1 / 30

DATA4400
Data-Driven Forecasting

Building End-to-End Forecasting Solutions for Business Impact
Week 10
1 / 30

Learning Outcomes

No. Outcome
1 Apply a forecasting technique to a real-world scenario
2 Understand that forecasting is a part of strategic planning
3 Analyse forecasting as a driver for financial planning
4 Evaluate the Business Case Framework
2 / 30

DATA4400 Roadmap

Week 1-4
Foundations
Week 5
Assessment 1
Week 6-8
Advanced Methods
Week 9
Assessment 2
Week 10
Implementation
Week 11
Adaptive Methods
Week 12
Assessment 3
Today: Bridging technical skills with business implementation
3 / 30

The Real-World Challenge

You have learned the techniques... but how do you apply them?

What You Know

  • Multiple methods: Smoothing, Holt-Winters, ARIMA, SARIMA, VAR, Prophet
  • Different platforms: Excel, Exploratory, Python, Tableau, Orange
  • Statistical validation: MAE, RMSE, MAPE, MASE
  • Model diagnostics and testing

What Organizations Need

  • Business context and requirements
  • Stakeholder buy-in and trust
  • Organizational constraints
  • Actionable recommendations
  • Measurable business value
How do you integrate everything into a solution that creates value?
4 / 30

The Complete Forecasting Pipeline

1

Define Business Problem

Identify decisions, stakeholders, and requirements

2

Collect Data

Gather historical data and external variables

3

Explore & Analyse

Visualize patterns, test stationarity

4

Select Method

Choose appropriate forecasting technique

5

Build Model

Develop and train forecasting model

6

Validate Results

Test accuracy on holdout data

7

Check Accuracy

Calculate performance metrics

8

Monitor Performance

Track ongoing forecast quality

9

Recommendations

Translate forecasts into actions

Each stage is critical for business success
5 / 30

Stage 1: Define the Business Problem

Start with WHY, not HOW

Critical Questions

  • 📋 What business decision needs to be made?
  • ⏱️ What is the forecast horizon?
  • 🎯 What level of accuracy is required?
  • 💰 What are the costs of forecast errors?
  • 👥 Who are the stakeholders?

Example: Retail Inventory

  • Decision: How much inventory to order?
  • Horizon: 3-6 months ahead
  • Accuracy: MAPE less than 15%
  • Cost: Overstock vs lost sales
  • Stakeholders: Procurement, finance, stores
6 / 30

Stages 2-3: Collect & Explore Data

Data quality determines forecast quality

Data Collection

  • Historical data length (minimum 2-3 cycles)
  • Data granularity (daily, weekly, monthly)
  • Handle missing values appropriately
  • Include relevant external variables
  • Verify data source reliability

Exploratory Analysis

  • Visualize patterns (trend, seasonality)
  • Identify outliers and anomalies
  • Check for stationarity (ADF test)
  • Examine autocorrelation (ACF/PACF)
  • Test for structural breaks
7 / 30

Quiz 1: Data Quality Concepts

For monthly sales forecasting, what is the minimum recommended historical data length?
A) 6 months (sufficient for simple patterns)
B) 12 months (one full seasonal cycle)
C) 24-36 months (2-3 seasonal cycles)
D) 60+ months (5+ years)
8 / 30

Stages 4-5: Method Selection & Building

Matching forecasting methods to data characteristics
Data Characteristics Recommended Method Typical Use Case
No trend, no seasonality Simple Smoothing Stable demand products
Trend + seasonality Holt-Winters Seasonal retail products
Complex temporal patterns ARIMA/SARIMA Economic indicators
Multiple related variables VAR Market basket analysis
Multiple seasonality + events Prophet E-commerce, hospitality
9 / 30

Method Selection: Simple Smoothing

Simple Exponential Smoothing

When to use: Data shows no clear trend or seasonality
Platforms: Excel, Python (statsmodels)
Example: Forecasting demand for stable commodity products with consistent consumption
Level remains relatively constant over time
10 / 30

Method Selection: Holt-Winters

Holt-Winters Exponential Smoothing

When to use: Data shows both trend and seasonal patterns
Platforms: Tableau, Python (statsmodels), Exploratory
Example: Seasonal retail products like winter clothing, summer beverages
Key consideration: Choose between additive and multiplicative seasonality
11 / 30

Method Selection: ARIMA/SARIMA

ARIMA and SARIMA Models

When to use: Complex temporal patterns requiring sophisticated modeling
Platforms: Exploratory, Orange, Python
Example: Economic indicators, financial time series, stock prices
Requirements: Sufficient historical data, stationarity (or differencing)

ARIMA (p, d, q)

  • p: Autoregressive order
  • d: Differencing order
  • q: Moving average order

SARIMA (p, d, q)(P, D, Q)m

  • P, D, Q: Seasonal components
  • m: Seasonal period
  • Captures both non-seasonal and seasonal patterns
12 / 30

Method Selection: VAR & Prophet

Vector Autoregression (VAR)

