New Product = Gambling? Forecasting - Lancaster · PDF fileNew Product Forecasting with Analytics ... are most important ... 200Adapt pattern to new product Repository of clusters

Page 1

New Product Forecasting with Analytics© 2016 by Sven F. Crone - all rights reserved.

Universität Hamburg Institut für Wirtschaftsinformatik

Prof. Dr. D.B. Preßmar

Dr. Sven F. Crone

Dr. Nikolaos Kourentzes

Predictive Analytics for New Product Forecasting

More Statistics and less Judgment:More Analytics and less Judgment: a Case Study

New Product = Gambling?

Forecasting

Page 2


Statistics? Judgment?

Both?

Statistics vs. Judgment in New Product Forecasting?

e.g. Delphi

Szenarios

Market Tests

Prediction Markets

e.g. Norton-Bass

Classification

Clustering

Social-media analysis

…

How to forecast high-tech products (CPU,RAM) at Intel? Product sourcing & production lead times are longer than sales

period every product is “new”

Often requires building new production facilities (or even plants) high costs for under & overforecasting

100s to 1,000s of new products per season

2-4 innovation cycles (selling seasons) per year Requires semi-automated & standardized process

How to they do New product forecasting?

The Challenge

Page 3


Sales of products with continuous repeat purchasing?

Norton-Bass Model

Estimatetotal marketvolume fromfirst 3 sales

observations

Common Problem

Catalogue Retailers

order from China / India

3-6 months lead time

4 catalogues per year

Fashion Retailers

order from China / India

3-6 months lead time

3-12 seasons per year

Sourcing lead times are

longer than sales period

(for many products)

many products are “new”

Page 4


CDs, DVDs, BlueRays, Video Games …

Product launch & intial sales before, during & after go-live (i.e. pipeline-filling) are most important

Common Problem

Common Problem

… check out the scaled thumbnails of the first 8 weeks

of sales time series plots for 100 new DVD releases

Page 5


Few things are completely new!

Delphi

Szenarios

Market Tests

Prediction Markets

Forecast using Analogies

How to find similar products from history (aka Analogies) automatically?

Finding Analogies: you try it!

Now

find:

Cluster 1: Cluster 2:

aka statistical technique called „Clustering“

Page 6


Finds clusters of typical initial sales behavior

5 10 15 20 250

200

400

600

800

1000

1200

Time series with 27 months of observations

Hard to extract useful information!

1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 30

200

400

600

800

1000

1200

1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 30

200

400

600

800

1000

1200

No additional information

Find time series that

behave similarly

1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 30

200

400

600

800

1000

1200

… using Kmeans clustering

1 2 30

500

1000

1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 30

200

400

600

800

1000

1200

1 2 30

500

1000

1 2 30

500

1000

1 2 30

500

1000

Focus on the first three months (new products)

Analogies in Time Series?

Need to be careful in specifying SIMILIARITY !

New product to be launched

Only one initial order Match to most similar cluster

1 2 30

500

1000

1 2 30

500

1000

1 2 30

500

1000

1 2 30

500

1000

(plus 10s of product attributes etc.)

Analogies in Time Series?

Compare

similarity

Multiple observations …

Take average shape of

cluster as forecast!

Page 7


What is “Similarity”?


Page 8




Page 9




Page 10


Distance

0.98

0.07

0.21

0.43

Rank

4

1

2

3

Database

n datapoints

n

iii sqSQD

1

2,

S

Q

Euclidean Distance between

two time series Q = {q1, q2, …, qn}

and S = {s1, s2, …, sn}

Find similar

shapes in

time series

pn

i

p

ii cqCQD 1

,

Manhattan

(p=1)

Max (p=inf) MahalanobisWeighted

Euclidean

Similarity in Time Series?

Empirical

Testing!

Theory?

Frost!

Tongue!

Not a

good

idea

How do we know that it works?

Page 11


• Curtains

• Tiebacks

• Cushion covers

• Valances

in various fabrics, linings, sizes, colours …

Case Study Task(s)

Clustering on the

first three observations

Clustering on the

first six observations

Clustering Approach Objectives?

Initial behaviour matching

Very short history

Longer history

Recalibrate matching

Clustering of the

1st pre-order

& product attributes

Forcast inital pipline filling

Without sales history

after the 6th month the history is long enough to use time series methods

support complete launch process!allocate to clusters on more data

Page 12


Sample new product – First 52 weeks of deliveries

Pipeline filling Normal behaviour

How to forecast with zero weeks of sales?

Know initial

delivery volume

Automatic forecasting by structured analogy

Products0 10 20 30

20

40

60

80

100

120

140

160

Sales history

0 10 20 3020

40

60

80

100

120

140

160

0 10 20 3020

40

60

80

100

120

140

160

X

5 10 15 200

50

100

150

200

Classification to

repository

5 10 15 200

50

100

150

200

Adapt pattern

to new product

Repository of

clusters (with

archetypical new

product sales)

Initial known orders

(interview, trial etc.)

3000 units

75000 units

45 units...

Calculate

safety stock

and use!

Stage I - Clustering

Stage II - Forecast

time series clustering

New Product Clustering

Page 13


Clustering Results3 first observations

I. 262 products used with history up to 27 months

II. Each product → Only first three months used

III. 8 clusters → Optimal number

IV. Each cluster → Average behaviour found

V. Average behaviour → New products will be matched with

1 1.5 2 2.5 3

200

400

600

800

1000

Cluster 1 Members: 14

1 1.5 2 2.5 3

200

400

600

800

1000


1 1.5 2 2.5 3

200

400

600

800

1000


1 1.5 2 2.5 3

50

100

150

200


1 1.5 2 2.5 3

50

100

150

200


1 1.5 2 2.5 3

50

100

150

200


1 1.5 2 2.5 3

50

100

150

200


1 1.5 2 2.5 3

50

100

150

200


Cluster 1 Cluster 2 Cluster 3 Cluster 4

Cluster 5 Cluster 6 Cluster 7 Cluster 8

Business intelligence can eliminate/correct clusters → Cluster 5 case

Clustering Results3 first observations

Page 14


Forecasting

Using the estimation for the first month value and the product type

find the correct cluster using the three first observations clusteringI.e.g. First value = 150, Unlined curtain Cluster 7!

Forecasting


find the correct cluster using the three first observations clusteringI.Use the % change of the average shape of the cluster to form the

2nd and 3rd valueII.

Second Value (Month):

150 * (1-0,4823) = 77.65

Third Value (Month):

77.65 * (1+0,2044) = 93.52

Page 15


Forecasting


find the correct cluster using the three first observations clustering.I.Use the % change of the average shape of the cluster to form the

2nd and 3rd value.II.For the 4th-6th month forecasts use the clustering of the first 6 months

to find the product membership (similar procedure!).III.

Update the forecasted values with real whenever they are available.

If possible re-evaluate membership when new values are available.IV.

To forecast after the sixth value/month use a statistical model. The

product history is now long enough!V.

Time series clustering always outperforms statistical approach

Time series clustering outperformed human demand planner

How accurate is it?

Evaluate on time series of the given data

Statistical ForecastingProposed Approach

1 1.5 2 2.5 30

200

400

600

800

1000

1200

Test1

Test2

Test3

Forecast1

Forecast2

Forecast3

1 1.5 2 2.5 30

200

400

600

800

1000

1200

Test1

Test2

Test3

Forecast1

Forecast2

Forecast3

New Product ForecastingExperimental Evaluation

Page 16


5 10 15 20 250

100

200

300

400

500

600

700

New Product Clustering

5 10 15 20 250

100

200

300

400

500

600

700

5 10 15 20 250

100

200

300

400

500

600

700

5 10 15 20 250

100

200

300

400

500

600

700

5 10 15 20 250

100

200

300

400

500

600

700

... ...

Historical Data

Clustering Profiles

Allows us to

measure safety stock!

5 10 15 20 250

100

200

300

400

500

600

700

Empirical EvaluationInventory Performance

Inventory Performance Results

150 200 250 30020

40

60

80

100

120

140

160Sales

Inventory

Sto

ck-o

ut

150 200 250 30020

40

60

80

100

120

140

160Sales & Features

Inventory

Sto

ck-o

ut

S1: Features

S2: Initial Orders

S3: Features & I. Orders

S4: Features & Sales (2)

S5: Features & Sales (6)

Clustering on

Page 17


Outcome: simple MS EXCEL application that can be automated

Clustering Results

A simple Forecasting Tool

Take aways

Many industries specialize onNew Product Forecasting opportunities to learn (steal fire)

Clustering creates insightful groups of similar products helpful to planners to improve judgment

Allows large-scale automation of new product forecasting

Predictive Analytics provides new solutions to forecasting challenges

… room for improvement![LCF ISF09, SAS New Product Presentation ISF 2008]

Looking for

companies for a

free-of-charge

pilot study

(to present at IBF?)

Page 18


Dr. Sven F. CroneAssistant Professor, Director

Lancaster University Management School

Research Centre for Forecasting

Lancaster, LA1 4YX

Tel. +44 1524 5-92991

[email protected]

Questions?

Dr. Sven F. CroneAssistant Professor, Director

Lancaster University Management SchoolResearch Centre for Forecasting

Lancaster, LA1 4YXTel. +44 1524 5-92991

[email protected]

Copyright © All rights reserved, Dr. Sven F. Crone

These slides are from a presentation on forecasting.

You may use these slides for internal purpose only, so long as they

are clearly identified as being created and copyrighted

© by Sven F. Crone.

All rights are reserved! You may not use the text and images in a

paper, tutorial or external training, duplicate or replicate or alter

them, or make them accessible as is, without explicit prior

permission from the author.

Documents

New Product = Gambling? Forecasting - Lancaster · PDF fileNew Product Forecasting with Analytics ... are most important ... 200Adapt pattern to new product Repository of clusters