
Bella chatbot - Bridging conversation design, UX and AI to simplify complex HRMS workflows.
Designing intelligent interactions that help people get things done
Project overview
Bella was my first chatbot project, back in 2021, and my introduction to the world of conversational AI.
It was an opportunity to move beyond traditional UX writing and think about how people interact with products through conversations. I worked across conversation architecture, chatbot personality, user journeys and technical handoff, collaborating closely with product and engineering teams to bring Bella to life.
Looking back, this project shaped how I approach conversational experiences: starting with user intent, understanding the system behind the interaction and designing for everything that can happen between a user's first message and task completion.
Bella was an AI-powered conversational assistant built for a leading AI-first, full-suite HRMS and payroll software platform.
The idea behind building Bella was to give employees, managers and HR administrators a simpler way to access HR services without having to navigate multiple sections of the platform. Through a conversational interface on the web application, users could perform everyday tasks such as applying for leave, filing income tax, requesting documents and managing their teams.
My role: Conversation Designer & UX Content Designer
My involvement: Product discovery, experience design, conversation architecture, prototyping, UX writing and engineering handoff.
Tools: Botmock, Botfront and conversation flow documentation.
Collaborators: Product Owner, Product Design, Engineering and AI teams.
The challenge
A full-suite HRMS platform comes with a lot of functionality. Employees, managers and HR administrators use it for different purposes, from managing leave and employee documents to handling payroll, taxes and team-related activities.
But having all these services in one platform doesn't necessarily make them easy to find or use. Users still need to understand where to go, which workflow to follow and what information is required to complete a task.
Bella was introduced to simplify these interactions through conversation.
The challenge was to design an assistant that could do more than answer questions. It needed to understand user intent, collect relevant information, navigate different HR workflows and guide users towards completing tasks.
This meant thinking beyond individual chatbot responses and designing the underlying conversation logic, interaction patterns and exception handling that would make the experience work.
My role
I worked on Bella from the early product discovery and experience design stages through to conversation design and engineering handoff.
My role involved translating product requirements and HR workflows into conversational experiences that were intuitive for users and structured enough for engineers to implement.
I collaborated closely with the Product Owner, product designers and engineering teams, contributing to both the chatbot experience design and its conversational architecture.
My approach
01. Bringing product and engineering together
I established daily stand-ups with the Product Owner and engineering team to align on requirements, discuss dependencies and review the progress of Bella's conversational experience.
These sessions were important in understanding the underlying HR workflows, clarifying product behaviour and identifying technical considerations early.
They also helped us make decisions together rather than treating conversation design as a separate activity that happened before development.
02. Defining Bella's personality and conversational identity
I wanted Bella to have a recognisable personality, not sound like a generic HR chatbot.
I used the Myers-Briggs Type Indicator (MBTI) as a framework to explore personality traits and help define how Bella should communicate, respond and interact with users.
To bring everyone onto the same page, I also conducted internal tone-of-voice workshops with stakeholders and team members. These sessions helped us discuss the kind of assistant we wanted Bella to be, how she should come across in different situations and what her communication should feel like.
Based on these discussions, I worked on establishing Bella's conversational identity, including:
Personality traits and communication style.
Tone of voice and language principles.
Response patterns across different HR scenarios.
How Bella should respond to user frustration, confusion and incomplete requests.
Consistency between her personality and the actions she performed.
The aim was to create an assistant that felt approachable and helpful without losing the clarity and professionalism expected from an HR platform.
03. Building the conversational architecture
I translated product requirements into a structured conversational framework using intents, entities and conversation logic.
This involved understanding:
What users might ask Bella to do.
How different requests could be expressed in natural language.
Which intents Bella needed to recognise.
What entities and information she needed to collect.
How user inputs mapped to specific HR workflows.
What should happen when a request was incomplete or unclear.
I worked with the AI and engineering teams to align the conversation structure with the product's underlying functionality.

