BAN3200 Module 6 — PRD

Course: BAN3200 Machine Learning and Predictive Analytics Module: 6 — Predictive Analytics Capabilities Degree level: BBA (Undergrad) AST type: AST-5 — Data Interpretation & Advisory Status: Draft — pending content gap remediation before live deployment

CLOs assessed

CLO	Statement	Coverage
CLO-3	Analyze model performance using appropriate metrics to evaluate predictive accuracy and bias	Primary
CLO-5	Interpret predictive modeling results to generate actionable business recommendations for stakeholders	Primary
CLO-6	Evaluate the ethical implications, bias potential, and fairness requirements for responsible model deployment and governance	Primary

Scenario

Company: Costco (C0039) — membership warehouse retail, US + UK operations Context: Junior analyst leading pre-deployment review of an Executive Member churn prediction model (logistic regression) before Q2 retention campaign launch.

Key seeded data:

Test cohort: 8,000 Executive Members, 1,200 at-risk (15%)
Overall: accuracy 85%, recall 50% at threshold 0.55
Disaggregated: <2yr tenure = 25% recall (fairness concern); ≥2yr = 75% recall, F1 = 0.60
Budget: 1,200 interventions → escalated to 900
SHAP top features: warehouse_visits (0.34), executive_tier_age_months (0.28)

Governance anchors: NIST AI RMF Measure 2.5 (US); UK GDPR Article 22 (UK)

Weighting

Area	Weight
Exercises (knowledge MCQ + decision tracing FIBs + AI audit)	55%
Performance task (pre-deployment report)	45%

Assessment variants — locked

	Variant 0	Variant 1
Company	Costco (C0039)	Amazon (C0004)
Membership tiers	Executive ($130/yr) vs Gold Star ($65/yr)	Prime ($139/yr) vs Standard (free)
Churn definition	Executive downgrade or non-renewal	Prime cancellation
Accuracy	85%	82%
Recall at threshold 0.55	50%	40%
<2yr tenure recall	25%	20%
≥2yr tenure recall/F1	75% / 0.60	60% / 0.48
FIB — missed members	600	720
FIB — false positive rate	50%	60%
Escalation delta	120	96
Metrics question order	mq1, mq2, mq3	mq2, mq3, mq1
Governance context	NIST AI RMF (US) + UK GDPR Art. 22 (UK)	same

What stays identical across variants: step structure, rubric, MCQ correct letter answers, SHAP feature ordering, data quality issue types, AI audit errors, governance frameworks.

Confirmed by: Fadl Al-Tarzi · 2026-05-24

Known content gaps (blocks live deployment)

CLO-1: Hands-on dataset preparation not supported by current content
CLO-2: ML model application — tools named but not demonstrated
CLO-4: End-to-end workflow design — content stays at Apply level

Assessment tests what content WILL support once gaps are remediated. Not to go live until confirmed by Head of LXD.

Industry frameworks referenced

EU AI Act (2024) — bias monitoring and deployment governance (EU context)
NIST AI RMF 1.0 (2023) — Measure 2.5 bias monitoring (US)
UK GDPR Article 22 (UK) — automated decision-making human oversight
CRISP-ML(Q) (2021) — Evaluation → Deployment gate framing
SHAP (industry-standard interpretability tool)