When to use: Multiple related time series variables
Platform: Orange Data Mining
Example: Market basket analysis (forecasting sales of complementary products)
Key advantage: Captures interdependencies between variables

Prophet

When to use: Multiple seasonality + holiday effects
Platform: Exploratory
Example: E-commerce sales, hotel bookings, restaurant revenue
Key advantage: Handles missing data and outliers robustly
Decision Principle: Match method complexity to data characteristics and business requirements
13 / 30

Activity 1: Method Selection Exercise

Instructions

Form groups of 3-4 students. Review the three business scenarios on the next slide.
For each scenario, recommend:
  • Best forecasting method and justification
  • Preferred software platform
  • Key data requirements
  • Expected challenges
Time allocation: 15 minutes for group discussion
14 / 30

Activity 1: Business Scenarios

Scenario A: E-commerce Sales

  • 5 years daily sales data
  • Weekly seasonality + holiday peaks
  • Multiple product categories
  • 90-day forecast needed
  • Marketing campaigns affect demand

Scenario B: Energy Demand

  • 10 years hourly data
  • Daily and seasonal patterns
  • Weather-dependent
  • Day-ahead forecasts needed
  • High accuracy critical (±2%)

Scenario C: B2B Sales

  • 3 years monthly data
  • Trending growth but volatile
  • Economic indicators influence
  • Quarterly revenue forecasts
  • Limited historical data
Consider data characteristics, business requirements, and implementation feasibility
15 / 30

Quiz 2: Method Selection

A retail chain has daily sales data for 3 years showing strong weekly patterns (weekend peaks) and yearly seasonality (holiday shopping). They need to forecast 90 days ahead for inventory planning. Which method is most appropriate?
A) Simple Exponential Smoothing (captures basic level)
B) ARIMA (handles complex patterns)
C) Prophet (handles multiple seasonality + events)
D) VAR (captures multiple variables)
16 / 30

Implementation Challenges: Technical

Technical excellence does not equal organizational success

Data Integration

Combining data from multiple systems (ERP, CRM, external sources) with different formats, granularities, and quality levels

Computational Resources

Managing scalability for large datasets, real-time forecasting requirements, and computational efficiency

Model Deployment

Automating forecast generation, scheduling updates, and integrating with business systems

Interpretability vs Complexity

Balancing model accuracy with stakeholder understanding and trust

Version Control

Tracking model changes, maintaining reproducibility, and documenting assumptions

Model Maintenance

Monitoring performance degradation, retraining schedules, and adapting to concept drift

17 / 30

Implementation Challenges: Organizational

Resistance to Change

"We've always done it this way" mentality. People may trust their intuition over statistical forecasts, especially when models contradict experience.

Trust in Forecasts

Building confidence in model predictions, especially after initial errors. Requires demonstrating consistent value over time.

Skill Gaps

Organization may lack technical expertise to maintain models or interpret results. Training requirements can be substantial.

Incentive Alignment

Existing KPIs may discourage forecast adoption. For example, sales teams may prefer conservative forecasts to easily beat targets.

Stakeholder Expectations

Managing unrealistic accuracy expectations. Forecasts are not crystal balls - communication is critical.

Political Dynamics

Forecasts may threaten existing decision-making authority or reveal uncomfortable truths about business performance.

18 / 30

Change Management Strategies

Getting people to use your forecasts

1. Executive Sponsorship

Secure leadership support and establish urgency with metrics-driven storytelling. Quantify current costs of poor forecasting.

2. Stakeholder Engagement

Involve users early in the design process and create a coalition of data champions across departments.

3. Training & Support

Develop comprehensive training programs tailored to different user groups. Provide ongoing support channels.

4. Gradual Implementation

Start with pilot projects in receptive departments, demonstrate value, then scale across organization.

5. Communication Plan

Regular updates on forecast performance, success stories, and continuous improvement efforts.

6. Performance Metrics

Show tangible business value: cost savings, revenue gains, improved service levels, waste reduction.
19 / 30

Stages 6-8: Validation, Accuracy & Monitoring

Performance Metrics

Accuracy Measures

  • MAE: Mean Absolute Error
  • RMSE: Root Mean Square Error
  • MAPE: Mean Absolute Percentage Error
  • MASE: Mean Absolute Scaled Error

Model Selection

  • AIC: Akaike Information Criterion
  • BIC: Bayesian Information Criterion

Business Metrics

Business Impact

  • Cost of forecast errors
  • Service level achievement
  • Inventory turnover improvement
  • Stockout reduction

Comparison Benchmarks

  • vs. naive forecast
  • vs. current method
  • vs. alternative approaches
  • Confidence intervals
20 / 30

Building a Monitoring System

Forecasts degrade over time - monitor continuously

Real-time Dashboard

Track accuracy metrics and compare forecast vs actual values. Visualize performance trends over time.

Automated Alerts

Notification when accuracy degrades below threshold or unusual patterns emerge.

