Predicting customer behaviour

KHALID SHEIKH continues his series on Customer Relationship Management

ONE of the most important functions performed by a CRM system is to help predict customer behaviour. Customer behaviour prediction as implemented in most CRM systems consists of the following activities:

• Capturing all relevant customer information.
• Customer behaviour modelling.
• Customer value assessment.
• Capturing relevant customer information.

In the past, information related to a single customer was distributed across the entire company within its different departments. This information had to be harmonised from both the business and technical point of view. Today, customers can interact with a company in a number of ways—telephone, fax, e-mail, EDI, Web, etc. Hence, analytical solutions must include the ability to flexibly and consistently integrate all data from a variety of customer interactions across all touchpoints into a consolidated view of the customer. This consolidated view, which should be available to every employee who needs to interact with the customer, should include information from external sources, interaction information captured at different touchpoints, and information from the back-office systems (ERP/SCM) of the company. CRM solutions integrate information from multiple sources to create a consolidated customer view and then make this customer knowledge base available as source data for the numerous CRM analytical applications. The analytical process can then generate new insights on customer behaviour with every new customer interaction.

Customer behaviour modelling

Customer behaviour modelling involves the following three steps:

• Observe customer behaviour as depicted by the consolidated customer view contained in the customer knowledge base.
• Identify relevant behavioural patterns from the observed customer behaviour using profiling and scoring techniques.
• Create predictive models that can be used to acquire, grow and retain attractive and profitable customers.

Analytical techniques for customer behaviour modelling

Association

Association techniques identify affinities among the collection as reflected in the examined records. These affinities are often expressed as rules. For example, 60 percent of records that contain item A also contain B. The percentage of occurrences (60) is the confidence factor of the association. The association technique is often applied to market basket analysis, where it uses point-of-sales transaction data to identify product affinities.

Clustering/Segmentation

Clustering is the method by which like records are grouped together. Usually this is done to give the end-user a high-level view of what is going on in the database. This technique segments records in a database into subsets (or clusters) based on a set of attributes. One way of using it is: In the process of understanding its customer base, an organisation may attempt to segment the known population to discover clusters of potential customers based on attributes never before used for this sort of analysis. (For example, the kind of school they attended, or the number of vacations per year.) Clusters can be created either statistically or by using artificial intelligence methods, and can be analysed automatically by a programme or by using visualisation techniques.

Scoring

Data mining software creates a predictive model based on historical data. This model can then be applied to new data in order to make predictions about unseen behaviour.

* Scoring: The process of using a predictive model to make predictions about behaviour is called scoring.

* Score: The output of the model—the prediction—is called a score. Scores can take just about any form, from numbers to strings to entire data structures, but the most common scores are numbers. For example, the probability of a customer segment responding to a particular promotional offer.

* Scoring engine: Scoring involves a software application—often called the scoring engine—that can evaluate mathematical functions on a set of data inputs. The scoring engine takes a predictive model and a dataset and produces a set of scores for the records in the dataset.

The scoring process

• A marketing user identifies a segment of customers of interest in the customer database. The records representing the customer segments might be copied into a separate database table.
• The selected group of customers is then scored by using a predictive model. For example, a model that predicts the customer’s likelihood of switching to a premium level of service might be used to get the probability that a customer will indeed switch if he receives a brochure describing the new service.
• The scores are placed in a database table and then sorted by their score value. For example, in a descending order of the probability of the customer switching to the premium service.
• The top x percent are then chosen, for example, as targets for a promotion. The information necessary for promotion (contact information) is pulled out of the data warehouse for sending the brochures.

The author is an associate

professor of supply chain management at the S P Jain Institute of Management & Research, Mumbai. His e-mail ID is khalid_sheikh@hotmail.com