Food AI Recognition: Snap the meal to get the nutrition fact

Designing end-to-end features and conducting user research

Project Overview

Objectives

To assess launch readiness based on user outcomes and feedback, and identify pain points or drop-off risks ahead of the public launch.

Challenge

The two-month timeline made it challenging to continuously gather user feedback and track engagement without disrupting their experience.

Duration

Nov. 2024 - Jan. 2025 2 months

Outcome

Identified 5 major pain points and proposed solutions achievable within one month before launch.

My Role & Deliverables

Researcher

  • Planned and executed the entire research strategy, including qualitative interviews, diary studies, quantitative surveys, and app data analysis.

  • Analyzed interview results for product decisions.

UX Designer

  • Design the mechanism to enhance the sense of feedback upon task completion.

Background

Tracking meals can increase diabetics’ awareness of food choices and help healthcare providers offer more precise guidance. However, diet log users only make up 10% of our monthly log users. Furthermore, we have consistently received complaints of our current meal-logging options.

Here are the problems of our current meal-logging options:

Food Database

Users are unsure which nutritional information is accurate, and it is difficult for them to determine portion sizes in grams.

Food photo

Simply taking a photo of the food does not provide any nutritional information, such as calories and the three macronutrients.

Food category

Users need to have a certain level of health knowledge to convert portion sizes into servings.

Design Process

1. user Interviews × 12 : Understand user behaviour of diet log and gather feedback on prototype

Acted as a researcher, planned research, conducted interviews and spotlighted insights

4. Deliver new feature: From wireframe to UI design

Served as an UI/UX designer, worked closely with PM, data analyst and dev teams to launch the feature

2. Design iteration: Optimize the design and align product team’s expectations

Acted as a designer, iterated design based on research findings

5. Track metrics: Measure success

Define prodcut metrics with PM and built a dashboard to monitor metrics

3. intercept interviews × 21: Understand users’ preference for 2 design solutions

Acted as a researcher, planned and conducted interviews.

6. Distribute satisfaction survey: Identify issues

Served as a researcher, drafted the survey and proposed optimization solutions

12 User Interviews

Key Research Findings

Research goal 1: Understand user behaviour of diet log and motivation

We concluded that only those who need insulin injections or weight control are motivated to track calorie information. Additionally, we found that the attitudes of healthcare professionals and encouragement from the community are the most significant driving forces.

We defined four personas and user journeys based on users' level of health knowledge and the nutritional information they require. This analysis helps us align product goals with our team.

Research goal 2: Identify usability issues of existing features

Through testing 2 usability tasks, we identified several issues with our current tracking methods. The most surprising findings are the low discoverability of our entry points, and the non-intuitive search logic of the diet database. Many high-priority issues have already been addressed right after the research report.

Research goal 3: Gather user feedback of food recognition

We evaluated whether this feature could increase the tracking frequency and discovered that users want to estimate calories and food portions. Additionally, we asked users to rank the importance of various nutritional information.

21 intercept Interviews

Key Research Findings

Research goal : Understand users’ preference for 2 versions of solutions

After gathering the first round of feedback, we developed two design versions. The primary difference is the summary information: one displays an overall health score, while the other highlights carb and calorie details. To quickly gather substantial feedback, we directly visited clinics to collect insights.

Final Design Solution

Final Design Solution

Key frames of the food recognition flow

Design Details

Project Results

The proportion of users logging diets increased by 2%, and the percentage of paid members rose by 7%. However, the conversion rate of the result page is lower than expected.

Investigation

We send out satisfaction surveys to understand the breakpoints. We found that, in addition to dissatisfaction with recognition accuracy, 60% of users did not know how to modify the recognition results.

Iteration

We allow users to add foods and enable modifications for incorrectly recognized foods through food database. This adjustment led to a 15% increase in the conversion rate.