Role: Product Manager | Personalization | Recommendation Systems | AI | B2C
Background > Problem > Use Cases > Personas > Phase 1 > Business Value > Phase 1
Background
Problem: Friction Points in Finding the Right Products
Shopper Frictions:
Shoppers who have Sprouts mobile app account with dietary restrictions have no reliable way to shop the Sprouts app confidently. Every product requires manual label checking.
No dietary filtering, no health-aware search, and no personalized guidance exists in the app today.
The in-store trust Sprouts has built around health and nutrition does not carry over to the digital experience.
Business Frictions:
Without a health personalization foundation, Sprouts' digital experience is indistinguishable from any conventional grocery app — failing to deliver on its health-forward brand promise. Three business risks stand out:
Purchase conversion for dietary-restricted shoppers sits at ~9% (Industry benchmarks), well below what a personalized experience should achieve for this high-intent segment.
Repeat shoppers default to the same items every order, leaving higher-margin Sprouts Brand products undiscovered by the customers most likely to buy them.
Health-focused competitors like Thrive Market are investing in personalized discovery, putting Sprouts at risk of falling behind its own brand positioning.
AI recommendations need three things to work: clean product data, dietary filtering rules, and a saved shopper health profile. None of these exist today. Phase 1 builds all three so that when the AI runs in Phase 2, it runs safely and accurately. Prototype with what Phase 2 implementation will look like included.
Phase 1: Build the data foundation: ingredient parsing pipeline, dietary rules engine, and health profile storage to support all downstream AI features.
Phase 2: Ship the core AI layer: health match scorer, dietary filtering, health-filtered search, and the For Your Goals personalized discovery rail.
Phase 3: Deepen personalization: hybrid recommendation ranker, smart cart health check, health-safe substitution engine, and personalized deal discovery.
Phase 4: Scale and optimize: model retraining cycles, multivariate experimentation, Sprouts Brand ranking tuning, and 12-month performance review against defined success metrics.
Business Value:
Value Driver
How Phase 1 Enables It
New item discovery
Health profile data enables Phase 2 to surface new or unfamiliar items that match a shopper’s goals and expand basket size beyond the default 10–15 repeat items.
Browse-to-cart conversion lift
Accurate dietary filtering removes the need for manual label-checking that causes health shoppers to abandon browse sessions. Phase 1 makes filtering possible
Sprouts Brand private label attach rate
AI recommendations in Phase 2 can surface Sprouts Brand items within health-compliant results. Phase 1’s product data processing tags Sprouts Brand items for this logic.
Phase 1: Components to Build
Product Data Processing Pipeline
As a shopper, I want product ingredients automatically processed and structured so that the app can accurately identify which products match my dietary needs without me checking every label.
Each product record includes parsed ingredients, dietary flags, nutritional data (calories, protein, carbs, fat, sugar, fiber), certifications, and a Sprouts Brand indicator.
Dietary Filtering Logic — Rules Engine
As a shopper, I want a dietary filtering layer applied before any recommendations surface so that I never see products that conflict with my health profile.
User Health Profile Storage
As a shopper, I want my dietary restrictions saved in the app so that every recommendation is personalized to what is safe for me to eat.
Once the foundation is in place and Grovewise launches across Phase 2 and Phase 3, success is measured by three metrics:
Purchase conversion for dietary-restricted shoppers improves by 25–35% based on industry benchmarks for health-personalized grocery experiences.
45% of monthly active users complete a health profile within 90 days of launch.
**Published onboarding activation benchmarks from Appcues and Mix panel
Shoppers discover 4 new items per order on average, up from 1.3 today.
Final Design and Thoughts
If I had more time on this capstone I would focus on:
Validating the health match scoring weights with real Sprouts loyalty purchase data to confirm that shoppers with specific health goals actually buy the products the model scores highest and refine the weights based on real behavior rather than expert estimates alone.
Fully speccing the Phase 3 personalization ranker, including how it handles new users with no purchase history and how it adapts when a shopper changes their health goal mid-journey.
Adding a detailed A/B test design for Phase 2 launch: sample size, traffic split, test duration, and decision criteria and to go beyond defining metrics and show how the experiments would actually be run.