Learning Objectives
By the end of this lesson, learners should be able to:
- Define predictive analytics and prescriptive analytics.
- Distinguish descriptive, diagnostic, predictive and prescriptive analytics.
- Explain how predictive analytics supports strategic planning.
- Identify common predictive modeling applications.
- Explain forecasting and probability-based decision-making.
- Interpret predictive model outputs.
- Explain the role of optimization in prescriptive analytics.
- Apply scenario analysis to strategic decisions.
- Evaluate model assumptions and limitations.
- Explain uncertainty and confidence in analytical predictions.
- Connect predictive and prescriptive analytics to business strategy.
- Evaluate ethical and governance considerations when using predictive models.
1. Introduction to Predictive Analytics
Organizations increasingly need to understand not only what has happened, but what may happen next.
Questions may include:
- What will demand be next quarter?
- Which customers may leave?
- Which transactions may be fraudulent?
- Which products are likely to sell?
- Which equipment may fail?
- What revenue might the company generate?
Predictive analytics uses historical and current data, statistical techniques and machine learning to estimate future or unknown outcomes.
2. Four Types of Business Analytics
Business analytics can be viewed as four broad categories:
Descriptive Analytics
What happened?
Example:
Sales declined by 8%.
Diagnostic Analytics
Why did it happen?
Example:
Sales declined primarily because of reduced demand in two major markets.
Predictive Analytics
What may happen next?
Example:
Sales are forecast to decline another 3% next quarter.
Prescriptive Analytics
What should we do?
Example:
The organization should adjust pricing and marketing investment in selected markets based on expected outcomes.
3. Predictive Analytics Process
A typical process is:
Business Problem
↓
Data Collection
↓
Data Preparation
↓
Exploratory Analysis
↓
Model Development
↓
Model Evaluation
↓
Prediction
↓
Business Interpretation
↓
Decision
↓
Monitoring
The process should begin with a clearly defined business objective.
4. Forecasting
Forecasting estimates future values using historical and other relevant information.
Applications include:
- Sales forecasting.
- Demand forecasting.
- Revenue forecasting.
- Cash-flow forecasting.
- Workforce forecasting.
- Inventory forecasting.
Forecasts are not guarantees.
They are estimates based on assumptions and available evidence.
5. Time-Series Forecasting
Time-series data consists of observations collected over time.
Examples:
- Daily sales.
- Monthly revenue.
- Weekly customer traffic.
- Quarterly profit.
Time-series analysis may identify:
- Trend.
- Seasonality.
- Cycles.
- Irregular fluctuations.
6. Trend
A trend represents a long-term direction in a dataset.
For example, monthly sales may generally increase over several years.
Trend analysis can help organizations identify whether growth or decline is occurring.
7. Seasonality
Seasonality refers to predictable patterns that repeat at regular intervals.
Examples:
- Increased retail sales during holiday periods.
- Higher travel demand during certain seasons.
- Increased energy consumption during particular months.
Ignoring seasonality can result in poor forecasts.
8. Regression for Prediction
Regression models can be used to estimate relationships between variables.
For example, a company may investigate whether sales are associated with:
- Advertising expenditure.
- Price.
- Customer traffic.
- Product availability.
A simplified linear model can be represented as:
Y = a + bX
where:
- Y = predicted outcome.
- X = explanatory variable.
- a = intercept.
- b = estimated relationship.
Multiple regression can incorporate several explanatory variables.
9. Classification
Classification models predict categories.
Examples:
- Customer will churn / will not churn.
- Transaction is suspicious / not suspicious.
- Loan applicant is likely to default / not default.
- Customer is high-value / not high-value.
Classification is widely used in business decision systems.
10. Probability-Based Prediction
Many predictive models produce probabilities rather than simple yes/no decisions.
For example:
A model may estimate:
Probability of customer churn = 0.72
This means the model estimates a 72% probability under its assumptions and modeling framework.
It does not mean that the customer is guaranteed to churn.
11. Customer Churn Prediction
Suppose an organization identifies customers with characteristics associated with churn:
- Declining product usage.
- Increased complaints.
- Reduced engagement.
- Shorter sessions.
- Payment issues.
A predictive model may assign each customer a churn probability.
Management can then prioritize retention resources.
12. Demand Prediction
A retailer may use historical data to estimate future demand.
Inputs may include:
- Historical sales.
- Price.
- Promotions.
- Seasonality.
- Inventory availability.
- Market conditions.
Predictions can support inventory and staffing decisions.
13. Predictive Maintenance
Manufacturing and infrastructure organizations can use predictive analytics to estimate equipment failure risk.
Relevant data may include:
- Temperature.
- Vibration.
- Operating hours.
