer conversational experience.
- Helped build a scalable Conversational UX foundation that shaped how I approach information architecture today.
The challenge
The main challenge was designing a chatbot that could support multiple financial services without overwhelming users or breaking the assistant's response quality.
- User challenge: users needed quick answers and support, but open-ended questions could easily lead to confusion, wrong interpretations, or dead ends.
- Business challenge: Banco Falabella needed one assistant that could serve different business lines while maintaining a consistent experience and supporting customer inquiries and transactions.
- Design challenge: the chatbot could not simply absorb unlimited content. Overtraining created poor responses, so the experience needed a clear structure, better intent organization, and guided interaction patterns.
Problem statement
How might we help Banco Falabella users get support across banking, CMR, travel, and insurance services through one friendly assistant, while keeping the conversation structured enough for the chatbot to respond accurately?
Goals & success criteria
User goals
- Help users find answers without navigating multiple channels.
- Reduce confusion caused by open-ended chatbot interactions.
- Make financial support feel more guided, friendly, and easy to follow.
Business goals
- Support multiple business lines through one conversational assistant.
- Increase the chatbot's ability to resolve inquiries and transactions autonomously.
- Create a scalable structure for adding and improving intents over time.
Design goals
- Define clear conversational flows and intent structures.
- Use guided templates to reduce ambiguity.
- Create a consistent tone and behavior for Lía.
- Improve chatbot training through better information architecture.
Process
01 — Discover
Understanding how users ask for help across financial services
I started by understanding the different business lines involved and the types of support users needed from each one. The first step was to identify where conversations could become ambiguous, where users needed guidance, and which topics required more controlled flows instead of open-ended questions.
- Reviewed the needs of Banco, CMR, Viajes, and Seguros.
- Identified recurring user requests, inquiries, and transactional needs.
- Mapped where open-ended conversations could create misunderstanding.
- Analyzed how information should be grouped to make the chatbot easier to train and easier to use.
- Looked for patterns that could work across business lines without making the assistant feel fragmented.
How I used AI
This was years before AI tools entered my workflow. Everything here came from UX analysis, information architecture, stakeholder input, and a lot of iterative chatbot training.
Early discovery structure used to understand how different financial services could live inside one conversational assistant.
02 — Define
Turning scattered services into a structured conversational architecture
After understanding the business needs and user requests, I focused on defining the structure behind the assistant. This meant organizing intents, deciding when to use templates, and shaping a conversation model that helped users make choices instead of forcing them to describe everything from scratch.
- Grouped intents by business line, user need, and expected outcome.
- Defined guided paths for frequent inquiries and transactions.
- Identified where open text input created risk and where templates worked better.
- Created a clearer content structure to avoid overtraining the chatbot.
- Helped define how Lía should behave when she did not understand a request.
How I used AI
No AI in this phase. Whatever value came out of it was from structuring the information carefully, cutting ambiguity, and turning business services into conversation paths the chatbot could actually handle.
Intent architecture used to organize multiple financial services into a single conversational model.
03 — Develop
Designing guided flows instead of relying on fully open conversations
The most important design decision was to avoid treating the chatbot as a fully open question-and-answer system. I learned that giving users too much freedom could make the experience less reliable. Guided flows and templates helped Lía understand the user better and reduced the risk of wrong answers.
- Designed conversational flows for key use cases.
- Created guided templates to help users choose the right path.
- Refined chatbot responses to feel friendly, clear, and useful.
- Adjusted flows when training results showed confusion or wrong responses.
- Balanced user freedom with the structure needed for better chatbot performance.
How I used AI
Again, no AI at the time. Deciding how much structure each conversation needed came down to testing, watching how the chatbot behaved in training, stakeholder feedback, and plain UX judgment.
Guided flow example showing how structured options reduced ambiguity compared with open-ended questions.
04 — Deliver
Launching a friendlier, more scalable assistant
The final solution was a conversational assistant that could support users across multiple financial businesses with a friendlier and more structured experience. Lía was designed to guide users through clear paths, manage different types of requests, and improve over time through a more intentional training methodology.
- Delivered conversational flows, templates, and intent structures.
- Supported chatbot training and refinement.
- Helped create a more consistent assistant behavior across business lines.
- Improved the quality of interactions by reducing overtraining and ambiguous inputs.
- Built a reusable Conversational UX foundation for future improvements.
How I used AI
No AI on this project, but it's part of why I insist on structure first now. Before I let AI speed anything up, I want the information architecture, the intent model, and the decision logic already sorted out.
Final conversational experience simplified for portfolio purposes to show structure without exposing sensitive business logic.
Before / after
Before
Multiple financial services were difficult to translate into one assistant. Open-ended inputs and overloaded training could make the chatbot respond incorrectly or create confusing interactions.
After
Lía used clearer flows, guided templates, and structured intents to support users across Banco, CMR, Viajes, and Seguros in a more consistent and scalable way.
Final solution
The final proposal focused on three main experience improvements:
A guided conversational model
Instead of relying only on open-ended questions, Lía used templates and structured paths to help users move forward with less confusion.
A scalable intent architecture
The assistant was organized around business lines, user needs, and expected outcomes, making the chatbot easier to train and evolve.
A friendlier financial assistant
Lía was designed to feel approachable and helpful, making financial support less intimidating and easier to navigate.
Impact & results
User value
Users could access support across multiple financial services through one guided conversational assistant.
Business value
Lía supported Banco, CMR, Viajes, and Seguros through a shared conversational structure.
Product value
The chatbot achieved a 70% autonomous success rate in customer inquiries and transactions.
Team / process value
The project created a stronger foundation for chatbot training, intent organization, and scalable Conversational UX.
Obstacles & trade-offs
This project required constant trade-offs between user freedom, business coverage, and chatbot reliability. A fully open conversation sounded flexible, but in practice it could make the assistant less accurate. The strongest solution was not to add more content, but to structure the experience better.
- Avoiding overtraining while still covering multiple business needs.
- Balancing guided templates with the user's expectation of natural conversation.
- Creating one assistant experience across four different business areas.
- Maintaining a friendly tone while handling financial information.
- Designing flows that could scale without becoming too complex.
What I learned
This project shaped the way I think about UX, even beyond chatbots. It taught me that structure is not a limitation; it is what allows a product to scale.
- I learned that information architecture is critical in conversational experiences.
- I learned that overtraining a chatbot can make the experience worse, not better.
- I learned that guided flows can create more confidence than fully open input.
- I learned that a friendly tone only works when the underlying structure is clear.
- I confirmed that good UX is often about reducing ambiguity before it reaches the user.
Lía is where I really learned to turn complexity into structure, and it's stuck with me since. A good conversational experience isn't built by adding more answers. It's built by designing the right paths, the right content structure, and just enough guidance for users to get where they're going.
Stephanie Cacheo — Senior UX Designer