Kaplan Business School · DATA4400 · Trimester 1 2026

Forecasting Foundations
for Business
Decision-Making

Week 1 — Lesson 1
Data-driven Forecasting  ·  Use ← → keys or buttons to navigate

Where Are We Going?

DATA4400 · 12-Week Roadmap

You are at Week 1. Each week builds directly on the previous one.

Week 1
Forecasting Foundations
Week 2
Time Series Properties & GenAI
Week 3
Moving Averages & Correlation
Week 4
Exponential Smoothing & Metrics
Week 5
Assessment 1 (Group, in class)
Week 6
Prophet for Business Forecasting
Week 7
ARIMA & SARIMA Fundamentals
Week 8
VAR & Multivariate Methods
Week 9
Regression Models · A2 Due
Week 10
End-to-End Forecasting
Week 11
Adaptive Forecasting
Week 12
Assessment 3 Presentations

What You Will Learn Today

Week 1 · Objectives
  • 1Understand what forecasting is and why businesses need it
  • 2Understand data-driven decision making in practice
  • 3Identify and distinguish the three types of data used in analysis
  • 4Recognise the basic components of a time series
Business First

Every technique in this subject serves a business question. Always ask: what decision does this help us make?

What is Forecasting?

LO1 · Definition
Definition

Forecasting is a systematic process of predicting future values using past and current data.

Key word: systematic — it is not guessing. It uses structured methods that produce predictions we can measure and improve.

Businesses use forecasting to answer questions like:
  • How many customers will visit next month?
  • How much stock should we order for Easter?
  • When will our water supply drop below a safe level?
  • What will revenue be in Q3?
The Three Time Zones
Past
Historical data
What happened — the evidence we learn from
Present
Current observations
What is happening — the trigger for a decision
Future
The forecast
Our best estimate, with uncertainty quantified

The Three Laws of Forecasting

LO1 · Fundamentals

Accept these three facts before learning any technique. They apply to every forecast ever made.

I

Forecasts are always wrong

No model predicts the exact future. The goal is to be usefully close, not perfectly correct.

Disney forecast summer 2025 crowds from historical data — actual attendance still differed from the forecast.
II

Every forecast must include an error estimate

A prediction without uncertainty bounds is incomplete. Decision-makers need to know how much to trust the number.

"We forecast 50,000 litres ± 8,000" is far more useful than "We forecast 50,000 litres."
III

Short-term forecasts are more accurate than long-term ones

Accuracy degrades the further out you forecast. Too many unknowns accumulate over time.

Predicting tomorrow's dam level is far easier than predicting next year's level.

Why Businesses Need Forecasting

LO1 · Business Value
Business ChallengeWithout ForecastingWith Forecasting
InventoryOver-order (waste money) or under-order (lose sales)Order the right amount at the right time
StaffingToo many staff on quiet days; too few on busy daysMatch rosters to predicted demand
InvestmentJustify spending on gut feeling aloneBack investment decisions with projected revenue
RiskSurprised by downturns or supply problemsAnticipate problems and prepare responses early
Water supplyRestrictions imposed after levels are already criticalAction triggered 6 months before the threshold is breached
Key Insight

Forecasting is both a risk management tool and a growth enabler. The goal is always better decisions — not a perfect number.

Case Study: Why Disney Forecasts Everything

LO1 · Real-World Application
17.7M
Visitors in 2023
$6B+
Annual Park Revenue
2/10
Crowd score — June 2025
Magic Kingdom annual attendance 2006–2023. The 2020 drop (COVID closure) is a structural break — a forecasting challenge. Source: queue-times.com
If Disney gets the forecast wrong:
Operations & Staffing
Too few cast members on a peak day → 3-hour queue times → guest complaints → lost repeat visits
Revenue & Pricing
Dynamic ticket prices depend on predicted demand. A wrong forecast means leaving money on the table or empty parks.
Maintenance Scheduling
Ride maintenance is planned for forecast-low periods. Getting this wrong disrupts operations during peak days.
Food & Inventory
Food waste and stockouts both cost money. Each vendor needs daily demand forecasts to minimise both.
Summer 2025 Insight

Summer became the lowest-crowd season — a decade-long shift that only data revealed. Qualitative intuition would have missed it.

The Three Forecasting Horizons

LO1 · Case Study: Qantas
Qantas Airways
Australia's largest airline. The same business problem — "how do we manage passenger demand?" — leads to three completely different forecasting tasks depending on who is asking and when.
Strategic
"Should we order 20 new aircraft for the next decade?"
The CEO needs a 10-year passenger growth forecast to justify a $4 billion capital investment. Historical route data, population trends, and GDP projections feed the model.
2–10 years ahead
Operational
"How many flights should we schedule for next quarter?"
The operations team forecasts monthly seat demand 3–6 months ahead to set timetables, crew rosters, and fuel orders. Uses seasonal booking patterns and school holiday calendars.
Weeks – months ahead
Tactical
"Should we discount fares for tomorrow's half-empty flight?"
Revenue management uses a short-term model (hours to days) on current booking pace to decide whether to release discounted seats. Wrong decisions cost thousands per flight.
Hours – days ahead

Two Approaches to Forecasting

LO1 · Method Selection
Qualitative

Use when: historical data is scarce or the future will look very different from the past.

  • Relies on structured expert judgement
  • Used for new products, new markets
  • Cannot be validated statistically
Example: A start-up with no sales history uses the Delphi Method to estimate demand for a new product.
Quantitative

Use when: sufficient historical data exists and the future is expected to follow past patterns.

  • Uses statistical and mathematical models
  • Produces measurable error estimates (Law II)
  • Can be validated and compared objectively
Example: A water authority with 10 years of dam data uses SARIMA to forecast the next 5 years.
This Subject

DATA4400 focuses on quantitative time series methods. You need historical data and a defined business question.

Qualitative Methods: Which One to Use?

LO1 · Method Selection Matrix

When data is unavailable, businesses rely on structured expert methods. Use the matrix below to choose the right one.

Internal / Few Experts
External / Many Experts
Little or No Data

Executive Opinion

Senior managers openly discuss and agree on a forecast. Fast, but vulnerable to dominant voices.

Delphi Method

Anonymous expert rounds with feedback until consensus forms. Structured and bias-resistant.

Some Market Data

Grassroots / Sales Force

Sales reps closest to customers estimate demand. Ground-level insight.

Market Research

Surveys and interviews with customers to measure preferences and estimate demand.

Highlighted cells indicate methods most commonly used when quantitative data is completely absent — typical in early-stage or novel business contexts.

Methods in Practice

LO1 · Executive Opinion & Grassroots Forecasting
Executive Opinion
Company: JetStar Australia planning a new Pacific route with no prior passenger history.

The CFO, Head of Operations, and Head of Marketing meet and each give their estimate of year-one passenger numbers. Identities are known — seniority can skew the outcome.

Outcome: A single agreed number (e.g., 180,000 passengers/year) used as the planning assumption. Reviewed quarterly as real data arrives.

⚠ Risk: The CEO's estimate may dominate, even if less informed.
Grassroots Forecasting
Company: Woolworths forecasting fresh produce demand across 1,000 stores for the coming week.

Store managers and department heads each submit their own demand estimates based on what local customers are buying. These are aggregated bottom-up to a national forecast.

Outcome: Avoids over-stocking in slow stores and under-stocking in high-demand areas. Reduces food waste by matching supply to local patterns.

⚠ Risk: Individual estimates vary in quality — some managers are more reliable than others.

Methods in Practice

LO1 · Delphi Method & Market Research
Delphi Method
Context: The Bureau of Meteorology needs a 20-year rainfall forecast for South-East Queensland — a question with no historical precedent.

A panel of 15 climate scientists receives a questionnaire. Identities are hidden from each other. After Round 1, the facilitator shares the range of responses. Outliers revise their estimates in Round 2.

Outcome: After 3 rounds, the group converges on a consensus estimate: a 12% reduction in average annual rainfall by 2045, with an 80% confidence range.

✓ Strength: Anonymity prevents groupthink. Structured rounds improve accuracy over one-off opinion polls.
Market Research
Company: Tesla Australia estimating demand for a new $40,000 EV model before launch.

A survey of 2,000 Australian car buyers asks about purchase intent, preferred price points, and charging infrastructure concerns. Focus groups explore why respondents would or would not switch from petrol.

Outcome: Survey data estimates 14,000 units in year 1. This feeds production planning decisions 18 months ahead of launch.

⚠ Risk: Survey intent does not always convert to purchase. Actual demand may be 30–40% lower.

Knowledge Check

Quiz 1 of 4
Quiz · Qualitative Methods
A pharmaceutical company needs to estimate demand for a brand new drug — there is no sales history anywhere in the world. They assemble 20 independent medical experts across different countries and run anonymous questionnaire rounds until the group agrees on an estimate. Which method is this?

Data-Driven Decision Making

LO2 · DDDM
Definition

DDDM means making decisions backed by evidence, not just intuition.

Forecasting produces the evidence. But evidence only changes decisions if it is communicated clearly. A great forecast presented poorly will be ignored.

1

Define the problem

What decision needs to be made? Who makes it?

2

Gather data

Collect relevant, reliable historical data

3

Analyse & forecast

Apply the appropriate method

4

Communicate results

Present clearly with business implications

5

Decide & review

Act on evidence; revise as new data arrives

Assessment Reminder

In your reports, a number without a business recommendation is incomplete. Always connect your forecast to a real decision.

Three Types of Data

LO3 · Data Classification

Only two of the three types can be used in forecasting. Understanding the difference affects how you collect and structure your data.

Cross-Sectional

Many subjects, one point in time. No time dimension — cannot forecast.

Not used in this subject.
StoreSales (today)
Brisbane CBD$42,000
Southbank$31,500
Chermside$28,000
Time-Series ✓ Used in Forecasting

One subject, measured at regular intervals over time. Primary data type in this subject.

Months must be evenly spaced.
MonthDam Level (%)
Jan 202472.4
Feb 202469.1
Mar 202465.8
Panel Data ✓ Used in Forecasting

Many subjects, each measured over time. Used in multivariate forecasting (Week 8).

Can be unbalanced (some gaps allowed).
StoreMonthSales
BrisbaneJan$42k
BrisbaneFeb$39k
SouthbankJan$31k
Red headings = data types used in forecasting.

Knowledge Check

Quiz 2 of 4
Quiz · Types of Data
A dataset records the monthly passenger counts at Brisbane Airport from January 2015 to December 2024. What type of data is this?

What is a Time Series?

LO4 · Time Series Fundamentals
Definition

A sequence of values recorded at regular time intervals, where the order of observations matters.

The key property: a time series is correlated with itself. This month's value is related to last month's. This is called autocorrelation — and it is what makes forecasting possible.

A time series contains three layers:

  • Trend — long-run direction (up, down, flat)
  • Seasonality — pattern that repeats at fixed intervals
  • Noise — random variation that cannot be predicted
The chart shows monthly airline passenger data 1949–1960 — the dataset used in today's practical work. Notice the upward trend and repeating peaks.
Monthly airline passengers (thousands) — modified_flights.xlsx

Decomposing a Time Series

LO4 · Trend, Seasonality, Noise

Every time series is made up of three layers stacked on top of each other. Separating them helps us understand and forecast each part independently.

Trend

The overall long-run direction. In the airline data, passenger numbers grew steadily year-on-year due to post-war aviation expansion.

Seasonality

A pattern that repeats every 12 months — peaks in July/August (school holidays) and dips in November/February.

Noise (Remainder)

What remains after trend and seasonality are removed. If it still shows a pattern, the model is missing something important.

Knowledge Check

Quiz 3 of 4
Quiz · Time Series Components
Monthly dam water levels drop every summer and recover every winter — consistently, year after year. Which time series component best describes this pattern?

Data Quality: Why It Matters

LO3 · Data Preparation

Your forecasts are only as good as the data behind them. Four things can go wrong — all must be checked before fitting any model.

Problem 1 · Completeness

Missing values in the series

Time series models need a continuous sequence. Gaps in the middle cannot be left blank — they must be filled using imputation before any model is fitted.

