Personalization Engine — StyleHive
Visual: recommendation widgets, dynamic banners, and lifecycle email examples.
Client Overview
StyleHive is a niche US e-commerce startup specializing in curated independent fashion brands. Rapid customer acquisition highlighted gaps in retention and revenue per user; StyleHive wanted a unified personalization platform to increase repeat purchases and improve customer lifetime value.
Challenge
StyleHive had siloed recommendation widgets provided by third-party tools and inconsistent lifecycle messaging. The business wanted a single source of truth for personalization with control over experiments and business rules to balance discovery and margin.
Solution — Unified Personalization & Experimentation Platform
We built a modular personalization engine that unifies user signals (browsing, purchases, returns), serves contextual recommendations across touchpoints, and supports A/B/C experiments with business-rule overrides to protect margins.
Core features
Approach
- Define KPIs (repeat rate, AOV, email CTR) and map current experience gaps.
- Implement feature pipelines and offline model training for ranking candidates.
- Expose personalization API consumed by web widgets, mobile SDK, and email templates.
- Run rapid experiments and incorporate guardrails for margin and inventory constraints.
Technology stack
Implementation — Delivery
Phase 1 — Signals & Baseline (Weeks 1–6)
Aggregate browsing and purchase signals, build baseline heuristics, and define experiment cohorts.
Phase 2 — Ranking & APIs (Weeks 7–14)
Train ranking models and expose low-latency APIs for widgets and email personalization.
Phase 3 — Experiments & Guardrails (Weeks 15–22)
Run controlled experiments, implement margin/inventory constraints, and iterate on content templates.
Phase 4 — Scale & Automation (Weeks 23–26)
Automate model retraining pipelines and lifecycle orchestration for ongoing growth.
Impact & Results
24%
Increase in repeat purchase rate
14%
Lift in average order value for personalized sessions
32%
Email revenue increase (targeted sequences)
Weeks
Time to run controlled experiments
Qualitative outcomes
- Business could prioritize experiments tied to margin outcomes rather than raw click metrics.
- Unified feature store reduced duplication and improved model freshness.
- Personalized lifecycle messaging reduced churn in early customer cohorts.
Client Testimonial
Key Highlights & Learnings
- Experimentation that ties to revenue preserves business focus and avoids vanity metrics.
- Guardrails (inventory, margin) are essential for production-ready personalization.
- Feature freshness (session-level) is a major driver of recommendation relevance.