Feature Prioritization

1 Learning Objectives

Key terms

Out of scope

• Product marketing strategy
• Detailed financial modeling
• User acquisition strategy
• Long-term product roadmaps

2 MVP vs Full Production

Same feature, two completely different scopes. The MVP exists to answer: "is this worth building at all?" The production system exists to answer: "can it run reliably for everyone?"

3 Impact vs Effort Matrix

Two axes, four quadrants, one rule per quadrant. The decision happens before any code does.

Honest effort estimation

• Engineering days — not calendar days. A feature that needs 2 specialists you don't have is infinite effort.
• Hidden costs — data labeling, eval setup, ops burden, ongoing maintenance.
• If unsure, double it. AI features routinely take 2–3× the naïve estimate because of data wrangling.

4 Worked Example — Recommendation Engine

Here's a real feature list students often build for a recsys MVP. Use the interactive matrix tool to drag these around yourself.

• Personalized recommendations (high impact, high effort) — the core thing. Phase it: start with content-based, add collaborative filtering later.
• Top-N popular fallback (medium impact, low effort) — quick win, also unblocks cold-start.
• User feedback loop (high impact, medium effort) — needed to evaluate at all. Don't ship without it.
• Real-time updates (low impact for MVP, high effort) — fight the urge. Daily batch is fine for proof-of-value.
• Recommendation explanations ("why this?") (medium impact, medium effort) — defer unless your hypothesis depends on trust.
• Slick UI for browsing recs (low impact, medium effort) — optional polish.

5 3-Hour Session Schedule

6 Hands-On Labs

Part A — Feature List

Create week31_ai_system_playbook/feature_list.md. For your assigned system, brainstorm freely — quantity first, judgment later. Categorize:

• Core features — the system can't be called "the system" without these
• Advanced features — improve the experience or unlock new use cases
• Nice-to-have features — polish, edge cases, "wouldn't it be cool if…"

# Recommendation Engine — Feature List

## Core
- Top-N item recommendation API
- User profile lookup
- Item embeddings store
- Click logging

## Advanced
- Personalization based on viewing history
- Cold-start handling for new users
- Explanation ("why this?")

## Nice-to-have
- Real-time updates
- Browsable rec UI
- A/B framework

Part B — Prioritization Matrix

Create week31_ai_system_playbook/feature_prioritization_framework.md. Categorize every feature from Part A into one of the four quadrants. Justify every "high effort" claim with concrete reasons (data labeling? specialist needed? new infra?).

Part C — MVP Definition

Add a section titled "Why these features define the MVP". Answer:

• Which features must exist for the system to demonstrate value at all?
• Which features get postponed, and what triggers their inclusion later?
• What trade-offs did you accept?

7 Anti-GenAI Requirement

8 Pre-class Reading

• ProductPlan — Minimum Viable Product. Validate ideas quickly before building complex systems. productplan.com/glossary/minimum-viable-product
• Productboard — Prioritization Frameworks. Several frameworks for ranking features by value, impact, effort. productboard.com/blog/prioritization-frameworks
• Atlassian — Impact vs Effort Matrix. The classic 2×2 explained simply. atlassian.com/work-management/prioritization

9 Closing Checklist

Before you leave tonight, every team should have:

• week31_ai_system_playbook/feature_list.md
• week31_ai_system_playbook/feature_prioritization_framework.md
• A clear MVP scope with ≤ 5 features
• The "Why these features are critical for the MVP" section filled in

These artifacts plug straight into the capstone product brief once topics are revealed.