Churn Risk Scoring

This drill builds a quantitative churn risk model that scores every active user daily. The model ingests behavioral signals from PostHog, weights them based on historical churn correlation, and outputs a 0-100 risk score per user. Scored users are synced to Attio and PostHog cohorts so downstream systems (intervention emails, in-app messages, sales alerts) can act on them.

This is the detection layer. It answers "who is at risk?" The churn-prevention drill answers "what do we do about it?"

Prerequisites

• PostHog tracking active with at least 30 days of usage data for churned and retained users
• n8n instance for daily scoring workflow
• Attio configured with user/company records
• At least 20 churned users to analyze patterns against (fewer and the model is unreliable)

Steps

1. Identify churn signals from historical data

Using the posthog-cohorts fundamental, create two cohorts: "Churned Last 90 Days" and "Retained Last 90 Days." Compare their behavior in the 30 days before churn vs. the same 30-day window for retained users. Query via PostHog API or HogQL:

SELECT
  person_id,
  countIf(event = 'pageview' AND timestamp > now() - interval 14 day) AS sessions_last_14d,
  countIf(event = 'core_action' AND timestamp > now() - interval 14 day) AS core_actions_last_14d,
  dateDiff('day', max(timestamp), now()) AS days_since_last_activity,
  countIf(event = 'support_ticket_created' AND timestamp > now() - interval 14 day) AS support_tickets_14d,
  countIf(event = 'team_member_removed' AND timestamp > now() - interval 30 day) AS team_removals_30d,
  countIf(event = 'billing_page_viewed' AND timestamp > now() - interval 7 day) AS billing_views_7d
FROM events
WHERE person_id IN (SELECT person_id FROM cohort WHERE name = '{cohort_name}')
GROUP BY person_id

Run this for both cohorts. Calculate the median and standard deviation of each signal for churned vs. retained users.

2. Define the scoring model

Assign point values to each signal based on how strongly it correlates with churn. Start with these defaults and refine based on your data:

| Signal | Condition | Points | |--------|-----------|--------| | Usage decline | Sessions dropped >50% vs. user's 30-day average | +25 | | Core action absence | Zero core actions in last 7 days (was previously active) | +20 | | Login gap | No login for 7+ days when user averaged 3+/week | +15 | | Support spike | 3+ support tickets in 7 days | +10 | | Team shrinkage | Removed 1+ team members in 30 days | +10 | | Billing page visit | Visited pricing/cancellation page in last 7 days | +15 | | Feature abandonment | Stopped using a feature they used 5+ times previously | +10 | | Export spike | Exported data 3+ times in 7 days (data portability signal) | +10 |

Total possible: well above 100. Cap the score at 100. A user with score 0 shows no churn signals.

3. Build the daily scoring workflow

Using the n8n-triggers fundamental, create a workflow triggered by daily cron (06:00 UTC):

1. Query PostHog for all active users (at least 1 session in the last 60 days) with their signal metrics from step 1
2. For each user, calculate the composite risk score by summing applicable signal points
3. Classify into tiers:
   • Low risk (0-25): Normal behavior. No action needed.
   • Watch (26-45): One or two weak signals. Monitor but do not intervene yet.
   • Medium risk (46-65): Multiple signals firing. Queue for automated intervention.
   • High risk (66-85): Strong churn pattern. Trigger immediate automated intervention.
   • Critical (86-100): Imminent churn. Route to human for personal outreach.
4. Using posthog-custom-events, emit a churn_risk_scored event per user with properties: risk_score, risk_tier, top_signals (array of the signals that fired), score_date
5. Using attio-custom-attributes, update each user's Attio record with churn_risk_score, churn_risk_tier, and churn_risk_signals
6. Using attio-contacts, create a task for the account owner when a user moves to Critical tier

4. Create PostHog cohorts per risk tier

Using the posthog-cohorts fundamental, create dynamic cohorts that filter on the churn_risk_tier property:

• "At-Risk: Medium" — users with churn_risk_tier = medium
• "At-Risk: High" — users with churn_risk_tier = high
• "At-Risk: Critical" — users with churn_risk_tier = critical

These cohorts feed the churn-prevention drill's intervention triggers. Intercom and Loops can target these cohorts directly.

5. Validate and calibrate the model

Using posthog-funnels, measure model accuracy monthly:

• Precision: Of users scored high/critical, what percentage actually churned within 30 days?
• Recall: Of users who churned, what percentage were scored high/critical before churning?
• False positive rate: Users scored high/critical who did not churn. High false positives waste intervention resources and annoy healthy users.

Target: precision > 60%, recall > 70%. If precision is low, raise thresholds. If recall is low, add more signals or lower thresholds. Log calibration results as a PostHog custom event (churn_model_calibration) with properties: precision, recall, false_positive_rate, calibration_date.

6. Track score transitions

Monitor how users move between tiers over time. The most important metric is "Medium/High risk users who returned to Low risk within 14 days" — this is your intervention save rate signal. Build a PostHog funnel:

churn_risk_scored (tier=high) -> churn_risk_scored (tier=low) within 14 days

This funnel's conversion rate is one of the primary KPIs for the at-risk-intervention play.

Output

• A daily n8n workflow that scores every active user and syncs scores to PostHog + Attio
• Dynamic PostHog cohorts for each risk tier
• Attio records updated with risk scores for team visibility
• A model calibration process that runs monthly
• Score transition tracking that measures intervention effectiveness