1
2
1 Overview of Demand Forecast
The demand forecast is a fundamental component of an organization’s Sales and Operation Planning (S&OP) process.
Developing an advanced forecasting process can greatly enhance demand and supply alignment.
Continuous Supply Chain Configuration
Potential disadvantages of not adopting automated planning
Lower
Historical Sales Sales Plan/Budget Market Intelligence
Forecast
Quality
Higher
Inventories &
net working
capital
Multiple Scenarios
The lower
degree of
Supply Chain Configuration Forecast automation
Lower
Purchase
D adoption of
r
vi technological
e
m Production s possibilities
r
o
f
n
I Inventory
Higher
Balancing
Real demand
planning
Distribution headcount
Source: adapted from Deloitte (2019)
3
AstraZeneca: Forecast Evaluation Project
1 Context and Approach
The forecast process implemented by AstraZeneca has moved beyond manual forecasting, adopting forecasting tools with
advanced algorithms for demand forecast. The company seeks to find areas of improvement for their forecasting process.
Evolution of Demand Planning
1 No/Naïve 2 Manual 3 Statistical 4 Advanced 5 Bionic
Forecasting Forecasting Forecasting Planning Demand
Planning
• Entirely reactive • No analytics • Basic algorithm. • Advanced
• Machine
• Based on • Medium • Capabilities and algorithms.
Learning.
historical data. capabilities. maturities vary. • Challenge the
• Differentiated
• Low • Medium • Little free time. plan.
planning.
capabilities. maturity. • Achieve target is • Plan by
• Root cause
• Low maturity. • Few KPIs and the mission. scenarios.
analysis.
• No target little target • Basic processes • Plan by
• Analytics
measuring. measuring. in place. segmentation.
planning.
• Medium free • Unharmonized • Chase the data. • Plan by past
• Inherent
time. objectives • Continuous performance.
automated
across planning performance • Prioritisation
performance
areas. improvement. process.
improvement.
• Financial
planning.
Source: adapted from Deloitte (2022)
4
AstraZeneca: Forecast Evaluation Project
1 Context and Approach
Using sales volume data as a sample and demand forecasting best practices, the project will analyse the dataset and
identify accuracy metrics, product segmentation, and appropriate forecast models as the base of the forecast evaluation.
Simplified Best Practices in Demand Forecasting
OBJECTIVE DATA METRICS BASELINE REVIEW
• What is the • Data of • Relevant MODEL PROCESS
purpose of the unconstrained metrics to • Use forecast • Enrichment
forecast? demand. assess forecast model to phase by
• What decisions • Ideally include quality. generate various teams
are supported external factors • Ideally include baseline model (FVA).
by this forecast? affecting the external factors with wide range • Promote
• Who will use demand affecting the of insights. ownership and
this forecast? (promotion, demand accountability.
competitors, (promotion, • Measure how
pricing) competitors, the forecast gets
pricing) better or worse.
Source: adapted from Vandeput (2023)
• Determine the objective of • Data cleaning • Cross-validation (training • Produce forecast based on • Typical Collaborative
the forecast. • Product segmentation data and test data) all available data points. Planning, Forecasting, and
• Predict new product? • Calculate forecast metrics. Replenishments between
departments.
5
AstraZeneca: Forecast Evaluation Project
1 Context and Approach
Framework and project scope
6
AstraZeneca: Forecast Evaluation Project
2 Data and Forecast Overview
Based on the granularity of the dataset, the result of the forecast is better suited for tactical planning, commonly used to
determine monthly Sales and Operation Planning (S&OP) of the company.
Forecast Dimensions
Materiality Geography Temporality
• Sales volume of 90 • Sales volume of • Sales volume data
All SKUs (product). World AstraZeneca Year recorded monthly from
• Measuring metric in Australia 2018 to 2023.
units of sale.
Brand Country Month
Product Store Week
Material Geographical Temporal
Source: adapted from Vandeput (2023)
Planning Horizons and Types
Operational Tactical Long-term
Forecast Horizons
Typical Granularity SKU-customer-resource Product group, country Product line, region, asset type
• Forecast set to predict sales
Time buckets
Day/week Month Month/year volume in 2024 – 2025.
• Classified under tactical horizons
Horizons
4 – 12 weeks 12 – 36 months 3 – 10 years to determine monthly S&OP.
Unit
Volume Volume & value value
Use case Campaign, peak season Monthly S&OP Annual planning, Product plan Source: adapted from Sankaran G et al. (2019)
7
AstraZeneca: Forecast Evaluation Project
2 Data and Forecast Overview
Forecast Metrics Used and Forecast Models
8
AstraZeneca: Forecast Evaluation Project
2 Data and Forecast Overview
Forecast Metrics Used and Forecast Models
9
AstraZeneca: Forecast Evaluation Project
3 Product Segmentation and Forecast Models
By using two-dimensional ABC analysis for segmentation, AstraZeneca can focus the forecast effort on commercially
critical products that require further human intervention and attention for more accurate forecasts.
