Introduction
Bistrohunter is a chat service that simplifies restaurant search and booking in Spain through WhatsApp. The goal of this project was to improve the UX of an AI-powered restaurant recommendation chatbot.
Initial Approach
This project is an evolution of what started as an MVP, iterated and improved over time. Although the original product was B2B-focused and operated on a different platform, after listening to users and analyzing results, we realized it would work much better on WhatsApp and for a different audience. In the following iterations, we added layers of complexity, such as integrating AI to make conversations more natural and recommendations more automatic and scalable.
Research
For this stage, we followed multiple research approaches:
Interviews: Conducted several interviews to better understand users and their needs when using the chat.
Surveys: Used to confirm insights from interviews and track quantitative product metrics (NPS, ratings, etc.).
Data analysis: Examined database records on user requests (location, cuisine type, price range, etc.).
Conversation analysis: Studied user interactions to identify conversation flows and common patterns.
User Personas
Based on our research, we defined three types of users:
Travelers: People visiting a new city who don’t know any restaurants.
Explorers: Users looking to discover new places in their city or near their workplace.
Specific Seekers: Users searching for restaurants with particular features, such as gluten-free options or an elegant atmosphere.
Introduction
Bistrohunter is a chat service that simplifies restaurant search and booking in Spain through WhatsApp. The goal of this project was to improve the UX of an AI-powered restaurant recommendation chatbot.
Initial Approach
This project is an evolution of what started as an MVP, iterated and improved over time. Although the original product was B2B-focused and operated on a different platform, after listening to users and analyzing results, we realized it would work much better on WhatsApp and for a different audience. In the following iterations, we added layers of complexity, such as integrating AI to make conversations more natural and recommendations more automatic and scalable.
Research
For this stage, we followed multiple research approaches:
Interviews: Conducted several interviews to better understand users and their needs when using the chat.
Surveys: Used to confirm insights from interviews and track quantitative product metrics (NPS, ratings, etc.).
Data analysis: Examined database records on user requests (location, cuisine type, price range, etc.).
Conversation analysis: Studied user interactions to identify conversation flows and common patterns.
User Personas
Based on our research, we defined three types of users:
Travelers: People visiting a new city who don’t know any restaurants.
Explorers: Users looking to discover new places in their city or near their workplace.
Specific Seekers: Users searching for restaurants with particular features, such as gluten-free options or an elegant atmosphere.
Happy Path
Initially, we designed a simple flow for customer requests, but we quickly realized that real-world scenarios were more complex than expected. Ideally, users follow these four steps:
They send a message and accept the privacy policy.
They request a restaurant recommendation.
We provide a recommendation.
They request and manage a reservation (if they choose to).
Flowchart & Conversation Flows
Building on this happy path, we mapped out all possible variations users might encounter. Since this is a conversational interface, users interact in diverse ways—unlike UI-based platforms with buttons, text input allows for multiple phrasing variations.
Through experimentation, we found that the more guidance users receive on how to interact with the chatbot, the more likely they are to follow the intended flow. To address this, we introduced structured onboarding messages to educate users on how to use the chatbot effectively.
Messaging & Brand Tone
Beyond flow design, message content played a key role—especially in guiding users back into the defined paths. Maintaining a consistent tone aligned with the brand was crucial, as branding extends beyond visual identity to communications across social media, the website, and even chatbot interactions.
Implementation & Metrics
The development team focused on refining AI prompts for improved responses and setting predefined messages to maintain consistency.
To measure performance, we established key metrics, including reservation rates, user retention, and failure rates. Additionally, we continued to analyze chat conversations to identify new, unforeseen user behaviors.
Key Learnings
The main takeaways from this project were:
Understanding the limitations of conversational AI.
Recognizing how open-ended interactions significantly expand possible user flows.
Appreciating the importance of well-structured data for effective analysis.
Implementation & Metrics
The development team focused on refining AI prompts for improved responses and setting predefined messages to maintain consistency.
To measure performance, we established key metrics, including reservation rates, user retention, and failure rates. Additionally, we continued to analyze chat conversations to identify new, unforeseen user behaviors.
Key Learnings
The main takeaways from this project were:
Understanding the limitations of conversational AI.
Recognizing how open-ended interactions significantly expand possible user flows.
Appreciating the importance of well-structured data for effective analysis.