Building a consistent conversation framework
I defined a set of reusable conversation components to guide how Bella interacted with users across different HR scenarios. These included greetings, questions, information prompts, quick-reply chips, confirmations, suggestions, acknowledgements and error-handling responses.
The aim was to make conversations feel consistent and predictable while helping users understand what Bella could do, what information was needed and how to move forward when something went wrong.

04. Designing and prototyping conversation flows
This was a significant part of my work.
I used Botmock and Botfront to map and prototype end-to-end conversation flows for key HR scenarios.
Rather than writing standalone chatbot responses, I designed complete interactions, thinking through the different ways a conversation could unfold.
My flows covered:
User entry points and intent recognition.
Sample utterances and variations in user requests.
Entity collection and information validation.
Follow-up questions and clarification prompts.
Conditional logic and branching paths.
Confirmation and task completion messages.
Error handling and fallback scenarios.
Using these tools helped me make the conversation logic visible and easier to review with product and engineering teams.
More importantly, the flows served as working documentation for engineering handoff. Engineers could refer to the mapped interactions, understand the expected behaviour and use them during implementation.
05. Designing for exceptions and unexpected inputs
Not every user interacts with an assistant in the same way. Some provide incomplete information, others phrase requests unexpectedly or change direction midway through a conversation.
I worked on identifying these scenarios and designing alternative conversation paths.
This included clarification prompts, error messages, fallback responses and recovery journeys for situations where Bella couldn't understand a request or complete the expected action.
The focus was on making the experience resilient, so users weren't left at a dead end when a conversation didn't follow the expected path.
06. Working closely with engineers on implementation
I collaborated directly with engineers throughout the process, explaining conversation flows, clarifying interaction logic and resolving questions around expected behaviour.
I also contributed to the experience design of Bella's chatbot interface, considering how the conversations would translate into the actual web application experience.
This close collaboration helped ensure that what we designed wasn't just a collection of ideal conversation scenarios, but something that could be built and integrated into the HRMS platform.
Key design considerations
Different roles, different capabilities
Bella supported employees, managers and administrators. Each group had different responsibilities, permissions and access to HR services. The conversational experience needed to account for these differences and guide users towards the actions relevant to their roles.
Moving from intent recognition to task completion
Understanding what a user wanted was only the beginning. Bella needed to gather the right information, ask relevant follow-up questions and guide users through the steps required to complete an HR task.
Managing conversation complexity
HR workflows often involve multiple inputs, conditions and dependencies. I focused on structuring these interactions into manageable conversation steps without overwhelming users.
Designing beyond the happy path
A useful assistant needs to handle ambiguity, missing information and unsupported requests. I considered alternative paths and recovery patterns to make conversations more predictable and consistent.
Maintaining a consistent assistant personality
Across different HR scenarios, Bella needed to sound like the same assistant. Tone of voice and response guidelines helped maintain consistency across conversations.
The outcome
Bella brought a conversational layer to a full-suite HRMS and payroll platform, giving users another way to access everyday HR services through the web application.
My work helped establish the foundations of its conversational experience, from personality and intent structure to end-to-end conversation flows, exception handling and implementation documentation.
A measurable improvement in task completion
Following Bella's launch, the task completion metric increased by 55% within three months.
This was an important indicator of the experience's progress and reinforced the value of making complex HR workflows easier to navigate through conversation.
What I took away
Bella helped me understand the depth of work that goes into designing a functional AI assistant.
The conversation is only the visible part of the experience. Behind it are user intents, entities, business rules, conditional paths, system capabilities and all the situations where things might not go as planned.
It also reinforced the importance of bringing content design, product and engineering together early. The best conversation flows are not just well-written. They are grounded in how the product actually works and designed with implementation in mind.
That foundation continues to influence how I approach conversational AI and more complex agentic experiences today.
Power in Numbers
30
Programs
50
Locations
200
Volunteers
Project Gallery

