🎯 AI Decision Scenarios

Think like an analytics leader. Make the call. Defend your reasoning.

💰 The CEO's AI Initiative

Your CEO just announced a $2 million AI initiative at the all-hands meeting. Now the executive team needs to decide which projects to fund. You're the analytics lead—your recommendation carries weight.

📋 The Situation

Company: MidWest Manufacturing Co. — $450M annual revenue, 2,800 employees
Budget: $2 million for AI initiatives (Year 1)
Timeline: Board expects measurable results within 18 months
Data Maturity: Mixed — good ERP data, fragmented customer data, limited data science team (3 people)
Current Pain Points: Rising warranty costs, inventory inefficiencies, sales team spending too much time on low-value leads

👆 Your Task: Click on the use cases in order of priority (1 = highest). You have budget for approximately 1-2 projects. Then explain your reasoning below.

🔧 Predictive Maintenance

Use sensor data from manufacturing equipment to predict failures before they happen. Reduce unplanned downtime and optimize maintenance schedules.

The Pitch: "We're losing $3.2M annually to unplanned downtime. Industry benchmarks suggest 25-30% reduction is achievable."

Est. Cost
$850K
Time to Value
12-18 mo
Data Readiness
Medium
Risk Level
Medium

🚪 Customer Churn Prediction

Identify which customers are likely to leave so the retention team can intervene proactively. Integrate with CRM for automated alerts.

The Pitch: "Customer acquisition costs 5x more than retention. We lost 127 accounts last year worth $8.4M in annual revenue."

Est. Cost
$400K
Time to Value
6-9 mo
Data Readiness
Low
Risk Level
Medium

🤖 GenAI Sales Assistant

Deploy a GPT-powered assistant that helps sales reps generate proposals, answer technical questions, and prioritize leads using conversation analysis.

The Pitch: "Our competitors are doing this. Sales reps spend 35% of their time on admin tasks. This could free up 10+ hours per rep per week."

Est. Cost
$600K
Time to Value
3-6 mo
Data Readiness
High
Risk Level
High

📊 Demand Forecasting

Improve inventory planning by predicting demand at the SKU level. Reduce both stockouts and excess inventory carrying costs.

The Pitch: "We're carrying $12M in excess inventory while simultaneously losing $2.1M to stockouts. Better forecasting could recover $3-4M annually."

Est. Cost
$550K
Time to Value
9-12 mo
Data Readiness
High
Risk Level
Low

📝 Your Recommendation

You've selected: None yet

Total estimated cost: $0 of $2M budget

📉 The Model Is Drifting

You're the analytics manager at a regional bank. Six months ago, your team deployed a customer churn prediction model that was performing beautifully. This morning, you got an alert that stopped you cold.

⚠️ Model Performance Dashboard

🚨 DRIFT DETECTED
62%
Current Accuracy
↓ 24% from baseline
0.58
AUC Score
↓ from 0.84 at launch
47%
False Positive Rate
↑ 31% from baseline
2,847
Customers Flagged This Week
↑ 340% vs. average

📅 Recent Events Timeline

6 months ago
Model Deployed — 86% accuracy, AUC 0.84, stakeholders thrilled
3 months ago
New Competitor Enters Market — FinTech startup offering 4.5% savings rates
6 weeks ago
Marketing Launches Promo — "Stay & Save" campaign with retention bonuses
2 weeks ago
Fed Raises Rates — Interest rate environment shifts significantly
This morning
Drift Alert Triggered — Model accuracy dropped below 65% threshold

🎯 The Stakes

The retention team relies on your model's predictions to prioritize outreach. They have capacity to call ~200 customers per week. If the model is flagging the wrong customers:

  • High-risk customers aren't getting called (they'll churn)
  • Low-risk customers are getting unnecessary calls (wasted effort, annoyed customers)
  • The retention team is losing trust in the model

👆 Your Task: Select your immediate response, then explain your reasoning. What do you do RIGHT NOW?

A Immediately Retrain the Model

Pull the latest 6 months of data and retrain the model from scratch. Push to production by end of week. The model needs to learn the new patterns.

B Turn Off the Model, Revert to Rules

Disable automated predictions. Have the retention team use simple business rules (e.g., "call anyone who hasn't logged in for 30 days") until you can investigate.

C Investigate First, Then Decide

Before changing anything, analyze what's causing the drift. Is it data quality? Concept drift? Feature distribution shift? Different problems need different solutions.

D Adjust the Threshold

The model might still be ranking customers correctly—just with different probability values. Recalibrate the decision threshold to reduce false positives.

E Call an Emergency Meeting

Get stakeholders together—retention team, marketing, data team—to understand the full picture before making technical decisions. This is a business problem, not just a model problem.