Retention

Core Metrics

Cohort Analysis

Track by signup week, not calendar week:

• Horizontal axis: weeks since signup (0, 1, 2, 3...)
• Vertical axis: signup cohort (Jan W1, Jan W2...)
• Cell value: % of cohort still active

Identify:

• Which cohorts retain better (product changes, marketing source)
• At which week users drop off (week 2 cliff = aha moment too late)
• Seasonal patterns (holiday signups retain worse)

Churn Signals

Early warning indicators (flag before churn):

• Login frequency drops 50%+ from baseline
• Core feature usage stops
• Support tickets spike then go silent
• Billing page visits without upgrade
• Team member removals
• Data export requests

Engagement Loops

Retention requires habit formation:

Design for variable rewards - predictable = boring.

Lifecycle Stages

Reactivation Campaigns

Timing matters:

• 7 days inactive: Soft nudge ("We miss you")
• 14 days inactive: Value reminder + what's new
• 30 days inactive: Incentive offer (discount, extended trial)
• 90 days inactive: Last chance + feedback ask

Message formula:

[Acknowledge absence] + [New value added] + [Easy re-entry CTA]
"Your dashboard is waiting. We added [feature]. One click to resume →"

Feature Stickiness

Measure which features predict retention:

• Usage correlation: Users of feature X retain 2x better
• Time to feature: Users who reach feature X in day 1 retain 3x
• Feature breadth: Users of 3+ features retain 5x vs 1 feature

Double down on sticky features in onboarding.

Churn Prevention

When churn signal detected:

1. Immediate: In-app message acknowledging drop ("Need help?")
2. Day 3: Email from founder (personal, not marketing)
3. Day 7: Offer call or live support
4. Before renewal: Proactive outreach with usage summary

Cancel flow optimization:

• Ask reason (required, 4-5 options)
• Offer pause instead of cancel
• Show what they'll lose (data, history, price lock)
• Easy return policy ("reactivate anytime, data saved 90 days")

Retention Benchmarks by Model

Common Mistakes

• Measuring retention from signup, not activation
• Treating all churned users the same (voluntary vs involuntary)
• Reactivation emails without new value proposition
• Ignoring payment failures as churn (30-40% of churn is involuntary)
• No segmentation in cohort analysis (power users mask problems)