PetPlate is a premium dog food subscription service where users complete a questionnaire during the subscription flow to receive a personalized meal plan. Based on their answers, users are recommended one of four plans: FreshBaked, FreshCooked, FreshCombo, or Toppers.
*Please reach out for an in-depth overview of the case study.
Client
PetPlate
Role
Product Designer
Year
2025
Results
A/B tests resulted in a 5% boost in conversions and a 35% reduction in drop-offs.
Significant improvement in recommendation algorithm accuracy.
60% reduction in time spent on page.
Problem: High Drop off Rates
Analysis of the subscription flow funnel metrics revealed that steps 10 (Plans page) and 11 (Entrées page) consistently produced the highest drop-off rates, averaging 30% and 22% month over month.
Hypothesizing the problem.
To guide our qualitative research, we first generated hypotheses based on the quantitative findings. What factors were causing the Plans and Entrées pages to consistently see elevated drop-off rates?
Decision Fatigue
After answering ten questions, users are overwhelmed by having to choose from four plans and eight entrées.
Too much cognitive load? Overwhelmed?
Recommendation Alogrithm
What system are we using to recommend the best plan to our users?
How accurate are our recommendations?
Finding 1: Time spent on page
Time spent on page
On average, users spent over 250% more time on these two pages compared to the rest of the subscription flow
Extended time-on-page indicates users are slowed down by the volume of choices presented, leading to decision fatigue.
Finding 1: Flawed recommendation structure
While PetPlate’s plans were built for different use cases, the algorithm focused solely on cost. As a result, a customer looking for premium fresh-cooked meals could be misdirected toward a baked plan.
Flawed recommendation
The recommendation algorithm was not grounded in research and defaulted to suggesting only the cheapest plan.
Finding 2: FreshCooked not recommended at all
Although FreshCooked accounted for the most plan checkouts, it was absent from the algorithm’s recommendations. The reason: its higher price. The reality: customers actively sought out and purchased FreshCooked despite the cost.
The graph illustrates calorie groups by the plan recommended (highlighted in yellow) versus the plan ultimately checked out (bolded in red).
Although customer data clearly favored FreshCooked, it was excluded from the recommendation UI. The decision reflected a business focus on new customer acquisition, with leadership concerned that higher prices might deter sign-ups.
Concerns
FreshCooked, which had the highest checkout rate, was notably absent from the recommendation feature.
The algorithm prioritized recommending the lowest-cost plan, but it was developed without any supporting user research or data-driven insights.
Finding 3: Data decided not data driven
PetPlate took a bold step in creating the recommendation system, but the approach was not data-driven. The team bypassed thorough quantitative analysis and relied primarily on qualitative assumptions.
The recommendation algorithm wasn’t driven by deep quantitative data. Instead, it was built largely on insights from a 2023 survey and ongoing email feedback from users.
Proper due diligence was not conducted in the research phase to develop a truly research-driven recommendation algorithm.
PetPlate’s rationale was that since customers cited price as the primary barrier to entry, the recommendation system should prioritize cheaper alternatives to encourage trial.
Concerns
FreshCooked, which had the highest checkout rate, was notably absent from the recommendation feature.
The algorithm prioritized recommending the lowest-cost plan, but it was developed without any supporting user research or data-driven insights.
The data made it clear, the current recommendation algorithm wasn’t working.
Recommendation Algorithm update
To support the design changes, we updated the recommendation algorithm to deliver more accurate and personalized meal plans to users. Through a series of UX interviews and the development of a new point-based scoring system, we created a more holistic algorithm (one that goes beyond simply recommending the cheapest plan). We also ensured that FreshCooked, which had previously been excluded despite high performance, was appropriately integrated into the new recommendation logic.
UX interviews -> Themes -> Plan Preferences
We conducted a series of UX interviews to understand why users selected their respective plans.
Using the insights gathered, we identified key themes for each plan and developed a new question featuring these themes as multiple-choice answers.
Scroll Rate: Information architecture
To ensure visibility, the Plans page had to present plan type, caloric needs, and key details within the 95th percentile of the scroll view.
Scroll Rate: Information Archtecture
Scroll rate studies were conducted to ensure all critical information appeared within the 95th percentile of the viewing area
New Landing Page
A new landing page was created, reducing page length by nearly 60%.
The addition of the new question and the restructuring of the subscription funnel created a more intentional system—one that leverages every question to inform calorie, entrée, and plan recommendations.
Along with refining the recommendation algorithm to match users with the most suitable plans, we merged the Plans and Entrées pages into a single view—showing the best plan and entrée together to minimize decision-making.
Selecting Change Plan takes the user through a single flow: first choosing a plan, then selecting an entrée.
Instead of a single sequence, users are given two separate options: Change Plan and Edit Entrées.
Users selecting Change Plan are taken into a dedicated flow, but entrée updates happen directly within the plan screen—no additional flow required.
Design 3 was decided as the first iteration of testing.
*Design 2 was a strong favorite. For more detail on why we chose to test Design 3 first, please feel free to reach out.
Results
A/B tests resulted in a 5% boost in conversions and a ~35% reduction in drop-offs.
Significant improvement in recommendation algorithm accuracy.
60% reduction in time spent on the page.
Conduct A/B test for Design 2
Conversion rates
Drop off rates
Checkouts vs. recommended plan
Time spent on page
