Making real-time offers from big data customer insights

Author: Chia-Yui LEE, TIBCO Data Science

In the previous post of this series, we?ve discussed the challenge of performing customer analytics with big data - you cannot load every single record into an analysis all at once. We then introduced TIBCO?s accelerator for Apache Spark and explained how it helps you harness every pertinent piece of big data information to better know your customers. We also briefly discussed applying machine learning techniques on big data sources to find products your customers are likely to buy. Now that you have your customer insights and know whom to target in a campaign and with which product, how do you put this into action? In this post we will look at how you can make real-time offers with big data when a customer visits your website, interacts with your mobile app, shop in your retail store or calls your service center.

We shall use propensity analysis as an example again. With the accelerator, the best offers for a customer are determined by applying predictive algorithms to historical data on a big data platform. The algorithms learn from the data to predict if a customer is likely to make a purchase from a product category. The relationship between factors influencing customers? action and the actions is represented in what we call a model.

When performing a propensity analysis, an analyst may find several of these relationships for a particular category and has to find the best one. Using TIBCO Spotfire^®, a component in the accelerator, the analyst can compare the models using a model quality measure and select the best to use. The accelerator supports a gated process of model approval and revocation such that only approved models can be used in marketing campaigns.

When running a campaign involving several product categories, an analyst would construct the campaign by adding models (those that have been approved) for these categories. The models selected for the campaign are deployed and stored on the big data platform.

The activities we have described so far are about analyzing historical data on the big data platform and learning from data about the past to predict the future. We have deployed the models as part of a campaign and we want them to predict what the customers are likely to buy and make the offer to customers. This is where the rubber hits the road. It is time to use the insights to make decisions and act on them. The accelerator achieves this through another of its component, StreamBase. It is an event-processing environment set up to receive real-time inputs from customer interactions (such as purchase transactions), retrieve historical data from the big data platform and send this information to the models to determine the best offers to make, in real-time.

With real-time offers being made, it may seem like we have reached the end of our story. This, however, is not the case. What is the use of making predictions and creating offers without any way to track and monitor them? With the accelerator?s streaming analytics component, the offers can be monitored in real-time along with the customers? response. This real-time data is stored on the big data platform. Once that is done, the data becomes part of the historical data.

Customer behavior changes from time to time. That means the models used to generate the product offers may become obsolete and deteriorate in quality. With the accelerator, analysts may conduct the propensity analysis again at regular intervals. Each time, they would have the algorithms learn from a refreshed set of data that includes customer interactions in the more recent past. In this way, they can improve on existing models, retire obsolete ones and keep the models in their campaigns up to date. In other words, they go into a cycle of continuous improvement.

(This post is the sixth of a series of posts on the Customer Analytics Templates for TIBCO Spotfire^®)

Previous blog posts in the series:

Know Your Customers with Customer Analytics Templates for Spotfire^®

Segmentation Analysis with Customer Analytics Templates for Spotfire^®

Propensity Analysis with Customer Analytics Templates for Spotfire^®

Affinity Analysis with Customer Analytics Templates for Spotfire^®

Customer Analytics with Big Data