ABC XYZ Analysis on the dataset
Importance
Importance
Measure : Sales volume
8 1 37 High Volume : contribute to 80% of total
sales volume.
High Priority
Medium Volume: contribute to next 15% of
Medium
High Volume Volume Low Volume total sales volume.
Low Low Low Low Volume: contribute to 5% of total sales
Forecastability Forecastability Forecastability
y volume.
t
i
l
i 7 11 23
b
a
t
s Forecastability
Medium
a
c High Volume Volume Low Volume
e
r Moderate Moderate Moderate Measure : Coefficient of Variation
o Forecastability Forecastability Forecastability
F (variability of sales volume)
High Forecastability : CV less than 0.2
0 2 1
Medium Forecastability: 0.2 <= CV <= 0.5
Low Forecastability: CV more than 0.5
Medium Low Priority
High Volume Volume Low Volume
High High High
Forecastability Forecastability Forecastability
Source: adapted from Vandeput (2023) Trivial Many vs Important Few
Appendix 1: Calculating ABC XYZ Analysis
10
AstraZeneca: Forecast Evaluation Project
3 Product Segmentation and Forecast Models
Further identification based on Average Demand Interval (ADI) yields a result that at least 7 SKUs can be classified as
products with intermittent demand that require a different approach than products with a smooth demand.
11
AstraZeneca: Forecast Evaluation Project
3 Product Segmentation and Forecast Models
Determine the best approach and forecast models based on demand characteristics of each segment.
12
AstraZeneca: Forecast Evaluation Project
3 Product Segmentation and Forecast Models
Forecast Model Validation
13
AstraZeneca: Forecast Evaluation Project
4 Demand Forecast for Sales and Operation Planning (S&OP)
Integrating qualitative input from various departments and quantitative input from statistical forecasting is essential in
effective demand review in Sales and Operation Planning (S&OP).
Source: adapted from Kinaxis (n.d)
Source: adapted from Sankaran G et al. (2019)
14
AstraZeneca: Forecast Evaluation Project
4 Demand Forecast for Sales and Operation Planning (S&OP)
Judgmental adjustment is often used as part of
Source: adapted from Brau, R, Aloysius, J, & Siemsen, E (2023)
15
AstraZeneca: Forecast Evaluation Project
4 Demand Forecast for Sales and Operation Planning (S&OP)
Using FVA to manually adjust the forecast result and track the performance metric.
Source: adapted from Brau, R, Aloysius, J, & Siemsen, E (2023)
16
AstraZeneca: Forecast Evaluation Project
5 Forecast Tools
Forecast tools can improve accuracy and reduce cost of generation compared to manual forecasting.
Source: adapted from Mckinsey and Company (2022)
17
AstraZeneca: Forecast Evaluation Project
5 Forecast Tools
Forecast tools can improve accuracy and reduce cost of generation compared to manual forecasting.
18
AstraZeneca: Forecast Evaluation Project
5 Forecast Tools
Forecast tools can improve accuracy and reduce cost of generation compared to manual forecasting.
Source: Statista (2023)
19
AstraZeneca: Forecast Evaluation Project
5 Forecast Tools
Forecast tools can improve accuracy and reduce cost of generation compared to manual forecasting.
Source: adapted from SAS Institute (2013)
20
AstraZeneca: Forecast Evaluation Project
5 Forecast Tools
Integrating Machine Learning and AI with the forecast process and S&OP
Source: adapted from SAS Institute (2013)
21
AstraZeneca: Forecast Evaluation Project
6 Recommendation
AstraZeneca should focus the forecast effort on the most important products (by sales volume)
• Use unconstrained demand rather than constrained sales (Vandeput 2023)
• Use and review forecasts using FVA Framework
• Use value-weighted and forward-looking demand to classify segmentation,
instead of historical sales (what will be in high demand in the next several
months or year).
•
22
AstraZeneca: Forecast Evaluation Project
Appendix 1. Calculating Segmentation Using ABC XYZ Analysis Framework
23
AstraZeneca: Forecast Evaluation Project
References
• Abolghasemi, M, Ganbold, O, & Rotaru, K 2024, ‘Humans vs. large language models: Judgmental forecasting in an era of advanced AI’, International Journal of
Forecasting, p. S0169207024000700, viewed 10 October 2024,
• AstraZeneca 2024, H1 and Q2 2024 results, viewed 6 August 2024, announcement.pdf>. • Benhamida, FZ et al. 2021, ‘Demand Forecasting Tool For Inventory Control Smart Systems’, Journal of Communications Software and Systems, vol. 17, no. 2, pp. 185– 196, viewed 10 October 2024, • Boylan, J & Syntetos, AA 2021, Intermittent demand forecasting: context, methods and applications, First edition, Wiley, Hoboken, NJ. • Brau, R, Aloysius, J, & Siemsen, E 2023, ‘Demand planning for the digital supply chain: How to integrate human judgment and predictive analytics’, Journal of Operations Management, vol. 69, no. 6, pp. 965–982, viewed 3 October 2024, • Chase, CW 2021, Consumption-based forecasting and planning: predicting changing demand patterns in the new digital economy, Wiley, Hoboken, New Jersey. • Chopra, S & Meindl, P 2016, Supply chain management: strategy, planning, and operation, Sixth edition, global edition, Pearson, Boston Columbus Indianapolis New York San Francisco Amsterdam Cape Town Dubai London Madrid Milan Munich Paris Montréal Toronto Delhi Mexico City São Paulo Sydney Hong Kong Seoul Singapore Taipai Tokyo. • De Mattos, CA, Correia, FC, & Kissimoto, KO 2024, ‘Artificial Intelligence Capabilities for Demand Planning Process’, Logistics, vol. 8, no. 2, p. 53, viewed 3 October 2024, • Deloitte 2019, Supply Chain Planning 2025 No Planning, Continuous Planning and Beyond, viewed 20 September 2024, • Deloitte 2022, Bionic Demand Planning, viewed 2 October 2024, bionic-demand-planning.pdf>. • Hugos, MH 2024, Essentials of supply chain management, Fifth edition, Wiley, Hoboken, New Jersey. • IBISWorld 2023, AstraZeneca Holdings Pty Ltd, IBISWorld, viewed 3 August 2024, • Mckinsey & Company 2022, AI-driven operations forecasting in data-light environments, viewed 10 October 2024, • Mentzer, JT & Moon, MA 2007, Sales forecasting management: a demand management approach, 2. ed., [Nachdr.], Sage Publications, Thousand Oaks, Calif. • Petropoulos, F et al. 2022, ‘Forecasting: theory and practice’, International Journal of Forecasting, vol. 38, no. 3, pp. 705–871, viewed 4 October 2024, 24 AstraZeneca: Forecast Evaluation Project References • Rathipriya, R et al. 2023, ‘Demand forecasting model for time-series pharmaceutical data using shallow and deep neural network model’, Neural Computing and Applications, vol. 35, no. 2, pp. 1945–1957, viewed 14 August 2024, • Sankaran, G et al. 2019, Improving Forecasts with Integrated Business Planning: From Short-Term to Long-Term Demand Planning Enabled by SAP IBP, Springer International Publishing, Cham, viewed 3 October 2024, • SAS Institute 2013, Enhancing Sales and Operations Planning with Forecasting Analytics and Business Intelligence, SAS, viewed 4 October 2024, • SAS Institute 2015, Forecast Value Added Analysis: Step by Step, SAS, viewed 4 October 2024, whitepapers-ebooks/sas-whitepapers/en/forecast-value-added-analysis-106186.pdf>. • SAS Institute 2018, A Practical View of New Product Forecasting, SAS, viewed 4 October 2024, whitepapers-ebooks/sas-whitepapers/en/new-product-forecasting-109813.pdf>. • SAS Institute 2019, What Management Must Know About Forecasting, SAS, viewed 4 October 2024, whitepapers-ebooks/sas-whitepapers/en/management-forecasting-104529.pdf>. • Saunders, LW et al. 2024, ‘New product family demand planning: Addressing SKU ‐level spread bias’, Journal of Business Logistics, vol. 45, no. 2, p. e12373, viewed 3 October 2024, • Siddiqui, R et al. 2022, ‘A hybrid demand forecasting model for greater forecasting accuracy: the case of the pharmaceutical industry’, Supply Chain Forum: An International Journal, vol. 23, no. 2, pp. 124–134, viewed 10 October 2024, • Silver, EA, Pyke, DF, & Thomas, DJ 2021, Inventory and production management in supply chains, Fourth edition, first issued in paperback, CRC Press, Taylor & Francis Group, Boca Raton, FL London New York, NY. • Van Steenbergen, RM & Mes, MRK 2020, ‘Forecasting demand profiles of new products’, Decision Support Systems, vol. 139, p. 113401, viewed 10 October 2024, • Vandeput, N 2023, Demand forecasting best practices, Manning Publications Co, Shelter Island, NY. • Vandeput, N & Makridakis, SG 2021, Data science for supply chain forecasting, 2nd edition, De Gruyter, Berlin Boston. 25 AstraZeneca: Forecast Evaluation Project
