Week 3: Product management I

goals: predicting consumer's choices and market shares of existing products

Concept of attraction or utility

total benefits/ satisfaction customers derived from the products

increasing attraction => increases purchase probability and hence the market share

market share prediction

measure brand attractions => estimate purchase probabilities => predict market shares

  1. Measure brand attractions
  1. Multi- attributable framework (from survey data) - page 6
  1. Conjoint analysis
  1. Evaluate a product as a bundles of attributes
  1. Different evaluations to different attributes
  1. Form beliefs about relative strengths/ attractiveness of different products

attributes importance (brand name, price, ...) should sum to 1 or 100% (obtained using survey)

example: p.8

attraction = weighted sum attribute strengths of that brand across all attributes

  1. Estimate purchase probabilities

when predicting market shares from purchase probabilities, we have assumed that all consumers/ segments buy the same quantity

equal size segments (example: p.10)

unequal sizes segments (example: p.11)

estimated in several ways

  1. Share of attraction: brand's purchase probability = share of total attraction (p.14), low involvement, frequently purchased products
  1. Maximum attraction: consumers choose the brand with the highest attraction (p.13), suitable for high involvement, infrequent purchases

3. Multinomial Logit Model

1 and 2 are simplistic approaches => low predictive power

Multinomial Logit Model

Logit model assumes that choice probabilities as well as utilities are stochastic (have a random component)

example: p.16

probability of choosing a product does not increase linearly with utilities, the relationship is S shape

3 properties (p.17)

2.Invariant property (p.18): only differences in attractions affect choices, adding/ subtracting a constant to all attractions does not change purchase probabilities

predicting market share: example p.20

Limitations (p.21)

Table p.22

Logit choice model

similar to linear regression model

uses discrete variable (for dependent variable) while linear regression model uses continuous variable

formula: p.24

differ from predicting choices (p.25)

choice model illustration (p.26)

predicting a binary outcome (p.28)

  1. Probabilities are never 0
  1. S shape relationship between purchase probabilities and attraction

known the purchase probabilities but don't know the utility