- Maintenance history.
- Error codes.
The organization may schedule maintenance before a major failure occurs.
14. Credit Risk Prediction
Financial organizations may use predictive models to estimate the probability that a borrower will fail to meet obligations.
Potential inputs include:
- Credit history.
- Income information.
- Existing obligations.
- Repayment behavior.
- Account activity.
Models should be carefully governed because financial decisions can have significant consequences for individuals and organizations.
15. Model Evaluation
A predictive model should not be judged only by whether it produces predictions.
Analysts should evaluate:
- Accuracy.
- Precision.
- Recall.
- Error.
- Calibration.
- Stability.
- Generalization.
- Business usefulness.
The appropriate metric depends on the problem.
16. Training and Testing Data
Predictive models are commonly evaluated using separate data.
Training Data
Used to develop the model.
Testing Data
Used to evaluate performance on previously unseen observations.
This helps determine whether the model generalizes beyond the data used to build it.
17. Overfitting
Overfitting occurs when a model learns the training data too closely and performs poorly on new data.
A model may appear highly accurate during development but fail when applied to future observations.
This is a major risk in predictive analytics.
18. Underfitting
Underfitting occurs when a model is too simple to capture important patterns in the data.
An underfitted model may perform poorly on both training and unseen data.
The objective is to develop a model that captures useful patterns without memorizing noise.
19. Model Drift
A predictive model may become less accurate when business conditions change.
Examples:
- Customer behavior changes.
- Economic conditions change.
- Fraud techniques evolve.
- New competitors enter the market.
- Regulations change.
This is known as model drift or performance deterioration over time.
Models should therefore be monitored and periodically reviewed.
20. Prescriptive Analytics
Prescriptive analytics goes beyond prediction by evaluating possible actions and recommending decisions based on objectives, constraints and predicted outcomes.
It addresses:
What should the organization do?
Examples include:
- Which products should be stocked?
- How should employees be scheduled?
- Which customers should receive retention offers?
- How should delivery routes be optimized?
- How should an investment portfolio be allocated?
21. Optimization
Optimization identifies the best or most suitable solution under defined objectives and constraints.
An organization may want to:
- Maximize profit.
- Minimize cost.
- Maximize service quality.
- Minimize delivery time.
- Optimize resource allocation.
Constraints may include:
- Budget.
- Staff availability.
- Production capacity.
- Regulatory requirements.
- Time.
- Inventory.
22. Example of Resource Allocation
Suppose a company has a marketing budget of $1 million.
It must allocate the budget across:
- Search advertising.
- Social media.
- Email.
- Partnerships.
Historical analytics estimates different returns for each channel.
Prescriptive analytics can help determine an allocation that maximizes the expected business objective while respecting budget and operational constraints.
23. Scenario Analysis
Prescriptive decisions should often consider multiple scenarios.
Scenario A: Strong Demand
Demand increases significantly.
Scenario B: Expected Demand
Demand follows the baseline forecast.
Scenario C: Weak Demand
Demand declines.
Management can examine how different decisions perform under each scenario.
24. What-If Analysis
What-if analysis asks:
What happens if an assumption changes?
Examples:
- What if prices increase by 5%?
- What if demand falls by 10%?
- What if supplier costs increase by 15%?
- What if customer churn doubles?
This helps management understand potential consequences.
25. Simulation
Simulation models possible outcomes under uncertainty.
For example, an organization may simulate:
- Demand.
- Costs.
- Exchange rates.
- Delivery times.
- Interest rates.
Thousands of possible scenarios may be generated to estimate the range of potential outcomes.
26. Monte Carlo Simulation
Monte Carlo simulation repeatedly generates possible outcomes using probability distributions and assumptions.
It can help estimate:
- Probability of losses.
- Expected returns.
- Project completion times.
- Inventory requirements.
- Financial risk.
Simulation does not predict one guaranteed outcome.
It provides information about possible outcomes and their probabilities.
27. Decision Trees
A decision tree represents alternative decisions and possible outcomes.
For example:
Launch Product
→ Strong demand
→ Moderate demand
→ Weak demand
Each outcome may have an estimated probability and financial consequence.
Decision trees help managers compare strategic alternatives under uncertainty.
28. Expected Monetary Value
A simplified decision-analysis concept is:
Expected Monetary Value = Σ (Probability × Outcome Value)
Suppose a product launch has:
- 60% probability of generating $2 million profit.
- 40% probability of generating a $500,000 loss.
Expected value:
(0.60 × 2,000,000)+(0.40×-500,000)
= $1,200,000 − $200,000
= $1,000,000
The expected monetary value is therefore $1 million.
Decision-makers should consider risk tolerance and other strategic factors as well.
