LOCAL PREVIEW View on GitHub

14. MVP vs Future Versions - Phased Rollout

Release Timeline

gantt
    title MangaAssist Release Roadmap
    dateFormat YYYY-MM
    axisFormat %b %Y

    section MVP (V1)
    Design and Architecture     :a1, 2026-03, 1M
    Core Development            :a2, after a1, 2M
    Testing and QA              :a3, after a2, 1M
    Beta Launch (1% traffic)    :milestone, m1, after a3, 0d
    Gradual Rollout (100%)      :a4, after a3, 1M

    section V2
    Post-Purchase Features      :b1, after a4, 2M
    Personalization             :b2, after a4, 2M
    V2 Launch                   :milestone, m2, after b2, 0d

    section V3
    Proactive Intelligence      :c1, after b2, 3M
    Multi-language              :c2, after b2, 3M
    V3 Launch                   :milestone, m3, after c2, 0d

MVP (V1) - Useful on Day One

Timeline: About 4 months to beta and 5 months to full rollout.

Goal: A chatbot that can answer the most common manga shopping questions and give basic recommendations.

What's Included

Feature Description Why in MVP
Product Discovery "Show me popular shonen manga" Highest-value use case
Basic Recommendations "Something like One Piece" using seed-based collaborative filtering Core differentiator
FAQ / Policy Answers Return policy, shipping info, payment questions via RAG Deflects support tickets
Product Q&A "Is this in English?" "How many pages?" via catalog lookup Removes purchase friction
Human Escalation "Talk to a human" with context handoff Safety net
Chat Widget Bottom-right FAB, basic chat panel, quick chips The interface
Basic Guardrails PII filter, toxicity filter, price validation Non-negotiable for launch
Core Analytics Session count, intent distribution, latency, error rate Must measure from day one

What's Not in MVP

  • Order tracking
  • Returns and refund flow
  • Personalized greetings
  • Proactive messages
  • Cross-sell and upsell
  • Multi-language support
  • Mobile-specific optimizations
  • A/B testing framework

MVP Architecture

graph LR
    A[Chat Widget] --> B[API Gateway]
    B --> C[Orchestrator]
    C --> D[Intent Classifier<br>Rule-based only]
    C --> E[RAG Pipeline]
    C --> F[Recommendation Engine]
    C --> G[Product Catalog]
    E --> H[LLM - Bedrock]
    F --> H
    H --> I[Basic Guardrails]
    I --> A

MVP uses rule-based intent classification only. That keeps the initial scope small and still covers the limited intent set.


V2 - Full Shopping Assistant

Timeline: About 2 to 3 months after MVP launch.

Goal: Handle the full shopping lifecycle including post-purchase support, and add personalization.

New Features

Feature Description Why in V2
Order Tracking "Where is my order?" with real-time status High-volume support case
Returns and Refunds "I want to return this" with eligibility check Reduces escalation
Promotion Awareness "Any deals on manga?" Drives incremental revenue
ML Intent Classifier Fine-tuned BERT replacing rule-based routing Higher accuracy as intent set grows
Personalized Greetings "Welcome back! Vol 12 just released" Increases engagement
Cart-Aware Responses "The box set is cheaper than your 3 individual volumes" Cross-sell opportunity
Enhanced Product Cards Images, ratings, Prime badge, Add to Cart buttons Richer UX
Feedback Loop Thumbs up/down feeds prompt improvement Continuous quality improvement

V2 Architecture Additions

graph TD
    subgraph "New in V2"
        A[ML Intent Classifier<br>SageMaker]
        B[Order Service Integration]
        C[Returns Service Integration]
        D[Promotions Service Integration]
        E[User Profile Integration]
        F[Feedback Pipeline<br>Kinesis to Model Improvement]
    end

V3 - Proactive and Global

Timeline: About 3 to 4 months after V2.

Goal: Make the chatbot a proactive shopping advisor, support international users, and integrate more deeply with the store experience.

New Features

Feature Description Why in V3
Proactive Recommendations Suggest titles based on browsing without being asked Increases discovery
Cross-sell and Upsell "Readers who bought this also loved..." at checkout Lifts average order value
Multi-language Support Japanese, Spanish, French, German Serves international buyers
Voice Input Speech-to-text for mobile users Accessibility and convenience
Reading Order Guides "What order should I read Fate/Stay Night?" High-value content
New Release Alerts Pre-order suggestions for upcoming volumes Drives pre-order revenue
Wishlist Integration "Add this to my wishlist" Retains intent for future conversion
A/B Testing Framework Test prompts, recommendations, and UX flows Data-driven optimization
Advanced Analytics Funnel analysis, cohort tracking, LTV impact Long-term measurement

V3 Architecture Additions

graph TD
    subgraph "New in V3"
        A[Proactive Trigger Engine<br>Event-driven suggestions]
        B[Multi-language LLM<br>Prompt per locale]
        C[Voice-to-Text<br>Amazon Transcribe]
        D[A/B Testing Service<br>Feature flags + experiments]
        E[Advanced Analytics<br>Cohort analysis in Redshift]
    end

Evolution Summary

graph LR
    V1[MVP<br>Discovery + FAQ<br>+ Recommendations<br>+ Escalation]
    V2[V2<br>+ Orders + Returns<br>+ Personalization<br>+ ML Classifier]
    V3[V3<br>+ Proactive AI<br>+ Multi-language<br>+ Voice + A/B Testing]

    V1 -->|+3 months| V2 -->|+4 months| V3
Dimension MVP V2 V3
Intents supported 5 10 15+
Data integrations 3 7 10+
Languages English English English + Japanese + 3 more
Team size 10-12 15-18 18-22
Expected conversion lift +2% +5% +8%
Support deflection 30% 50% 70%