INI-005
·SubmetidaChurn prediction with predictive model
Patrícia Souza · CommercialAI analyticsCriticality MediumAutonomy: Suggests actions
71Overall
Automatic evaluation
Business Value78
AI Quality72
Security70
Privacy65
Opinion
Healthy use case, but governance work is needed before production. Classic ML is the right choice — no LLM justified here.
Aieval qualifies and recommends — the final decision is always human.
Gaps
1- Training data retention policy not defined
Risks
2- Bias against under-represented segments
- PII use requires documented legal basis
Recommendations
3- Validate legal basis (legitimate interest) with Privacy
- Bias audit by segment before production
- Document features and model card
Discovery
Answers collected during intake
- Problem
- Identify customers with high cancellation risk.
- Objective
- Anticipate churn 60 days ahead for proactive action.
- Impacted users
- CS team (25 people)
- Vendor / model
- Internal model (XGBoost) · Supervised ML
- Data used
- Transactional history, Product engagement
- Integrations
- Data Warehouse, CRM
- KPIs
- Recall · Lift over baseline · Churn reduction
- Estimated cost
- $60,000
- Risk flags
- Uses PII: YesCustomer-facing: NoProduction access: YesExternal sharing: No