Demystifying predictive analytics: design your analysis

This is part 3 in a series designed to show your organization how to create a customer-centric organization with predictive analytics. Get started by analyzing the status quo and evaluating your infrastructure.

Saken Kulkarni

Saken Kulkarni

Before entering into a career in Big Data and analytics, I was interested in becoming a Foreign Service Officer. I was fascinated by ancient history, international development, and War Games. As a college student, I remember that the most interesting classes in the course registrar required some less-than-intriguing prerequisite courses. I tried to get around taking these classes, to no avail. But looking back, it’s a good thing that my university enforced these prerequisite courses. Developing a well-rounded knowledge base enables you to think contextually and critically to drive new ideas forward.

Just like in higher education, your organization must pass its predictive analytics prerequisite classes (analyzing the status quo and evaluating infrastructure) before it reaches the most exciting part of the program: analysis design.

But designing your customer-centric advanced analysis can be daunting. In my experience, retailers understand that they need to obtain a multidimensional view of their customers, but don’t know where to start. To help you get started, I’ve gathered five high-impact customer-related analyses. These analyses look at customers through three different lenses:

Analysis WheelLet’s go through each of these in detail.

Market basket analysis

If you log into Amazon.com or explore movies on Netflix, you may see a section that suggests products that you like based on products that you have browsed or bought in the past. This is market basket analysis at work.

The purpose of market basket analysis is to answer a simple question: what items are most likely bought together? Statistical software, like SAS, can conduct type of analysis by placing items together in a “basket” based on purchase patterns. Analyzing this behavior can help retailers optimize promotions, store displays, and web advertisements to maximize upsell and cross-sell potential.

Share of Wallet

This analysis helps retailers understand the percentage of a customer’s business that they get from that customer. Share of Wallet (SOW) analytics helps retailers understand customer loyalty as well as opportunities to deepen relationships with unsaturated customers.

In its most simple form, SOW can be calculated using the following formula:

Share of Wallet CalculationWhile SOW analysis can range from 0%-100%, it is very rare for retailers to have 100% of a customer’s “wallet” due to the vast number of purchasing options on the market. Data from SOW calculations come from two sources: customer share (the numerator) from internal transactional data, and customer wallet (the denominator) from survey data that augments data from your warehouse.

Customer Lifetime Value

In customer analytics, Customer Lifetime Value (CLV) calculations predict the profitability of a customer over the course of time. Mark Jeffrey, Director of Technology Initiatives at the Kellogg School of Management, elegantly defines CLV in his book, Data-Driven Marketing: The 15 Metrics Everyone in Marketing Should Know:

Customer Lifetime Value calculationSegment CLV calculates your preferred segmentation criteria (initially, even quartiles will do) and bucket customers into “Very High,” “High,” “Medium,” and “Very Low” CLVs to better understand customer loyalty.

This type of analysis is critical for retailers. Competition and discounting have driven retailers to razor-thin margins. Understanding and targeting your most loyal customers will help you create more personal and authentic relationships to drive higher margins. There will be no need to unnecessarily mark down prices if you can create strong relationships with your customer evangelists.

Customer churn

Customer churn quantifies the percentage of customers who stop purchasing a product or service during a given time period. Calculating customer churn can help you identify segments, sales territories, or products when customers are leaving in droves. Most important, combining churn analysis with CLV analysis can help you identify high-value customers with high churn rates, and design loyalty campaigns to reactivate those profitable relationships.

For the sake of simplicity, a generic equation can be used:

Customer Churn calculationSentiment analysis

Also known as text analytics, sentiment analysis tools mine websites, online product reviews, call center logs, blogs, and social media to better understand what customers are saying about your products or services. In its simplest form, unstructured text data can be categorized as “positive” or “negative” through natural language processing algorithms. The explosion of social media data provides another rich data source. Retailers can mine public social media sites and identify what customers are saying about their products. If their systems are sophisticated enough, they can then match those individual customers with the customers in their data warehouse or data mart to obtain a richer view of their customer.

As a retailer, you might think it’s difficult to conduct these analyses using only the data you collect from source systems, like your Point-of-Sale (POS) system. And you’d be absolutely correct. Luckily, there is a way to augment your current data sources with external sources. This type of data enrichment provides demographic, psychographic, ethnographic, and social media data and is gathered by third-party data source providers. The use of data augmentation can separate retailers who lead from those who lag behind.

 Data Sources Homework is done—let the analysis begin

Using these analyses can help your organization get started on the journey toward customer centricity. It’s important to note that while these analyses are part of your analytics “toolbox,” they are by no means comprehensive. You can always explore options with your data science team to create a more robust approach. Most important, be patient with the process—there is no need to stuff poor quality data into a model or report shoddy results just to show advanced analytics prowess.

Stay tuned for my next post, which will discuss how to create data-driven tools through visualization and discovery, creating the ultimate user experience for the business user.

About sakenkul
Saken is a Business Analytics consultant at Slalom Consulting. He focuses on the intersection of customer engagement, analytics, and data visualization

One Response to Demystifying predictive analytics: design your analysis

  1. Pingback: Demystifying predictive analytics: train and drive adoption | The Slalom Blog

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: