Skip to main content
Goal: Segment users into cohorts by signup period and calculate retention metrics to identify patterns in user engagement and product stickiness. This skill runs using CORE memory only. No integrations required. Trigger: Run on demand when the user requests cohort analysis or retention metrics.

​
Setup

Search memory for:
  • “What date range should cohort analysis cover?”
  • “What retention metrics does the user care about?”
  • “How should users be grouped (daily, weekly, monthly cohorts)?”
If not found, ask once:
“To analyze cohorts, I need to know: (1) What time period should I cover? (2) Should I group users by day, week, or month of signup? (3) Which retention milestones matter most (7-day, 30-day, 90-day)?”
Store the response in memory. Do not ask again in future runs.

​
Step 1: Define Cohort Boundaries

Establish the date range and cohort grouping period from memory or setup response.
  • Cohort grouping: Day, week, or month (e.g., “all users who signed up in the week of Jan 1-7”)
  • Analysis window: Start date to end date (e.g., last 180 days)
  • Retention milestones: Days 7, 30, 60, 90 (or custom from memory)
If memory contains previous cohort preferences → use silently. If not → use defaults (monthly cohorts, 180-day window, 7/30/90-day retention).

​
Step 2: Extract Cohort Data

Pull user signup dates and activation status from available data sources. For each user, capture:
  • Signup date (to assign to cohort)
  • First action date (to confirm activation)
  • Last active date (to calculate churn)
  • Total events or sessions (engagement proxy)
If data source is unavailable → note “No user activity data available” and skip to output.

​
Step 3: Assign Users to Cohorts

Group users by signup period (day/week/month as defined in Step 1). For each cohort:
  • Count total users
  • Count activated users (first action within 7 days of signup)
  • Record activation rate as percentage
Output format: [Cohort Period]: [X users], [Y% activation]

​
Step 4: Calculate Retention Curves

For each cohort, measure how many users remain active at Day 7, Day 30, Day 60, and Day 90 post-signup.
  • Retention = (users still active on day N) / (total cohort size) × 100%
  • If a user has no activity on a given day, they are counted as churned
  • Round percentages to whole numbers
Flag cohorts with <50% 7-day retention as high-churn. Compare retention curves across cohorts.
  • Improving trend: Later cohorts show higher retention at comparable milestones
  • Declining trend: Earlier cohorts show higher retention (product regressing)
  • Plateau: Retention stable across all cohorts (consistent product health)
If retention improves by >5% month-over-month → flag as positive. If drops by >5% → flag as concerning.

​
Step 6: Calculate Cohort Lifetime Value Proxy

For each cohort, sum total events/sessions per user and calculate average.
  • Cohorts with higher average engagement tend to have stronger long-term value
  • Note early warning signs (very low engagement by Day 30)

​
Step 7: Compile Output

Structure the analysis into a cohort table and trend summary.

​
Output Format

Cohort Analysis — [Start Date] to [End Date] Cohort Retention Table
CohortUsersActivationDay 7Day 30Day 60Day 90
[Period 1][X][Y%][Y%][Y%][Y%][Y%]
[Period 2][X][Y%][Y%][Y%][Y%][Y%]
Trends
  • 📊 Activation rates: [Improving / Declining / Stable]
  • 📊 7-day retention: [Average X%, range Y–Z%]
  • 📊 30-day retention: [Average X%, range Y–Z%]
  • 🚨 High-churn cohorts: [List cohorts <50% 7-day retention, or “None”]
Key Insights
  1. [Most recent cohort health]
  2. [Strongest cohort and why]
  3. [Weakest cohort or trend concern]

​
Edge Cases

  • Insufficient data: If fewer than 3 cohorts or <100 total users exist, note “Cohort sample too small for reliable trends” and show available data only.
  • No retention data: If users have no activity after signup (all churn on Day 0), show 0% retention and flag data quality.
  • Future signups: Exclude any users with signup dates in the future; silently filter.
  • Custom time zones: If user has specified a time zone in memory, adjust cohort boundaries accordingly (e.g., “week starts Monday at midnight PST”).
  • Overlapping date ranges: If user asks for cohorts that overlap the current date, only include complete cohorts; note how many users are in the partial cohort but exclude from calculations.
  • No memory preferences: Default to monthly cohorts, last 180 days, 7/30/90-day milestones; continue without asking again.