
AI Chatbot - Designing an LLM-powered shopping assistant for a leading European online pet retailer
Designing the personality, behaviour and conversational experience of an AI shopping assistant.
The AI assistant was designed to help customers make more better shopping decisions through natural conversations. From exploring nutritional information and checking product discounts to finding alternatives for out-of-stock items and discovering products based on their pet's age or specific needs, chatbot brought multiple shopping scenarios into one conversational experience.
I worked on designing chatbot's conversational experience from the scratch, collaborating with a Experience Designer and the AI team at zooplus. My contribution spanned LLM experience design, conversation architecture, system prompt design, AI behaviour design, intent and entity modelling, and user evaluation.
The focus was to create an assistant that was useful, consistent with the brand and intuitive for customers to interact with.
Project details
Product | AI chatbot assistant |
Company | zooplus |
Industry | E-commerce · Pet retail |
Market | EMEA |
Year | 2025 |
Technology | Large Language Model (LLM) |
Role | Senior UX/CX Content Designer · Conversation Designer |
Team | Experience Designer, AI team |
Status | Launched |
The challenge
Shopping for pets involves more than finding a product. Customers often need to consider their pet's age, nutritional requirements, product availability, discounts and suitable alternatives before making a purchase.
Traditional product discovery relies heavily on navigating categories, applying filters and searching through product listings.
The opportunity was to introduce a conversational shopping assistant that could understand customer requests and help them explore relevant products through natural language.
The challenge wasn't simply to make the assistant sound human. It was to define how it should communicate, interpret requests and behave across different shopping scenarios while maintaining the brand's personality.
What this AI chatbot could do
Customers often struggled to find suitable product alternatives when an item was out of stock or identify products that matched their pet’s specific needs.
The AI chatbot was designed to support customers across different shopping scenarios, helping them discover relevant products, explore alternatives and make more informed purchase decisions.:
Product discovery
Suggest products based on a pet's age or specific requirements.
Nutritional information
Help customers explore the nutritional value of products.
Discount discovery
Provide information about available product discounts.
Product alternatives
Help customers discover similar items when their preferred product was out of stock.
These capabilities informed the conversation architecture and interaction patterns I worked on.
My role: Designing the conversational experience from scratch
I collaborated with a experience designer and the AI team to shape conversational experience, connecting customer needs, brand personality and AI behaviour.
01. Research & discovery
Started with requirement discovery to understand the product objectives, customer needs and intended role of the assistant within the shopping experience.
This helped establish the foundation for its capabilities, personality and conversational behaviour.
02. Personality & identity design
Developed Zoomy's conversational personality and identity, defining how the assistant should sound and interact with customers while remaining aligned with the brand.
The objective was to establish a recognisable personality without compromising clarity, relevance or usefulness.
03. Conversation architecture
Worked on conversation architecture to structure customer journeys and define how different shopping scenarios should be handled.
Created reusable sample dialogues and conversation flows to demonstrate expected interactions across different customer requests.
04. Intent and entity modelling
Worked on intent and entity modelling to structure customer requests and identify the information required to support different conversational scenarios.
This provided the AI team with a clearer framework for understanding user intent and mapping requests to relevant shopping experiences.
05. System prompt design & AI behaviour design
Contributed to system prompt design and AI instructions, translating conversational requirements, personality principles and behavioural expectations into structured guidance for the LLM-powered assistant.
Collaborated with the AI team to establish response guidelines and define how Zoomy should communicate across different interactions.
06. Tone of voice & content guidelines
Developed structured tone of voice and conversational documentation in collaboration with the AI team.
These guidelines helped establish consistent communication principles and maintain the brand's personality across customer touchpoints.
07. Conversational UX & error handling
Designed conversation flows, reusable dialogue patterns and responses for scenarios where customer requests were unclear, information was incomplete or the expected conversation path changed.
The focus was to make interactions understandable and provide customers with a clear way forward.
Key design considerations
Making AI feel like the brand
An AI assistant is an extension of the customer experience. Its personality should be reflected not only in greetings but also in how it responds, explains information and handles different situations.
I collaborated with the AI team to document tone of voice, language principles and conversational behaviour.
Designing reusable conversation patterns
Different shopping scenarios often require similar interaction structures. I created reusable dialogues and conversation flows to establish consistency across customer journeys.
Designing for real-world inputs
Customers don't always communicate in predictable ways. I considered incomplete requests, ambiguous inputs and unexpected conversation paths when defining the assistant's behaviour.
Outcome
Zoomy was launched as an LLM-powered shopping assistant for the European retail experience.
My contributions helped establish its conversational foundations through:
LLM experience design and conversational behaviour guidelines.
Conversation architecture and reusable interaction patterns.
Intent and entity modelling.
System prompt design and AI instructions.
Chatbot personality and tone of voice documentation.
Error handling and conversational recovery.
User evaluation through 10 customer interviews.
The project brought together content design, AI collaboration and conversational UX to create a more natural way for customers to interact with a retail assistant.
Reflection
Zoomy strengthened my understanding of designing AI experiences at the intersection of content, technology and customer behaviour.
It reinforced that successful Large Language Model experience design goes beyond generating relevant responses. It requires thoughtful conversation architecture, clearly defined behavioural principles and an understanding of how people actually interact with intelligent systems.
One of my biggest takeaways was that customers don't necessarily approach AI with new behaviours. They often bring familiar expectations from existing experiences, such as search engines and product filters.
If I were approaching the project again, I would invest even more in continuous conversation evaluation, testing ambiguous inputs and understanding post-launch customer behaviour.
For me, human-centred AI starts with a simple principle: the value of an intelligent assistant isn't measured by how intelligent it sounds, but by how effectively it helps people achieve what they came to do.
Power in Numbers
30
Programs
50
Locations
200
Volunteers
Project Gallery


