Aster: How to Define a Solid Use Case

Learn Data Science
Teradata Employee

A big issue with any analytic project is deciding on if I should execute on an idea or use case.  Many times we get excited about ideas that really don't have much value to an organization.  Ideas around social media for example:  when was the last time you really did something because a friend of yours liked something?  Now take that to scale, what does it really mean if a million people like a particular product?  Is that really significant?  Maybe if it is a new product by a new manufacturer that is getting BUZZ but if it is a product like hand soap that has been around a while, so what.

The point of this manuscript is to provide you with a tool that will help you decide if a project is a SCIENCE EXPERIMENT or a VALID USE CASE.  What is a SCIENCE PROJECT?  A science project is a very cool project that is interesting from a technical standpoint or from an informational standpoint.  They do not offer real value meaning: cost avoidance, value add(income or profitability), or both.  This is usually the goal for a commercial organization.  It could be rephrased for a public organization:  reduce citizen burden, increase citizen value, both.  SCIENCE PROJECTS are also difficult to make actionable or to make operational.  For example, the use case was able to generate the metrics and customer lists as output, but no one in the business is able to do anything with the scores or the lists because it is too late or not enough time.  We want to try to avoid SCIENCE PROJECTS and that is the point of post.

This article is about defining and scoring a use case for Aster analytics.  I will provide you with a scoring mechanism which will help you rank and score a collection of use cases.  The tool also provides easy and standard questions to ask with respect to the use case and its feasibility.  Lets dive in:


The first part is to give the use case a name and describe the use case in business terms:  for our example we will name our use case: MOBILE CHURN PREVENTION.  A description of this use case: The MOBILE CHURN PREVENTION use case will generate a list of customers with two scores:  1.  Customer 90 Day Churn score and a customer valuation score.  By doing this we hope to retain high value customers and reduce costs and raise revenue.


Question One:  Is this an Aster use case, is this an analytical use case:  Will I be able to leverage analytical operators and scoring methods to generate output, lists, or metrics?  We are going to apply regression, statistical functions, and prediction functions to achieve a list of 90 churn customers.  We also will provide a customer valuation score.

Question Two: What is the Business Value of the Use Case:  What quantifiable costs will I avoid, what profits or revenue will I add as a result.  Can I achieve both?  When will I realize this value?  An example:  we want to produce a 90 day score for customer value and churn risk.  We hope to save 10% of high value churning customer which will reduce costs by $500,000 each month, and increase revenue by 1.5 million per month.

Question Three:  Is the Use Case Actionable:  Is the business involved in the use case definition?  What will they do with the lists and scores when they get them?  Will the outputs be given to the business in a timely manor which will enable them to take action?  How will the business change its people, process, and technology as a result of receiving the scores?  How will they integrate the scores into their systems?  How is the business doing performing churn prevention today if at all?

Question Four:  Is the Data Available:  Where is the data located for the use case?  Is it internal?  Is it external to the company?  Is it consistent and of quality?  How much data is there?  Is it enough data?  How many sources of data do I have?  What political work will I have to do if I have to go beyond departments?  For our churn use case we will require: Customer Data, Billing Information, Web Customer Clickstream, Contact Center information, IVR, and store data.

Question Five:  Who is going to participate in the Use Case:  Do we have management support?  Do we have the technology and the data scientists?  Do we have the technology?  Who from the business is going to work with us on making the churn scores ACTIONABLE?  Without management, business, and IT support this use case will not go very far.


You can use your own approach to scoring the use cases but I like to use a three score key:

( 1)   - Positive or the resources are available with very little work and adaption and is easily repeatable.
( 0)   - Neutral or the resources are not obviously available and will  require more exploration.
(-1)  -  Negative or the resources are not available and will require significant work or is not repeatable.

The SUM of the five scores taken as an average allows you to take a mathematical approach to scoring and ranking your use case.  Obviously the closer to 1 your use case is the more VALID that use is.  As the score approaches zero or is negative the greater the likelihood that the use case is a SCIENCE PROJECT and should be avoided.

Here is an example of a spreadsheet tool I use today: