Prescriptive Analytics Workshop

Dave Wentzel
LinkedIn Profile
Spark - 20220218

Abstract

As analytics practitioners we tend to be enamored with "the tech". We love to talk about data lakes, MPPs, ETL, and ML algorithms. But none of this is particularly interesting to a project sponsor or executive. They are interested in business outcomes.

We build dashboards that forecast sales, but is that adding value?

we build ML algos to predict which customers will likely churn...but is that adding value?

Initially, most executives will say dashboards and predictive algorithms add value, but if you probe you'll likely hear something like this:

"The dashboard gives me a bunch of facts but doesn't tell me a story and more importantly what do I do next?" "Thanks for telling me which customers will churn, what do I do about it?"

Now imagine if we didn't talk about the tech at all, but just demo'd how to solve a really thorny analytics problem using simple tech that mere mortals can understand, with a repeatble set of patterns (People Process Technology). Suddenly data projects are viewed as less risky.

In this session we'll look at a few interesting use cases to understand how Prescriptive Analytics can help you compellingly demonstrate how an organization can be more data literate and insights-driven.

Demos

1. Prescriptive Analytics requires a simple analytics pattern. CETAS in Synapse works real well. I have 2 versions of this demo. Both do the same thing, one is simply a notebook where I can show better documentation and plots:
   • Demo SQL Script: import this to your Synapse workspace.
   • SQL Serverless Notebook
     • this uses a vscode devcontainer to ensure everything is setup perfectly. If you know how to use these then it should be easy.
     • I use this devcontainer "pattern" along with the imports and utils files in the scripts folder to create a totally reproducible "analytics sandbox" that can be used by anyone without dealing with installing python and configuring environments.
     • This works awesome for hands-on, interactive demos. I have a version of this that runs in the browser and is launched via ACI that will allow MTC session attendees to play with the code and adjust the parameters WITHOUT needing access to Azure, vscode, python, or Synapse. Let me know if this is interesting and maybe we can collaborate on it.
2. Avoiding Cognitive Mistakes. Prescriptive Analytics is primarily about answering the question what do we do next?. This requires collaboration and critical thinking skills. You don't want to make cognitive mistakes. Here's a case where we need to talk through the business problem so we tell management the RIGHT decision to make
   • we use Synapse Spark for this example, you will need to import the notebook into your Synapse workspace
3. Marketing Campaign Analytics. Let's use some data and critical thinking to recommend what should we do next?
   • Demo Notebook: load this up into your Synapse workspace and execute.
4. A primer on Prescriptive Analytics and Linear Programming...according to Dave. Prescriptive Analytics is an old field, but the term Prescriptive Analytics still means different things to different people. Some view it as using math and linear algebra (programming) to solve problems. I must say, Linear Programming is really easy to do and is VERY cool to demo.
   • Prescriptive Analytics Notebook: It's easier to run this in my devcontainer as a Jupyter notebook vs in Synapse Workspace. But it's easily adaptable. Honestly, start here if you have no clue what Prescriptive Analytics even is. This is my overview

WIP

• recommendation engine/reinforcement learning : wip