Skip to content

metric-plan-003.md

Latest commit

 

History

History
37 lines (30 loc) · 2.58 KB

metric-plan-003.md

File metadata and controls

37 lines (30 loc) · 2.58 KB

Metric Plan: AI Personal Finance Advisor (Issue 003)

North Star Metric

Daily Log Rate: The percentage of active users who successfully log an expense (or a 0 spend) within 12 hours of receiving the 8:00 PM nudge. (Ties directly to the hypothesis: Does a WhatsApp nudge reduce the friction of budgeting and maintain daily engagement?)

Supporting Metrics

  1. Day-7 and Day-14 Retention: Percentage of users still logging expenses 1 and 2 weeks after their first message.
  2. Weekly Goal Success Rate: Percentage of users whose Sunday Summary shows totalSpent <= weekly_goal.
  3. Ghosting Rate: Percentage of nudges sent that receive zero reply before the next nudge cycle.

Event Tracking Plan

Event Name Trigger Condition Properties to Capture
user_onboarded First message received causing new DB row. phone_number_hash, timestamp
daily_nudge_sent Cron job successfully fires API dispatch to a user. user_id, timestamp
expense_logged User replies with a valid numerical amount > 0. user_id, amount, timestamp
zero_spend_logged User explicitly replies with "0". user_id, timestamp
invalid_input_received User replies but regex fails to find a number. user_id, raw_text
weekly_summary_sent Sunday cron job successfully dispatches summary. user_id, total_spent, goal, budget_status

Funnel Definition

The Engagement Funnel

  1. User receives 8 PM Nudge (daily_nudge_sent).
  2. User replies with text (Drop-off point: Ghosting).
  3. Text is successfully parsed as an expense or 0 (expense_logged / zero_spend_logged). (Drop-off point: invalid_input_received).

Success Thresholds

  • Target North Star (Daily Log Rate): > 60%
  • Target Retention (Day-14): > 40%
  • Target Goal Success Rate: > 30% of users staying under budget.
  • Alert Threshold: If invalid_input_received occurs for > 15% of total messages, the regex parsing is failing real-world unstructured text and an LLM parser needs to be swapped in.

Implementation Notes

  • Tool: Since backend API routes are handling everything, server-side analytics (e.g., PostHog Node.js SDK) should be used inside the webhook and cron functions.
  • Database Proxying: Alternatively, for an MVP of 20 users, these metrics can be manually derived directly from the Supabase users and logs tables using SQL aggregation queries, avoiding the need to install a heavy analytics SDK. The logs.logged_at timestamp compared against users.created_at provides exact retention curves without code overhead.