Plant Care Through Observation
Problem
I didn’t start this project from a deep interest in plants.
It began as part of the Google UX Design course, where plant care was the most accessible theme as I entered product design.
To understand the space, I explored existing plant care apps and discovered that users still experience issues with diagnosis functions.
Most apps try to reduce uncertainty through plant identification or diagnosis features. Some guide users through multi-angle photo capture and additional inputs to improve accuracy.
However, across the apps I observed, the same issue persisted: users reported inconsistent results, incorrect diagnoses, and care advice that sometimes worsened plant conditions.
Initially, I treated this as a UX problem—unclear flows or insufficient guidance for capturing photos.
Reframing the problem
Plant health cannot be understood from a single moment.
Through further research, I realized the issue was not only UX—it was the approach itself.
Plant health depends on change over time, not a snapshot.
- Plants vary by age, light, season, and environment
- Visually similar symptoms can have different causes
- Even experienced growers avoid single-image judgments
The limitation was not interface clarity—it was attempting certainty from incomplete context.
Diagnosis creates false confidence
Diagnosis systems translate uncertainty into authoritative outputs.
The flow feels complete: capture → analysis → result.
This creates perceived certainty without corresponding understanding.
- The system implies better input produces correct truth
- Outputs are framed as definitive rather than contextual
- Uncertainty is hidden rather than communicated
Plant health, however, is temporal and evolving. A single-frame model reduces this complexity into oversimplified answers.
Why observation is the solution
Existing and potential users
I interviewed beginner gardeners, plant-curious users, and past plant care app users. I also analyzed app store feedback.
Three consistent patterns emerged:
1. Fear of making mistakes
Users acted from uncertainty—overwatering, underwatering, or missing early signs.
2. Emotional attachment
Plants were described as relationships, not objects.
“I feel connected with my plants when I sit in the sunlight with them.”
“They relax me when I see them.”
“I think of my plants like my babies.”
3. People don’t seek experts first
When problems appeared, users rarely went straight to experts. They asked friends, searched online, or checked communities first.
Despite this, most apps positioned themselves as authoritative answer systems.
People who don’t rely on apps
Experienced gardeners show a different model of care.
The difference is not knowledge—it is attention.
They:
- Observe continuously
- Detect subtle changes
- Adjust based on patterns, not rules
“It feels dry today. I’ll wait one more day.”
Care becomes pattern recognition over time, not task execution.
Confidence is accumulated, not delivered.
Product direction
The research shifted the focus away from diagnosis and toward confidence through observation.
Instead of answering:
“What is wrong with my plant?”
The product supports:
“What is changing over time?”
By making a plant’s story visible over time, users can:
- Bring previous moments into the current view
- Compare past and present states
- Compare their own with external stories
What this provides
- More context around the current state
- Better understanding of changes over time
- A stronger sense of connection and attentiveness
Over time
Observation → History → Context → Confidence
This also meant moving away from optimizing for diagnosis, plant identification, or expert-like prescriptions. Those approaches focus on providing answers.
This product focuses on helping users build understanding through observation and context over time.
The goal is not to tell users what’s wrong with their plants.
It’s to help them understand their plants better.
Monetization Model
As the product shifted from diagnosis to accumulated context, it raised a monetization question:
If the value comes from context rather than answers, what should users actually pay for?
Charging for diagnoses didn’t feel right. The product’s value wasn’t in answers, but in the context built over time.
The idea became clearer when I thought about how iOS Photos works. People technically pay for storage, but storage alone is not what makes Photos valuable. Features like Memories, People & Pets, and contextual search help users make sense of everything they’ve already captured.
The value isn’t the data itself. It’s the organization of that data.
The same applies here.
Free
If the product’s value is context, then continuity and relationship-building remain free.
- Full timeline
- Bring previous moments
- Photos and notes
- Logged care actions
- Community access
Paid
The paid layer focuses on organizing accumulated context. It doesn’t create new information. It reduces the cognitive load of understanding information that already exists.
