Blog

Top 3 Objectives Before Starting an Analytics Project

Gerhard Pilcher

October 26, 2018

BLOG_Top 3 Objectives Before Starting an Analytics ProjectUnderstanding the organization’s business objectives and requirements, converting this knowledge into a definition of a problem, and developing a preliminary plan to solve that problem is crucial to the successful application of analytics and machine learning. In order to construct a successful model, the data scientist must understand how the business functions and how it will use the data. Even the most technologically advanced analytics model will produce trivial and possibly misleading results if it is disconnected from the purposes and goals of the business.

Define The Business Impact

An important aspect before starting any analytics project is defining how the model will impact the workflow and decision-making processes of the business. To achieve maximum return on the investment in analytics, the results of the initiative must lead to advantageous changes in business operations.

Different constituencies within an organization typically want different information from a model. For example, the marketing group might want to know which products and geographic areas merit more promotional support; the customer service department might want to know which customers represent the highest risk for attrition; the financial department may be interested in knowing which customers are credit risks; and individual sales reps might want to know which customers have the biggest potential for
future purchases.

Connect Business Objectives to The Data

Connecting the business objectives to the input data usually requires considerable time and effort. The data scientist must interview subject-matter experts within the company to determine the objectives of the project, the resources available (people, data, and technology), the precise definitions of terms to be used, the projected costs and benefits of the project, and the projected return on investment (ROI).

Many companies can’t understand why so much time must be spent on understanding the business. They seem to think the data scientist should just be able to take their data, put it in a magic “analytical model” box, turn the crank, and churn out the desired information. We call this the “black box fallacy.” A data scientist who tries to build an analytic model without consulting SMEs is like a lawyer who tries to handle a case without consulting the client.

Clarify Your Objective

Take sufficient time and care to clarify the purpose of your data analytics initiative. If you don’t get your objective right, your whole project can be a waste of time and money. A team of our data scientists recently spent more than eight hours on the phone with several business and technical people who work at the home office of one of our clients. This provider of outpatient medical services needed to predict the rate at which subscribers were likely to be referred out of their network.

It took us a full day on the phone to understand their answers to a number of questions, and that was only one conversation among many. There’s just no shortcut around this phase of the process. A model built on a poor understanding of the business will likely deliver disappointing results.

Unfortunately, against our recommendation, this client rushed into the data understanding phase without fully completing the business understanding phase. Months later, during the model validation phase, the project team struggled to arrive at a stopping point for model accuracy. When one member realized that they had shortcut the business understanding phase, the team went back and defined how the model would be put into use by the business. It then became clear to them that they had wasted considerable time refining the model, because the extra accuracy obtained in the recent months of modeling was not germane to the ultimate business use of the model. The client team learned a hard lesson, but one that proved valuable on subsequent projects.


COVER-eBook_Top-10-Data-Mining-Mistakes-1-1

Want to explore other common mistakes when applying advanced analytics? Download our eBook The Top 10 Data Mining Mistakes

 


Related

Download the Mining Your Own Business EBook to preview Chapter 3 - Leading a Data Analytics Initiative

Read the blog  Jumpstart Your Data Science Team with Experts

Read the blog What is Data Wrangling and Why Does it Take So Long


About the Author

Gerhard Pilcher Chief Executive Officer Gerhard Pilcher enjoys predictive analytics and data mining, especially related to the areas of Fraud Detection, Financial Risk Management, and Health Care outcomes using various analytical methods, working with people, leading change, and timely management of complex projects. His work experience spans both private and government sectors including international experience. Gerhard teaches at Georgetown University as an adjunct faculty member in the Math and Statistics Masters degree program. He also is an instructor for the three day SAS Business Knowledge Series course "Data Mining: Principles and Best Practices" and been invited to teach at international conferences. Gerhard currently serves on the Institute for Advanced Analytics Advisory Board and George Washington University Masters in Science in Business Analytics Advisory Board. Gerhard has extensive industry experience in government oversight, financial, construction and telecommunication industries both as a business owner and executive.