Enter password to view
Turning experimental AI into scalable behaviour change
My role
Product Design Lead
Company
ZOE
Year
2024
Background
Meal logging is a central pillar of the ZOE experience, enabling members to track what they eat and receive personalised health insights. Over time, however, logging had become increasingly cumbersome - requiring users to manually construct meals by searching for individual ingredients.
The team had begun exploring large language models (LLMs) and optical character recognition (OCR) to simplify this process, particularly for homemade meals and recipes from books or websites. While early releases demonstrated strong technical potential, their real-world impact was limited. Members struggled to discover the features, understand when to use them and integrate them naturally into their everyday logging habits.
The core challenge wasn’t AI capability - it was driving behaviour change.
I led the design of a streamlined AI-assisted logging experience and activation strategy, focused on embedding these tools directly into natural user flows, accelerating time-to-value and laying the foundation for more passive forms of logging in the future.
Framing the problem
Early AI experiments demonstrated that meal logging could be dramatically simplified. However, low discovery and unclear first-time experiences meant that only a small subset of members were benefiting from the new capabilities.
Rather than continuing to ship isolated AI features, I led a focused discovery phase to understand where adoption was breaking down - combining customer interviews, behavioural data analysis and a deep review of the existing logging experience across key scenarios. This surfaced high-leverage behavioural opportunities:
Awareness and first-time value - members didn’t naturally discover or understand when to use the tools
Search as the dominant (and broken) entry point - many logging journeys stalled when meals or recipes couldn’t be found
A long-term shift toward passive capture - strong demand for logging by photo rather than manual input
The strategic goal became clear: move logging from manual construction toward progressive AI-assisted capture, while making value immediately obvious in the moments members already used.
Shaping the solution
I aligned the team on a multi-pronged approach to drive adoption and long-term behaviour change, focused on embedding AI into the moments members already used rather than treating it as a standalone capability.
1. Progressive activation through onboarding
Guide members to early ‘wow moments’ by walking them through high-value logging actions, reducing reliance on passive feature discovery.
2. AI embedded directly into search
Turn failed searches into moments of creation - enabling members to generate meals or recipes instantly when content wasn’t available.
3. Laying foundations for passive logging
Explore image-based capture as the long-term evolution of meal logging, reducing effort further over time.
Rather than delivering all changes at once, we deliberately sequenced these bets - prioritising the interventions most likely to shift behaviour and validating impact through experimentation before scaling.
Insights from behavioural data were continuously paired with member interviews to understand not just what moved metrics, but why - allowing us to course-correct quickly and compound gains over time.
Driving discovery and first-time value
To accelerate adoption of AI-assisted logging, we introduced a new ‘Get more from logging’ experience - a progressive onboarding flow designed to guide members to early high-value actions across the logging journey.
This was anchored by a simplified logging screen that:
Clarified entry points for different logging scenarios (recipes, websites, homemade meals)
Connected common use cases directly to relevant AI-powered features
Reduced friction between discovering a capability and successfully using it
Alongside this, we embedded lightweight, in-context education within each flow - using example content to help members get started without pulling them out of task.
Through user testing, we deliberately deprioritised a full guided product tour. While initially appealing, it didn’t meaningfully improve activation and introduced friction at moments where speed and momentum mattered most.
Focusing instead on task-driven activation significantly reduced time-to-value and drove higher adoption of AI-powered logging tools.
Turning search dead-ends into AI creation
Search was the most common starting point for meal logging - and also the most frequent point of failure. Members regularly couldn’t find meals or recipes and were forced into manual construction, creating significant friction.
To remove this bottleneck, we embedded AI directly into the search experience. When results didn’t return useful matches, members could instantly generate a meal or recipe using AI - turning dead ends into moments of creation and dramatically reducing effort.
Through iteration and testing, we refined how this capability was surfaced:
Evolving from a simple prompt button to a richer contextual card
Clearly communicating what the AI would generate using the member’s query
We also learned that in-search educational content reduced conversion. In a task-focused flow, members prioritised speed over learning. As a result, we shifted education for this feature into the broader onboarding experience instead.
This change drove a 41% increase in usage of the AI meal generation feature.
Building the future of logging through images
Strong member demand and discovery insights pointed to image-based logging as the natural long-term evolution of meal capture - reducing effort further and moving toward more passive tracking.
I led early work to de-risk this strategic shift, designing an image-based experience that allowed members to photograph meals and automatically generate ingredient lists.
In parallel with foundational product design, we ran a focused technical proof of concept to validate:
Image-to-ingredient accuracy
Generation speed
Overall feasibility at product quality standards
Once viability was confirmed, I partnered closely with product and engineering on a rapid prototyping phase, shaping an experience that:
Captured meals in seconds
Generated editable ingredient lists
Significantly simplified saving recipes
Early user testing showed strong engagement and clarity, and the work directly informed roadmap prioritisation - positioning image-based logging as a core future initiative.
Outcomes
By shifting focus from isolated AI features to behaviour-driven design, we transformed experimental capability into meaningful member value.
Progressive onboarding, embedded AI in search, and clearer first-time experiences led to:
Significantly higher adoption of AI-powered tools
Improved overall logging conversion
Faster time-to-value for members
Beyond immediate metrics, this work established a clear strategic direction for logging - moving from manual construction toward increasingly passive, AI-assisted capture.
The foundations laid through this project directly shaped ZOE’s future roadmap, positioning image-based logging and deeper AI integration as core long-term investments.
Previous
Next