Turning Data into Customer Insight

Intel Corporation, outlines how one company has used predictive analytics – a type of advanced analytics – to better understand and serve its customers.

 

The data deluge of recent years has created a host of challenges for companies across all industries. It’s an ongoing headache for many to collect, store and secure data in multiple formats and from an increasingly diverse range of sources – but doing it is no longer optional. Unlock the value in all this new data and you can boost business performance, customer loyalty, and revenues. The key is in going beyond traditional, backward-looking business intelligence (BI) practices, and adopting advanced analytics practices that can help you make smarter, faster business decisions.

That’s the theory, at least, but let’s take a look at one company that has put advanced analytics into practice. We’ll look at how it used predictive analytics to help it understand and engage with its customers on a whole new level.

Building an analytics algorithm

A French household electronics retailer had a lot of data about its customers – what they had bought in the past when they’d made purchases, when they’d requested services and maintenance and so on. However, it lacked insight into what all these existing customers might do next. Specifically, it wanted to identify the top 500 customer households that were most likely to purchase appliances within the next 12 months, and which product lines they were most likely to buy from.

In order to gain this insight, the company decided to use the Cloudera distribution of Hadoop (CDH) and model codes developed by Intel to process large amounts of raw customer data from nine different sources. This data was merged and aggregated to create 106 variables covering basic customer information, household information and segmentation, previous purchase behavior, previous service and maintenance requests and events triggered by them, and government demographic information.

It then applied a random forest algorithm to each segment of customer data separately to determine which had the strongest impact on their purchasing habits. Each customer was allocated a buy/no buy flag, indicating whether each variable had a positive impact on their likelihood to make a purchase, and which product line they would purchase from.

The customer households were then ordered based on their probability of purchase, from highest to lowest. The company tested the accuracy of the model by taking customer data from 2005 and using it to predict purchasing behavior and then comparing it against actual purchase data for the period. The predictions had a 68 percent accuracy rate.

Insight drives customer engagement

With this predictive analytics model in place, the retailer now has a reliable method of predicting customer behavior, which can be adapted to take into account different variables as needed. With this additional insight, it can offer much more timely and targeted communications to its most valuable customers, increasing loyalty and boosting revenue opportunities. For example, if its data-driven predictions indicate that a particular customer is likely to purchase a dishwasher soon, the company can proactively send the customer an email with details of the latest models and features, encouraging the customer to come back to the company to make the purchase, and possibly to buy a slightly better model than before. This type of customer-centric approach is a focus for all sorts of organizations – from healthcare to financial services.

Take a look at Intel’s ‘Getting Started with Analytics’ Planning Guide to learn more about how you can start to use advanced analytics to achieve these sorts of results within your own business.

Predictive Analytics Model
Predictive Analytics Model (Click image to enlarge