29. Predictive Versus Prescriptive Analytics
|
Predictive Analytics |
Prescriptive Analytics |
|
Estimates what may happen |
Recommends what may be done |
|
Produces forecasts or probabilities |
Evaluates actions and alternatives |
|
Focuses on future outcomes |
Focuses on decisions |
|
May use statistical and ML models |
Often combines prediction, optimization and constraints |
|
Example: forecast demand |
Example: determine optimal inventory |
The two approaches can be used together.
30. Business Strategy Applications
Predictive and prescriptive analytics can support:
Marketing
- Customer churn prediction.
- Campaign targeting.
- Budget allocation.
Finance
- Cash-flow forecasting.
- Credit-risk prediction.
- Investment analysis.
Operations
- Demand forecasting.
- Inventory optimization.
- Predictive maintenance.
Human Resources
- Workforce forecasting.
- Employee turnover analysis.
- Recruitment analytics.
Supply Chain
- Demand prediction.
- Supplier risk.
- Route optimization.
31. Strategic Decision-Making
Analytics should not replace strategic thinking.
A predictive model may indicate that a market has strong growth potential.
Management must still consider:
- Competitive conditions.
- Regulation.
- Brand strategy.
- Organizational capabilities.
- Capital availability.
- Long-term objectives.
Analytics provides evidence; strategic leadership determines how that evidence should influence action.
32. Model Interpretability
Managers may need to understand why a predictive model produces a particular result.
Interpretability is particularly important where decisions have significant consequences.
Examples include:
- Credit decisions.
- Insurance.
- Employment.
- Fraud investigations.
Highly complex models may produce strong predictions but be difficult to explain.
33. Uncertainty
Predictions always contain uncertainty.
Sources include:
- Incomplete data.
- Measurement errors.
- Changing behavior.
- Model assumptions.
- Unexpected events.
Analysts should communicate uncertainty rather than presenting predictions as guaranteed outcomes.
34. Prediction Intervals and Confidence
Forecasts may be accompanied by ranges rather than a single value.
For example:
Expected sales: $10 million
with a plausible forecast range of:
9million–11 million
A range communicates uncertainty more effectively than presenting $10 million as a guaranteed result.
35. Model Governance
Organizations should establish processes for:
- Model approval.
- Documentation.
- Validation.
- Monitoring.
- Version control.
- Performance review.
- Security.
- Responsible use.
Model governance is particularly important for high-impact decisions.
36. Bias in Predictive Models
Predictive models can reproduce or amplify biases present in:
- Historical data.
- Sampling.
- Feature selection.
- Labeling.
- Business processes.
Analysts should evaluate whether model outcomes are systematically unfair or inappropriate.
37. Privacy
Predictive analytics may use large quantities of personal or behavioral information.
Organizations should consider:
- Data minimization.
- Lawful data use.
- Security.
- Transparency.
- Access controls.
- Appropriate retention.
Responsible analytics requires balancing business value with individual rights and organizational obligations.
38. Strategic Value of Predictive Analytics
Predictive analytics can help organizations:
- Anticipate demand.
- Identify emerging risks.
- Improve resource planning.
- Reduce unexpected costs.
- Improve customer retention.
- Identify opportunities.
Its value increases when predictions are connected to appropriate actions.
39. Strategic Value of Prescriptive Analytics
Prescriptive analytics can help organizations:
- Allocate scarce resources.
- Optimize operations.
- Compare strategic alternatives.
- Improve decision consistency.
- Quantify trade-offs.
However, recommendations are only as good as the:
- Data.
- Models.
- Objectives.
- Constraints.
- Assumptions.
40. Best Practices
Business analysts should:
- Start with a clearly defined business problem.
- Use appropriate and reliable data.
- Select methods appropriate to the decision.
- Separate training and evaluation data where appropriate.
- Monitor model performance.
- Communicate uncertainty.
- Test assumptions.
- Consider multiple scenarios.
- Evaluate business consequences.
- Integrate predictions with optimization where appropriate.
- Maintain human oversight for important decisions.
- Address privacy, fairness and governance.
- Monitor model drift.
- Review models as business conditions change.
Lesson Summary
Predictive analytics estimates what may happen, while prescriptive analytics helps determine what should be done.
Predictive analytics commonly involves:
- Forecasting.
- Regression.
- Classification.
- Probability estimation.
- Time-series analysis.
- Machine learning.
Prescriptive analytics commonly involves:
- Optimization.
- Simulation.
- Scenario analysis.
- What-if analysis.
- Decision trees.
- Resource allocation.
The greatest strategic value occurs when organizations connect:
Data → Prediction → Decision → Action → Measurement → Learning
Analytics should support strategic judgment rather than replace it.