Problem 2 · Accuracy

Outliers and data entry errors

A single wrong value (e.g., 9,000 instead of 900) can distort an entire model. One-off events like COVID closures also need to be identified and handled deliberately.

Problem 3 · Consistency

Irregular or mixed time intervals

All observations must be at the same frequency. Mixing some monthly and some quarterly entries breaks the time series structure and makes most methods unusable.

Problem 4 · Relevance

Wrong variables for the question

Including variables with no logical link to the target creates noise, not signal. Every variable in your model must be justified by the business problem.

The next four slides show each problem with a real example from the airline passenger dataset.

Problem 1: Missing Values

Data Quality · Completeness
✗ The Problem
Mar–Apr 1950 and Jul 1952 are missing from the passenger data. The gap causes most forecasting software to either throw an error or produce misleading results.
✓ The Solution
Interpolation fills the gaps using the surrounding values. The imputed points (shown as circles) fit smoothly into the existing pattern — far better than using the series average.

Problem 2: Outliers

Data Quality · Accuracy
✗ The Problem
Jan 1951 shows 850 passengers — an impossible spike. This is likely a data entry error (9,000 typed instead of 900). A model trained on this data will overestimate future peaks.
✓ The Solution
The outlier is replaced with an interpolated value consistent with the local trend and seasonal pattern. Always document what you changed and why — this is part of your methodology.

Problem 3: Inconsistent Intervals

Data Quality · Consistency
✗ The Problem
PeriodPassengersIssue
Jan 2024168,000Monthly ✓
Feb 2024177,000Monthly ✓
Q1 2024522,000Quarterly ✗
Apr 2024193,500Monthly ✓
Q2 2024577,500Quarterly ✗
Mixing monthly and quarterly records creates an uneven time axis. ARIMA, Holt-Winters, and Prophet all assume equally-spaced observations — this will cause errors or silent model failures.
✓ The Solution
PeriodPassengersNote
Jan 2024168,000Monthly ✓
Feb 2024177,000Monthly ✓
Mar 2024177,000Derived ✓
Apr 2024193,500Monthly ✓
May 2024192,000Derived ✓
Convert quarterly figures to monthly by disaggregating (e.g., dividing by 3 and interpolating). The series must be uniform before any model is applied.

Problem 4: Irrelevant Variables

Data Quality · Relevance
✗ The Problem

Forecasting monthly dam water levels using:

VariableLogical Link to Dam Level?
Monthly rainfall (mm)✓ Direct — rainfall fills dams
Population size✓ Indirect — more people = more drawdown
Stock market index✗ None
Average daily temperature✗ Weak — only via evaporation
Number of cricket matches played✗ None
Including irrelevant variables can cause the model to find spurious correlations — patterns in the data that are coincidental, not causal.
✓ The Solution

Only include variables with a clear justification:

Monthly rainfall (mm)
Primary driver of inflows. Directly increases storage levels.
Population (quarterly estimate)
More residents → higher water consumption → faster drawdown.
Month of year (seasonal factor)
Captures known seasonal demand patterns without adding noise.
Every variable in your model must be justified by a logical business or physical mechanism.

Knowledge Check

Quiz 4 of 4
Quiz · Data Quality
You are preparing monthly dam storage data for analysis. You discover three months of data are missing from the middle of the series. What is the correct approach?

Week 1 Summary

Key Takeaways
  • Forecasting is systematic and data-based — not guessing. Forecasts are always wrong, always need an error estimate, and degrade over longer horizons.
  • Businesses need forecasting to manage inventory, staffing, investment, and risk. Disney is a real-world example of forecasting applied at every operational level.
  • The three horizons (strategic, operational, tactical) require different methods and have different accuracy expectations.
  • Time-series and panel data are used in forecasting. Cross-sectional data cannot be used on its own.
  • Every time series has three components: trend, seasonality, and noise. Understanding them drives every method from Week 2 onwards.
  • Data quality must be fixed before modelling — missing values, outliers, inconsistent intervals, and irrelevant variables all cause model failures.
Next Week

Week 2 — Time Series Properties, Data Preparation & Generative AI. You will learn how to describe, clean, and transform a time series, and how GenAI assists with synthetic data generation for Assessment 1.