Use the SWAT Framework (Strategic questions, Wrangle data, Analyze, and Take action) This will be "business led analytics" where the questions are developed first. Then sponsors think about how they will Take action and the value of the intelligence is tracked. Finally the Wrangling of data and Analysis is done. Because its intelligence, it is fast! It needs to be, customers and ships can't wait...
The team structure enables the right delivery. Since intelligence is always "optional" so you need a rapid response business team under the Business Answer Lead and a enterprise operational team with the Data Lead in charge who enables standardization. Business and IT need to work very carefully together. Too often, "data scientists" are hoisted with the entire job, but they lack business acumen/credibility, and don't enjoy data pipelining, so caught in between the leave. An entire organization has the right structure. This structure has been used by the best intelligence operators in the world. Verses the pure data scientist teams we are seeing today which were suggested by vendors/consultants with less stake in sustained business value as an outcome.
"Sometimes we need to see to feel."
A SanKey diagram shows the flows of data from one state to another. Customers, represented by data can be shown how they move from webpage to webpage or channel to products, the best outcomes of increased engagement or worst outcomes of drop or lost customer.
Here is a good list of the talent and resources you need to create and sustain your actionable intelligence journey.
Purists will say machine learning starts with code and equations. Business leaders need to move fast and typically are highly visual. Rapidminer brings these camps together. For fast POCs(Proof of Concepts) Rapidminer lets you drag and drop Gradient Boosted Trees (a more robust and highly descriptive M/L) and it lets the programmers go behind the scenes to implement R or Python.
Try it out! If you are in my class... its free! (Click Here)
Using Neural Networks, computers can "see", match complex patterns. As kids, we learn, we are handed shapes and are taught their name. Computers are being taught the same way.
This picture by Professor Andrew Ng, says it all. The colors on the door handle are simply numbers to the computer. Learning is hard, but with math, and decent processing power, its now possible.
Play around with Neural Net Vision here:
Each cloud platform has its strengths. In these "early days" try them all! Each has a free trial, so you have everything to gain.
Check out this tensorflow playground! Great way to visualize how neural networks handle classification problems.
Keith B. Carter