Technology

We are often asked how we are able to make such accurate predictions of a customer’s future buying behavior. Without delving into the underlying algorithms, here is a concise description of the processes we use and the unique tools we have developed.

Methodology

We start with transaction data—who bought what, when, and for how much. Transaction data is more predictive of future buying behavior than demographic data, survey data, or web browsing data. Further, every company already has this data for every one of its customers. It is a goldmine you already own.

The customer transaction data is fed into our Analytic Engine, a set of algorithms we have built, honed, and tested for over a decade. The Analytic Engine creates a model for each customer. Each model holds a record of all the transactions for the associated customer and predicts what that customer will buy next. This approach, with an individual model for each individual customer, is what lets us make predictions at the customer account level. It is a completely different approach than traditional methods which typically analyze a group of [hopefully] similar customers, a segment, with the assumption that each segment member (a customer) will behave in the same way.

The output from the Analytic Engine is a set of data tables. One table holds all the scoring metrics, with one record per customer and as many fields as needed to capture the scores for each customer. Another table stores the top up-sell and cross-sell purchase propensities—which products or services we expect the customer to purchase. Sample output is shown in Figure 1.

Sample output from the Analytic Engine

Figure 1. Sample output from the Analytic Engine, click image to enlarge.

 

Finally, these data tables are used by Longbow, our campaign builder and reporting tool. Longbow helps customers put together individualized direct marketing campaigns that target customers ready to buy (or close to defection!) and that offer these customers products we predict they will find attractive. Longbow tracks the campaigns and reports on the results. The complete process is diagrammed in Figure 2.

The Loyalty Builders Process

Figure 2. The Loyalty Builders Process, click image to enlarge.

 

Architecture

The Loyalty Builders Technology Architecture, Figure 3, is a different way to look at this process. At the bottom of the stack are the fundamental tools. Your data is stored in a highly secure MS SQL Server database. Code for the Analytic Engine is written in C#, Transact SQL and MatLab. The next layer up is the transaction data stored in the database, which in turn is operated on by the Longbow Analytic Engine. The engineers on our data teams work at this level.

Loyalty Builders Architecture

Figure 3. Loyalty Builders Architecture, click image to enlarge.

 

The Analytic Engine output, the scores and purchase propensities, are the input for the Longbow Campaign Builder and Report Tool, software delivered over the Internet as SaaS—Software as a Service. Our customers work at this level to build their direct marketing campaigns and to view the reports produced by Longbow.

Also see The Mathematics of Customer Loyalty.