Retention & Engagement in AI

Core thesis

Retention = f(Activation, Engagement, Resurrection). You cannot work on "retention" directly — you pull one of three levers. Each requires different metrics, strategies, and investment. AI products face the same retention math as every prior tech wave, but with three new realities: switching costs have collapsed, capability jumps every ~7 months (forcing bridge rebuilds), and products fail silently (every prompt gets an answer, even when it is wrong).

Retention strategy sequence

Five Parts of a Use Case

• Problem definition: specific job, not "be productive"
• Persona: demographics, skill level, workflow context
• The Why: speed, quality, scale, accessibility, consistency
• Alternatives: manual, human expert, traditional software, nothing
• Natural frequency: daily/weekly/monthly/quarterly/yearly

Initial vs Core use cases

Initial Use Case: low friction, high frequency, easy to activate on — builds the habit. Core Use Case: high frequency, often higher friction — drives stickiness and switching cost. The retention strategy is to engineer the Initial → Core transition. Do not try to serve every use case equally.

Three Activation Moments (design backwards)

• Setup: minimum information or actions to enable the aha moment
• Aha: first experience of core value within a time period
• Habit: repeated behavior around core value — X actions in Y time period

AI-native activation challenges

• Probabilistic output: same input, different quality each time
• Context-dependent performance: cold start problem
• Expectation-reality gap: AGI hype vs real limits
• Value comprehension challenge: delayed value, quality variance, personalization lag
• Habit formation paradox: prompting cognitive load, use case discovery friction

Engagement States

Minimum three segments: Casual, Core, Power. Every transition between states must create measurable business value. Power users create disproportionate value: 30x UGC pages, 40x viral touchpoints, 8x MRR, dramatically higher switching costs.

Four-Way Engagement Impact

• Retention multiplier: 2-3x longer retention curves
• Acquisition loop acceleration: UGC, viral, referrals
• Monetization expansion: usage, transaction, tier, ad inventory
• Defensibility creation: integration, data, habit

Quiet quitting detection

Resurrection reality check

Three flawed assumptions: pool size (real reachable pool is ~10-15% of "inactive"), response rate (2-5% for dormant, not 20-30%), and "nothing to lose" (aggressive campaigns damage deliverability for active users). Prevention beats resurrection by 5-10x ROI. Resurrection priority: at-risk prevention → involuntary recovery (payment, technical) → capability-match resurrection (model upgrade as "Everything Changed" campaign) → generic improvements (last resort).