|Modules||Core Business Issue
|1||Intro to Data Science, ML, AI and DL (Deep Learning)||1|
|2||Innovation, Positioning and Data Visualization||2-3|
|3||Predictive Analytics, Machine Learning (ML) and AI||3-4|
|4||Data Ecosystem, Enhancement, Scoring, and Imputation||6-7|
|5||Product Design, Product Line Optimization||7-8|
|6||New Customer Capture||9-10|
|7||Customer Experience (CeX), Satisfaction and Loyalty||11-12|
In the first week we will introduce the core integrative concepts of applied data science - which are drawn from mathematics/stats, computer science and business respectively. Next, we will define the data science landscape, its competitors, assets, approaches, and current generation challenges. In this step, we will orient participants to their laptops, the software and platforms used in the bootcamp and the seven discrete modules that follow. Last, we will lay out our own "rules of data science engagement" (rather akin to the rules of the fictional NCIS agent Jethro Gibbs). And for the data science scholars among you, there will be plenty of bibliographies for optional reading.
Warning! It's a jumble of acronyms out there! While much of the action in the data ecosystem seems to be about harnessing distributed data via Hadoop and its ken (and we WILL cover that content), we will also carefully explore whether corporations have failed to empirically develop and validate an ontological / philosophical model of knowledge discovery. We will ask the fundamental, tough questions: "WHAT should be collected," and "HOW can we do a better job of empirically and verificationally capture how customers (and segments of customers) actually THINK and BEHAVE?"
Data visualization that drives insight is far more than simple charts and graphs. It requires using the uniquely honed human perceptual capabilities (i.e., our capability to notice fine variations in color, shape, size, and motion) and encode data in those ways. To do so, this module will use a half dozen types of multivariate data reduction (from GLM to SOM to MDS, to various kinds of scaling, to T-SNE) and will also explore how decision support platforms can be built on top of this "math." Quite importantly, we will learn how to visualize demand in the product space.
Ah yes. This is the work horse. The "home base" where most of the money in data applied science is made and wasted. We will cover the usual suspects here (supervised and unsupervised learning, ensemble methods, baggers, boosters, various flavors of validation, etc.), but we will probe into other timely topics. Might we get far further and far faster, for instance, if we build ML platforms that automatically vary input variables, layers, and transformations, rather than relying on machine learning optimization algorithms alone? We will also ask whether (and how) you can do it without big box players because virtually all of differentiating data science lies in the public domain.
When it comes down to it, people buy products they like. Econometricians call this utility theory, and psychometricians term these measures of preference "part-worths." Regardless of what it is called, the goal of this two-week module is to quickly spin us up on the optimal approaches (designs) to utility estimation, simulation modeling and demand prediction. This will position us to build decision support platforms -- because as another fictional character (Terrance Mann) said, "If you build it they will come." Last, since the corporate bottom line derives from all your products and product lines we will delve into the data science of product line optimization.
Ahh, here hangs a tale. New Customers are the enigma of corporations. After all, how do we capture the very people that, by definition, we know the very least about? There are two ways to do this, actually. First, one develops a multivariate demand space built up from the utilities of the market. And yes, that requires designed experiments. (It may be heresy to say so, while boatloads of
social data can help, nothing replaces designed experiments. As such, NCC draws from utility theory and integrative experimental design.) Second, we will visualize and move toward opportunity in a multivariate reduced space. This uncovers "hotspots of demand" by using a number of strategies (expectation), a variation on a transform of preference, etc. The endgame is both a predictive AND and a visual decision support system.
Way back when Millennials were just a gleam in their parents' eyes, there was Deming and the House of quality (HOQ). The HOQ worked. Mostly. It worked because it captured something intrinsically true about how people behave, namely, when you perform better, people will buy from you again. The problem? Many decision makers attempted to base product development decisions on a deeply impoverished measurement model (it only had performance and satisfaction.) Fast forward to today. What happens if we add preference, expectation, loyalty, habituation and some other sundries? Some pretty cool outcomes, actually. The result? A approach to the customer that captures, retains and grows the customer relationship. As before, we will end up with a decisioning approach and system. Why? Because that is the endgame of all ML-fronted APPLIED data science.
Fortunes have been made and lost on this simple concept. The question isn't as much "How do we segment the marketplace?" After all, iterative, multi-algorithmic classification algorithms from all parts of the quantitative compass exist (flavors of clustering, Q factor analysis, Bayesian and non-bayesian classifiers, different brands of neural nets, etc.). The question, rather, boils down to what those measurement inputs actually yield - and how we can use them for competitive advantage. Here again, we will borrow insights from a deeper data ecosystem (discussed in the module four: the data ecosystem). Because, after all, a world-class data ecosystem combined with a comprehensive and rationalized classification ontology is the key to ultra-high performance market segmentation.