KIWI-PLUS
How we designed an AI-powered dinner service for 280,000+ monthly active users, and the strategic pivot that made it scale.
Project information
My role
UX Lead
Team
1 Product Owner
2 full-stack developers
2 data scientists
Product
AI-generated dinner recipes for KIWI-PLUS app users
The assignment
​KIWI wanted to help their customers answer one of everyday life's most common questions: what should we have for dinner tonight? The goal was to develop a personalised dinner service in the KIWI app that suggests relevant meals based on each user's purchase history from the Trumf loyalty program, using an AI model. The service needed to feel simple, inspiring and personalized.
Project overview
My Role
UX Lead responsible for the full design process from research to delivery. Working closely with the product owner, developers, data analysts and business stakeholders in a cross-functional Scrum team.
​
Process
We started with qualitative interviews and surveys to understand user needs. The research showed that people want simple, familiar everyday meals. We ran inclusive workshops with the team to design the preference and filter model, using card sorting and collaborative prioritisation to decide what had to be included at launch.
​
The AI model was tested across two rounds of beta testing. The first round revealed that individually tailored suggestions were not scalable. We moved to a cluster-based personalisation model, grouping users with similar purchase patterns, without compromising the feeling that suggestions were made just for them.​
Results
650
beta users validated the core hypothesis
74%
Perceived recommendations as relevant
64%
Reported increased decision confidence
73%
Preferred recognizable product imagery over finished dish photography
75%
Found the suggested products relevant to their habits
10x
Scalability through cluster-based personalization
without reducing perceived relevance