Product Managers and Demand Planners often find themselves in the position of forecasting demand of new products with unavailable historical information. This presentation will discusses how demand of analogous products can be used to arrive at more accurate forecasts using “Forecasting By Historical Analogy.”
1. How to Forecast With
Limited Historical Data
Saba Dowlatshahi
Data Analyst
DataScience Inc.
2. About us
Saba Dowlatshahi
Data Analyst
Background in quantitative
finance and statistics
Passionate about cutting
edge methods to improve
business and marketing
forecasting
DataScience Inc.
Data Science as a
Service
Customers from
JustFab to Belkin
Ranked #1 among Best
Places to Work in Los
Angeles for 2015
4. Failed Forecasts
4
“It is a perennial sport to make fun of bad forecasts by
experts – but this misses the point. It’s not that our
forecasts are wrong – it’s that the future keeps
surprising us, teaching us shocking, amusing and
unexpected things about modern life. People need to
learn from bad forecasts to make better decisions, for
their organizations and for themselves personally.” -
Eric Garland
5. Challenges In Forecasting With Limited or No
Data
Challenges with quantitative
methods:
▪ Limited or no historical data (e.g. new
products, products with short life
cycles)
▪ Changes in the underlying structure of
the data
▪ Missing data and bad information
5
Challenges with qualitative
methods:
▪ Forecasting without proper data can
be subjective
▪ Judgments are frequently biased
with over-optimism and clouded by
personal and political agendas
▪ Group psychology effects such as
herding influence forecasts
6. Effective Approaches When There is Limited
Data or No Data
▪ Historical Review
▪ Test Markets
▪ Diffusion Modeling
▪ Before-After Trial
Simulation
▪ Executive Judgement
▪ Statistical/Probability
Based Modeling
6
7. Historical Review: Analogous Series
Often, groups of products are
analogous in ways that make them
follow similar time series patterns
as a result of:
✦ Similar consumer tastes
✦ Competition levels
✦ Local economic cycle
Therefore, their time series co-move
(strongly correlate
positively over time).
7
9. Forecasting By Historical Analogy
Forecasting by analogy is a
forecasting method that uses
additional information from
equivalence groups of analogous
series to make a more accurate
forecast than what can be made with
a single series or judgement alone.
9
10. What Type of Data is Required?
▪ Product attributes for prior and
new product
▪ Expert judgement/Product
attributes
▪ Historical data for prior
product releases
▪ New data as it becomes
available
10
11. How Does It Work
▪ Selection of an equivalence groups
▪ Validating equivalence groups
▪ Construction of a model based on
information from the group
▪ Forecasting or simulation from the
available model
12. What Are the Benefits
12
▪ Includes guided statistical analysis
that incorporates human judgment. As a
result, it reduces judgement bias.
▪ Can incorporate a variety of prior information
▪ Forecasts are a blends of a quantitative and
qualitative approach guided by a structure
▪ Can incorporate new data as it becomes
available
13. Case Study: Samsung Product Galaxy
Launch
Overview:
▪ In June 2010, Samsung released the first of its S, "Supersmart,"
high-end Android smart phones named Samsung Galaxy S.
▪ Annually, new versions of the phone were released between the
end of April and beginning of June.
Data:
▪ Consumer interest data in Samsung products were gathered from
Google search. Search results were limited to consumer searches
queries and their misspellings completed within United States.
13
16. Modeling From the Equivalence Group
16
● A Model was created
using a cluster of
GALAXY S4 & S5
● Weights were
determined using
variability of the data