This is a hands-on, practical full-day software training covering the depth of features and settings available within Sawtooth Software’s choice simulator, also known as the market simulator. That’s why we call this an A-Z training!
Come learn how market simulations can tackle business questions and find near-optimal solutions (e.g., pricing, product configurations, targeting specific segments) with realism and specificity to the competitive effects in the marketplace as well as the variety of consumer preferences. Conducting market simulations is the important next step beyond just computing and summarizing utility scores that you need to extract the most value from your conjoint analysis studies.
Sawtooth Software’s desktop choice simulator is an integrated feature of the (Windows-based) Lighthouse Studio platform, and also is available for analysts or clients in the (Windows-based) standalone package. Among the many things we’ll cover in this workshop, we’ll discuss:
The pros/cons of the different market simulation models (first choice, share of preference, randomized first choice)
IIA (the “red-bus/blue-bus” issue) and how to reduce its ill effects
Setting up segmentation variables and weights
Interpolation and extrapolation
Reporting options, including shares of preference, revenues, profits, and costs
Adjusting the scale factor (exponent—flatness or steepness of the simulated shares of preference)
Adjusting the model to account for differential product distribution and also awareness
Calibrating the market simulator to predict actual market shares (and the challenges involved)
Product optimization searches, including multi-objective genetic algorithms (MOGA)
Exporting/creating client simulators in Excel
Solve for share function (to find the price that achieves a target share of preference result for the client’s product offering)
Willingness to Pay (WTP) functions in the simulator
And much more!
Segmentation can be the most frustrating kind of study marketing researchers face. Much of the difficulty is inherent in the complexity of what we’re trying to do with segmentation. But some is difficulty we make for ourselves by not having a clear process that takes into account the foreseeable complexities or by using analysis methods that reduce our chance of finding a good segmentation solution.
This session begins with a general description of what segmentation seeks to accomplish and a general process that textbooks advise. We’ll then cover a number of hard truths about the segmentation enterprise that help explain why so many segmentation studies fail. It’s largely because people don’t understand some of these hard truths that they don’t use robust enough segmentation processes and they open themselves up to failure. In addition, we’ll identify a number of smaller pitfalls that a successful segmentation process will avoid.
Next, we’ll show a process that takes the hard truths above into account and that also avoids the smaller pitfalls. We’ll start with a new way to conceptualize different kinds of segmentation, because some traditional methods are mis-aligned with the goals of some types of segmentation (and some are mis-aligned for any use in segmentation whatsoever). We’ll describe some newer methods, many borrowed from the machine learning literature, that can help us navigate the hard truths and the pitfalls, to increase our chances of successful segmentation. We’ll also provide advice about statistical software packages for running segmentation and we’ll provide R code for running the analyses we describe.
The bulk of the time we’ll spend demonstrating different kinds of analysis using different kinds of software:
Variable selection using R software to run
Unsupervised random forests
An automatic variables selection method for metric data called ClustVarSel
An automatic variable selection method for mixed scale data called VarSelLCM
Supervised random forest for predictive segmentation
Segment generation
Needs-based segmentation: latent class MNL in Sawtooth Software
Supervised segmentation: Tree-based methods in R
Unsupervised segmentation
Cluster ensemble analysis in Sawtooth Software’s CCEA package
Model-based clustering using Latent Gold software
Typing tool creation
MaxDiff typing tool for needs-based segmentation
Discriminant analysis in SPSS
MNL in SPSS or R
Tree-based classification in R
Supervised random forest for segment assignment
SVM for segment assignment
Finally, we’ll end with a decision flow chart showing the decision points and options to guide you through our recommended segmentation process.
