Business Problems & Data Science Solutions
Somsak Chanaim
International College of Digital Innovation, CMU
June 6, 2025
Business Problems
Business Problem refers to a situation or obstacle that prevents a business from achieving its goals.
Examples include:
declining profits
falling sales
losing customers
rising costs
inefficient operations
If left unaddressed, these issues can affect a company’s competitiveness or long-term survival.
Why Do Businesses Need Data Science?
Businesses have large volumes of data but lack deep analytical insights
Enables data-driven decision-making
Helps reduce costs, improve efficiency, and forecast future trends
Data Driven Business Problem Solving
Data-Driven Business Problem Solving
This approach involves using available data to analyze root causes and make strategic decisions based on evidence, rather than just “intuition” or “experience” alone.
Steps in Solving Business Problems with Data
Real-World Data Use Case
Problem: Customers Are Not Returning
Analyze Customer Lifetime Value (CLV)
Perform segmentation and send targeted promotions
Build a recommendation system to encourage repeat purchases
Examples of Business Problem Types
Marketing: Low customer retention, low conversion rates, low brand awareness
Operations: Overstocked inventory, demand fluctuation, slow processes
Finance: Poor cost control, inaccurate profit forecasts, cash flow issues
Human Resources: High employee turnover, low engagement, lack of tech talent
Government / Public Sector
Unequal access to public services: Slow complaint handling, staff shortages
Misallocated budget: Inability to measure policy impact
Lack of data for policy decisions: e.g., outdated data on low-income citizens
Corruption and low transparency: Weak internal auditing systems
Urban issues: Traffic, PM2.5, public health — require data-driven spatial and behavioral management
Personal finance: Overspending, lack of savings, unsure about investments
Career development: Unaware of market-relevant skills, lack of career planning
Health & behavior: Lack of personal health data, risky habits
Lifelong learning: Don’t know how to upskill or choose the right courses
Life decisions: Choosing a house/car, planning for family life
Translating Business Problems → Data Problems
| Customer churn |
Predict churn |
Classification |
| Poor product sales |
Forecast sales |
Time Series Forecasting |
| Ineffective promotions |
Segment customers |
Clustering |
| Fraud detection |
Identify abnormal behavior |
Anomaly Detection |
Key Business Metrics to Know
1. Customer Lifetime Value (CLV)
The total value a customer is expected to generate for the business throughout their entire relationship.
Formula:
\[
\text{CLV} = \text{Average Purchase Value} \times \text{Purchase Frequency} \times \text{Customer Lifespan}
\]
Example:
Average purchase = 500 THB/month
Buys every month → 12 times/year
Remains a customer for 3 years
\[
\text{CLV} = 500 \times 12 \times 3 = 18{,}000 \text{ THB}
\]
2. Customer Retention Rate
The percentage of existing customers who continue to use the service over a given period.
Formula:
\[
\text{Retention Rate} = \left( \frac{E - N}{S} \right) \times 100
\]
\(E\): Number of customers at the end of the period
\(N\): Number of new customers acquired
\(S\): Number of existing customers at the beginning
Example:
Start of year: 1,000 customers
End of year: 1,200 customers (400 new)
\[
\text{Retention Rate} = \left( \frac{1{,}200 - 400}{1{,}000} \right) \times 100 = 80\%
\]
3. Return on Investment (ROI)
Measures the return on an investment to evaluate its efficiency or profitability.
Formula:
\[
\text{ROI} = \left( \frac{\text{Net Profit}}{\text{Investment Cost}} \right) \times 100
\]
Example:
Investment: 100,000 THB
Net profit: 30,000 THB
\[
\text{ROI} = \left( \frac{30{,}000}{100{,}000} \right) \times 100 = 30\%
\]
4. Inventory Turnover Rate
Measures how quickly inventory is sold and replenished. A higher rate means lower holding time.
Formula:
\[
\text{Inventory Turnover} = \frac{\text{Cost of Goods Sold (COGS)}}{\text{Average Inventory}}
\]
Example:
Annual COGS = 1,000,000 THB
Average inventory value = 250,000 THB
\[
\text{Inventory Turnover} = \frac{1{,}000{,}000}{250{,}000} = 4 \text{ cycles/year}
\]
5. Click-Through Rate (CTR)
Measures the percentage of viewers who clicked on an ad or link.
Formula:
\[
\text{CTR} = \left( \frac{\text{Number of Clicks}}{\text{Number of Impressions}} \right) \times 100
\]
Example:
Ad impressions: 10,000
Clicks: 300
\[
\text{CTR} = \left( \frac{300}{10{,}000} \right) \times 100 = 3\%
\]
Types of Business Data
Behavioral data (e.g., customer actions)
Transaction data (e.g., purchase history)
Real-time logs (e.g., system or clickstream data)
External data (e.g., weather, social media)
Choosing the Right Model for the Problem
| Classification |
Logistic Regression, Random Forest |
To predict categories |
| Regression |
Linear Regression, XGBoost |
To predict numeric values |
| Clustering |
K-Means, DBSCAN |
To find hidden groups |
| Recommendation |
Collaborative Filtering |
To recommend products |
| Time Series |
ARIMA, Prophet |
To forecast future trends |
How to Present the Results?
Create a dashboard for clear and interactive insights
Use data storytelling
Focus on business impact, not just model accuracy
Ethics & Pitfalls
Data Privacy & GDPR (General Data Protection Regulation)
Bias in data can lead to flawed models
Ensure model results are explainable and understandable to stakeholders
![]()
Sales Forecasting
Business Problem: Need to forecast daily sales per branch to optimize inventory management.
Time Series Models: ARIMA, SARIMA
Gradient Boosting: XGBoost, LightGBM
Deep Learning: LSTM (Long Short-Term Memory)
🎬 Netflix: Personalized Recommendation System
Collaborative Filtering: Matrix Factorization (SVD, ALS)
Content-Based Filtering: TF-IDF + Cosine Similarity
Deep Learning: Autoencoders, Neural Collaborative Filtering
☕ Starbucks: Location Analytics for New Store Placement
Geospatial Clustering: K-Means, DBSCAN
Regression Models: Random Forest Regressor
Predictive Modeling: Gradient Boosting, Decision Trees
🚗 GrabCar: Dynamic Pricing Strategy
Problem:
Passengers complain about unexpected fare increases or excessively high prices during peak hours
However, without price adjustments, drivers may reject requests → the challenge is to balance passenger satisfaction vs. driver incentives
Regression Models: Linear, Ridge, Lasso, Gradient Boosting
Reinforcement Learning: Multi-Armed Bandit, Deep Q-Network (DQN)
Demand Forecasting: XGBoost, LSTM (for time-dependent patterns)
Geospatial Analytics: Heatmaps, Clustering
Reduced passenger wait times → Higher satisfaction
Maintained driver motivation → More accepted rides
Improved ride matching → Lower cancellation rates
Increased average driver earnings during peak hours
🛍️ Zara: Inventory Optimization
Clustering: K-Means, Hierarchical Clustering
Classification: Random Forest, Logistic Regression
Forecasting: Time Series + Machine Learning
Chapter Summary
Understanding the business problem is the first step
Match the problem with the right Data Science techniques
Present actionable and practical results
“Data is valuable only when turned into action.”
