Cluster analysis allows us to effectively segment our target market into clusters. K means clustering we specify how many clusters until everyone is apart of a cluster. For hierachical we build up everyone is in their own cluster and we build up the clusters and we stop when there is no information loss. We utilize a dendogram to assess information loss. Intuitively information loss reflects upon the fact that everyone is special in their own way. When we group two people together we lose some individuality which is information loss as we ignore the individual differences. As marketers we target by groups so therefore we must group and utilize a dendogram to get a sense of when we push the tradeoff too far. We cluster people on bases variables focusing on preferences not what we look like.
DIscriminant analysis describes people on the outside (e.g. hey Wendy can’t see your preferences but I can target you because I see what you do.
Classification analysis is when we did a survey on our attitudes so we can cluster based on that. We segment based on part worths and describe based on variables. We do cluster and discriminant variables which provides a mapping and we then have new descriptor variables (i.e. new market) which we then run through the model we have already created. We go to different market and utilize the model we have created from our previous survey of descriptor variables and map people to different clusters.
*should be able to interpret cluster means, assignment, dendograms, discriminant function.
Logit Choice analysis is a general tool where we model and predict choice based on their past behavior. They made a choice in the past which we analyze and use to our advantage. We use the logit rule for scoring and targeting but we also use RFM (poor because it only uses a few variables), scoring for targeting, lift charts.
Must know output such as coefficients, elasticities, and choice probabilities (we will be given)
Bass model estimates parameters, adoption process, innovation and imitation effects. This is for durables (e.g. this is for cars where we buy once, sodas)
P – coefficient of innovation
Q – coefficient of imitation (multiplies the ratio of people in the market that have adopted the product). Therefore P acts individually intrinsically. We can either use historical data or look at similar products that have been adopted in the past.
Chain ratio method – narrowing down to a specific market. The chain ratio works well with the bass model because in the bass model we need market potential which is from the chain ratio method.
Understand the applications (entry decisions, pricing, product design
Conjoint is good because it is forcing the respondent to make trade offs. Part worths are attributes, in a regression they are coefficients. Partworths are coefficients in a utility function for example. We run a regression
Types of conjoint
Rankings based We use linear regression
Ratings based – We use linear regression
Choice based We use logit regression
Price elasticity of the ln regression is Beta (B)
Price discrimination – don’t need to know first VS second VS third but would impress him
Market demand model (log log regression, logit model, Gabor Granger) – need to collect data (when we don’t have historical data)
Likelihood of purchase scale – we convert 1,2,3,4,5 to map these numbers to probabilities of buying, it relies heavily on the assumption. We assign the probabilities based on historical data. The analyst makes the likelihood of purchase assumption and who we survey for the past data can directly impact potential problems in the assumptions.
Use with costs – can use the model for marginal costs.
Use for price differentiation – segment on demographic variables such as age
Online marketing analytics
Targeting (Bookbinders) – found optimal point to target
Keyword bidding (correlations, regression for CTR)
Tools are valuable if we know
Where the #s come from
Believe the numbers?
Tools aid decision making but don’t dictate decisions
Advisory not prescriptive
Must use judgment to interpret