- Collapsed timelines (stress, stability, recovery)
- Generated previous moments (you saw something similar / your moments tell)
- Long-term context preservation
- Story filtering
- More plants over time
The product offers continuity for free and organization for pay.
External Patterns & Internal Patterns
External patterns come from community observations and provide useful reference points before enough personal history exists.
Example: Yellowing often appears 5–7 days after relocation.
Internal patterns come from a user’s own plant history.
By combining system signals with past observations, the app can detect recurring relationships between actions, conditions, and outcomes.
System Signals + Plant History = Internal Patterns
Final Logic
Each layer builds on the one before it:
Observation → History → Context → Structure → Confidence
- Free builds the first two layers (observation + history)
- Paid enhances the later layers (context+ structure)
The product is not designed to answer plant problems.
Testing
I tested key assumptions with five participants and iterated prototypes.
Observation flow vs diagnosis flow
I compared two flows using a yellowing-leaf scenario.
The diagnosis flow:
- Scanning-based capture
- Diagnosis results
- Guidance frames
The observation flow:
- Removed guidance frames that hinder observation
- Added hidden frames to keep the focus on the plant itself
- Removed verdict-like outputs
- Added previous moments from the plant’s history
Result
Participants felt observation flow helped them spend time with their plants, but preferred diagnosis because it provided immediate answers, direction, and reassurance.
Initially, this seemed to challenge the direction. After review, I realized the issue wasn’t the observation flow, but the scenario.
Yellowing leaves signal urgency, where users naturally want immediate guidance.
This revealed that diagnosis and observation serve different needs:
- Diagnosis for urgent, problem-solving moments
- Observation for noticing, documenting, and understanding change over time
Rather than treating this as a compromise, diagnosis was designed as an extension of the same temporal system. Previous moments were included in diagnosis results so users could still understand current conditions within the context of change over time.
When retested with a blooming scenario, participants clearly distinguished the two paths and found revisiting past moments meaningful.
Comparison gallery
I explored whether temporal comparison is effective for understanding plant history.
Participants viewed current plant states alongside historical moments and they identified changes more clearly.
This indicated that comparison over time improves perception of plant development.
Original timeline vs Paid model
I was unsure whether users would value a the paid model (collapsed timelines and previous moments attached to story)enough for it to become part of a paid model.
I tested original timeline and paid model. Participants responded positively to both. Some preferred detailed chronological context, while others preferred summarized phases.
Rather than choosing one, I designed both views as complementary perspectives.
This approach aligned better with the product goal of supporting understanding rather than enforcing a single structure, while leaving room to further evaluate its long-term value within the product.
Onboarding
A key risk was conceptual clarity, given most plant apps focus on diagnosis. I introduced observation as the core behavior.
All participants understood the concept after onboarding, which gave me confidence to move forward with the direction.
Design decisions
Co-observing layer
I introduced co-observing companions, as the idea of a plant’s “story” became central. They are not decorative characters—they are embedded into decision points.
Capture entry points
Core states:
- Looking good
- Need help
- Something new
- Same as before
Instead of acting as overlays on capture entry points, companions shape interpretation at the moment of decision.
This shifts interaction from selecting an option to framing perception.
Plant current state
When users record a story, the system generates plant state sentences from user inputs. These reflect the actual meaning of those inputs within context.
Visual system
Corner radius system
Because co-observing characters appear within cards, the radius system avoids ambiguous nesting between similarly rounded elements, which can blur visual hierarchy and create tension.
Characters sit within containers that either clearly match or clearly contrast their geometry, such as large surfaces (38px) or fully rounded forms.
Mid-size cards echo the character’s rounding language, while smaller components scale down using a golden-ratio-based hierarchy.
- Character surfaces: 24
- Large cards: 38
- Mid cards: 24
- Small cards: 15
Typography
Typography with minimal character is chosen to balance expressive UI elements.
- Poppins: headings
- Inter: body and labels
Card system
The interface is built around a unified card system spanning present and past states.
The primary view represents the current plant state. Previous moments exist within the same structure as expandable states. It is not a separate layer, but a continuation of the same system.
This preserves continuity across time, allowing users to move between moments without changing mental models.
