Churn Score Reflexivity Index (CSRI)

GMI-TEL-010TelecommunicationsInfrastructure-Dependent

Definition

Correlation between churn-score-driven interventions and the outcomes those scores predict, measuring how much operational response to the score distorts the score's own validity.

The Measurement Gap

Measurement Gap and Decision

What CSRI Measures: CSRI quantifies the degree to which operational responses to churn scores causally influence the outcomes those scores predicted. It compares churn rates of low-scored subscribers who receive degraded service against a baseline of low-scored subscribers who receive standard service.

Decision Supported:

Is the churn model's apparent accuracy partly an artifact of its own influence on operations, and if so, how large is the reflexive effect? Without CSRI, a model can appear accurate because it causes the outcome it predicts: subscribers labeled "low risk" receive worse service, become dissatisfied, and churn. Confirming the model's prediction retroactively.

Why This Is Unmeasured:

No standard operational metric tests whether a predictive model is contaminating its own outcomes. Churn models report accuracy as a fixed property of the model, without examining whether operational responses to the scores are changing subscriber behavior in ways that inflate apparent accuracy. When a low-risk subscriber receives longer hold times, fewer proactive touches, and slower issue resolution because the system deprioritizes them, their subsequent churn validates the score, creating a feedback loop that entrenches the model's biases rather than revealing them. This reflexivity is politically uncomfortable because it implies the organization's AI-driven operations are creating customer harm (Journal of Marketing Research, "Algorithmic Self-Fulfilling Prophecies in Customer Management," 2024; Telecommunications Policy, "Feedback Loops in Predictive Customer Analytics," 2023).

Formula

Plain Language: Identify subscribers scored as low churn risk. Measure whether they receive systematically different service quality compared to a baseline. Then measure whether their actual churn rate correlates with the degree of service degradation they experienced.

Notation:

CSRI = ρ(Service_degradation_index, Churn_outcome) | low_risk_cohort

Variable Definitions:

  • ρ = point-biserial correlation coefficient (correlation between a continuous variable and a binary variable)
  • Service_degradation_index = composite measure of service quality received by each subscriber relative to the population average, incorporating hold time percentile, resolution time percentile, and proactive contact frequency (dimensionless composite score)
  • Churn_outcome = binary variable: 1 if the subscriber churned during the measurement period, 0 if retained (dimensionless)
  • low_risk_cohort = subscribers in the lowest churn risk score band as defined by the churn model (the population subset for measurement)

Units:

Correlation coefficient, ranging from -1 to +1. CSRI near 0 = no reflexive effect (low-risk subscribers churn at the same rate regardless of service quality received). CSRI significantly positive = subscribers who received worse service churned more, suggesting the score influenced the outcome through operational response.

Sign Convention:

Lower absolute value is better. CSRI near zero means the model's predictions are not contaminated by operational feedback. Positive CSRI indicates the model is partly creating what it predicts.

Edge Cases:

  • Service degradation may be caused by confounders (geography, plan type, time of contact). Control for observable confounders using matching or regression adjustment.
  • A small positive CSRI does not prove the model is wrong: the low-risk cohort may be genuinely low-risk AND receive worse service without the service difference causing the churn.
  • Minimum sample: 500 or more low-risk subscribers with complete service quality data per measurement period.

Proposed formula, pending validation pilots.

At a Glance

Unit
Correlation (-1 to 1)
Trend Direction
Lower; approaching 0
Reporting Frequency
Quarterly
Category
AI-Driven Customer Value and Retention Integrity
← Marginal Commitment Risk Index (MCRI)Retention Offer Cannibalization Rate (ROCR) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment