14
Sweet Tooth Bakery Sweet Tooth Bakery Cake preference conjoint analysis Cake preference conjoint analysis Michael Bystry, Stacy Comrie, Shannon Goyda, Prochi Jain, Amy Kasten

Conjoint class project

Embed Size (px)

DESCRIPTION

This is conjoint class project to assist a bakery in optimizing its product offerings. It was completed as part of a Conjoint/Discrete Choice for the MMR program at UGA. Any questions can be sent to [email protected].

Citation preview

Page 1: Conjoint class project

Sweet Tooth BakerySweet Tooth BakeryCake preference conjoint analysisCake preference conjoint analysis

Michael Bystry, Stacy Comrie, Shannon Goyda, Prochi Jain, Amy KastenMichael Bystry, Stacy Comrie, Shannon Goyda, Prochi Jain, Amy Kasten

Page 2: Conjoint class project

Background/Problem StatementBackground/Problem Statement

Sweet Tooth Bakery wants to create a new cake offering for its customers

The cake will be 9”x13” and sell for $17.99In order to build the most appealing cake, the

bakery needs to understand consumer preference for different cake components

Sweet Tooth Bakery wants to create a new cake offering for its customers

The cake will be 9”x13” and sell for $17.99In order to build the most appealing cake, the

bakery needs to understand consumer preference for different cake components

Page 3: Conjoint class project

Background/Problem StatementBackground/Problem Statement

Cake components:• Cake Flavor

• White, Chocolate, Yellow, Marble• Frosting/Filling Flavor

• Chocolate/Raspberry, Vanilla/Raspberry, Vanilla/Lemon, Cream Cheese/Raspberry, Cream Cheese/Lemon

• Number of Layers• Two or three

Cake components:• Cake Flavor

• White, Chocolate, Yellow, Marble• Frosting/Filling Flavor

• Chocolate/Raspberry, Vanilla/Raspberry, Vanilla/Lemon, Cream Cheese/Raspberry, Cream Cheese/Lemon

• Number of Layers• Two or three

Page 4: Conjoint class project

DesignDesign

Full-profile conjoint◦Minimum design size=9 profiles◦Full factorial design=40 profiles◦Selected Design=20 profiles

D-efficiency=99.5% with only one violation No canonical correlations >.316

◦Two hold-out tasks added to the design Final design size=22 profiles

◦D-efficiency=98.7%◦No canonical correlations >.316

Full-profile conjoint◦Minimum design size=9 profiles◦Full factorial design=40 profiles◦Selected Design=20 profiles

D-efficiency=99.5% with only one violation No canonical correlations >.316

◦Two hold-out tasks added to the design Final design size=22 profiles

◦D-efficiency=98.7%◦No canonical correlations >.316

Page 5: Conjoint class project

Analysis and ResultsAnalysis and Results

44 usable completesValidation procedure

◦Correlations between 20 test profiles and two holdout profiles were examined Seven respondents removed

◦Responses to holdout tasks not consistent with responses to test profiles

37 remaining respondents used for analysis of importance scores and market share

44 usable completesValidation procedure

◦Correlations between 20 test profiles and two holdout profiles were examined Seven respondents removed

◦Responses to holdout tasks not consistent with responses to test profiles

37 remaining respondents used for analysis of importance scores and market share

Page 6: Conjoint class project

Analysis and ResultsAnalysis and Results

Importance and Part-worths◦Average, maximum, and minimum part-worths

calculated for each attribute◦Importance scores calculated for each attribute

Importance and Part-worths◦Average, maximum, and minimum part-worths

calculated for each attribute◦Importance scores calculated for each attribute

Cake Component

HighestAverage Part-Worths

Lowest Average Part-Worths

Importance Scores

Cake White (5.2) Marble (-4.69) 37%

Frosting/fillingChocolate/Raspberry

(11.12)Cream cheese/lemon

(-14.23)55%

Layers Two (0.98) Three (-.098) 8%

Page 7: Conjoint class project

Analysis and ResultsAnalysis and Results

Page 8: Conjoint class project

Share Simulator UsageShare Simulator Usage

Base StateBase State Updated with new selections Updated with new selections

Analysis and ResultsAnalysis and Results

Page 9: Conjoint class project

Analysis and ResultsAnalysis and Results

Customer clusters◦Segmentation based on importance scores as

derived from individuals’ utility functions◦Two-, three-, and four-cluster solutions

examined Three-cluster solution gave best results

Customer clusters◦Segmentation based on importance scores as

derived from individuals’ utility functions◦Two-, three-, and four-cluster solutions

examined Three-cluster solution gave best results

Page 10: Conjoint class project

Analysis and ResultsAnalysis and Results

Segments of cake buyers1. Frosting/filling segment (38%)

High importance given to frosting/filling attribute

2. Cake flavor segment (19%) High importance given to cake flavor

attribute3. Holistic segment (43%)

• Higher than average importance give to layers• Relative equality given to cake flavor and

frosting/filling

Segments of cake buyers1. Frosting/filling segment (38%)

High importance given to frosting/filling attribute

2. Cake flavor segment (19%) High importance given to cake flavor

attribute3. Holistic segment (43%)

• Higher than average importance give to layers• Relative equality given to cake flavor and

frosting/filling

Page 11: Conjoint class project

Analysis and ResultsAnalysis and Results

Average Importance Percentage by ClusterHolistic Cake Frosting

Cake Segment 39.28 67.93 17.87

Layer Importance 11.57 6.38 4.07 Frosting/Filling Importance 49.15 25.69 78.06

Page 12: Conjoint class project

LimitationsLimitations

Due to the nature of conjoint analysis◦No information about purchase intent

Only preference information◦All possible options not included in design

Some cake, filling, and frosting options removed◦No information about possible interactions

Frosting/filling may interact with cake flavor◦The model doesn’t capture pricing information

Conjoint not well suited to capturing price

Due to the nature of conjoint analysis◦No information about purchase intent

Only preference information◦All possible options not included in design

Some cake, filling, and frosting options removed◦No information about possible interactions

Frosting/filling may interact with cake flavor◦The model doesn’t capture pricing information

Conjoint not well suited to capturing price

Page 13: Conjoint class project

LimitationsLimitations

Due to limited time and resources◦Respondents drawn from convenience sample◦Sample size is too small for statistically-

meaningful results Results can not be projected onto the general

population◦Sample size is too small to allow for use of

holdout sample for validation Holdout tasks were used instead

Due to limited time and resources◦Respondents drawn from convenience sample◦Sample size is too small for statistically-

meaningful results Results can not be projected onto the general

population◦Sample size is too small to allow for use of

holdout sample for validation Holdout tasks were used instead

Page 14: Conjoint class project

Questions/Comments?Questions/Comments?