Regular Reviews

Weekly performance reports and monthly deep-dive analyses with stakeholders.

Feedback Loop

Collect user feedback, analyze systematic errors, implement continuous improvements.

21 / 30

Quiz 3: Performance Metrics

Your retail forecast has MAPE of 18%, but management requires less than 15%. The cost of understocking (lost sales) is $50 per unit, while overstocking costs $10 per unit. What should be your priority?
A) Focus only on reducing MAPE below 15%
B) Accept current MAPE and focus on minimizing understocking errors (bias adjustment)
C) Abandon the forecasting project
D) Use a simpler model to achieve 15% MAPE
22 / 30

Embedding Forecasts into Strategic Planning

Strategic Planning Cycles

Annual and quarterly business reviews use forecasts as key inputs for strategic decisions about market expansion, product portfolio, and resource allocation.

Scenario Analysis

Use forecasts to simulate different market or economic conditions (best case, base case, worst case) for risk management.

Balanced Scorecards

Integrate forecast metrics (sales, churn, demand) as lead indicators of future performance alongside lagging financial metrics.

Example: Qantas Airways

Qantas uses passenger demand forecasts as a core input to route planning, fleet allocation, and partnership negotiations. Forecasts inform not only what flights to schedule but also strategic partnerships and pricing elasticity models.

23 / 30

Forecasts as Drivers of Financial Planning

Forecasting accuracy has direct implications for budgeting, capacity planning, and resource allocation

Adaptive Budgeting

Forecasts feed rolling budgets or driver-based planning models rather than static annual budgets.

Resource Allocation

Staffing, inventory, and logistics align with predicted demand peaks or troughs.

Customer Management

Forecasts of churn or demand inform retention campaigns and pricing strategies.

Example: Woolworths Australia

Woolworths integrates predictive demand models to optimize inventory across 1,000+ stores. Forecasts reduce food waste by 15-20% and improve shelf availability, directly impacting profit margins and sustainability goals.

24 / 30

Activity 2: Strategic Implications

Scenario

Work in pairs. Assume you forecast a 12% increase in online sales next quarter for a retail business.

Outline the implications for:

  • Budget: What additional costs will be incurred? (e.g., inventory, warehouse space, technology infrastructure)
  • Staffing: How many additional staff needed in which departments? (e.g., warehouse, customer service, delivery drivers)
  • Customer Strategy: How to prepare for increased demand? (e.g., communication, service levels, delivery capacity)
Time allocation: 10 minutes discussion, then share with class
25 / 30

Business Case Framework

Connect forecast to business improvement - the value proposition
1. Executive Summary: Problem, solution, ROI
2. Current State: Pain points and costs
3. Proposed Solution: Method and technology
4. Financial Analysis: Costs, benefits, ROI
5. Risk Assessment: Risks and mitigation
6. Implementation Plan: Timeline and milestones
Critical Principle: Business cases must be concise (1-2 pages), quantitative (show the numbers), and decision-focused (clear recommendation).
26 / 30

Quiz 4: Business Case Components

What is the MOST important element to include in the financial analysis section of a forecasting business case?
A) Detailed technical specifications of the forecasting model
B) Comparison of all available forecasting methods
C) Quantified costs, benefits, and ROI with clear timeline
D) History of forecasting in the organization
27 / 30

Activity 3: Create a Business Case

Instructions

Work in pairs. Develop a 1-page business case outline for implementing a forecasting solution in a business context of your choice (or use your A3 project).

Include:

  • Problem statement (What business problem needs solving?)
  • Proposed solution (Which forecasting method + which software tools?)
  • Expected costs (Implementation + ongoing operational costs)
  • Expected benefits (Quantified improvements in revenue, cost savings, or service)
  • Implementation timeline (3-6 month roadmap with key milestones)
  • Key risks and mitigation (Top 3 risks and how to address them)
Time allocation: 20 minutes, then share with partner for feedback
28 / 30

Key Takeaways

1. Forecasting ≠ Just Technical Accuracy
Success requires business context, stakeholder management, and clear value proposition
2. Business Context Drives Method Selection
Match forecasting techniques to data characteristics, business requirements, and organizational capabilities
3. Implementation is 50% People, 50% Technology
Change management, training, and communication are as critical as model accuracy
4. Your Forecast Must Answer: "So What?"
Always translate predictions into actionable business decisions with clear financial impact
5. Monitor and Adapt Continuously
Forecasts degrade over time - establish monitoring systems and feedback loops
29 / 30

Quiz 5: Course Synthesis

Which statement BEST describes the relationship between forecasting and strategic planning?
A) Forecasting is a technical exercise separate from strategic planning
B) Strategic planning should be completed before any forecasting begins
C) Forecasts are integral inputs to strategic planning cycles, scenario analysis, and resource allocation decisions
D) Only financial forecasts matter for strategic planning

Next Week: Adaptive Forecasting

Week 11 will focus on matching methods to business problems and adapting forecasts to changing conditions

Assessment 3 due Week 12

30 / 30