Customer modelling & scoring: improve client targeting

12 Sep 2016  |  by Ray Kirk

Improve the accuracy and effectiveness of your marketing with these four simple customer modelling/scoring steps.

In this blog, we look at why customer modelling and scoring is important, how it can be used and the 4 basic customer modelling and scoring techniques.  

Why Model and Score?

Many organisations use direct marketing (e.g. direct mail, telemarketing, email marketing) to contact prospective buyers (“prospects”) about their products or services. Often companies engaging in this activity have a large prospect database, however, marketing to all of these prospects is expensive and often counter-productive as potential customers may be put off by irrelevant information. However, customer modelling and scoring enables these companies to identify which message, product or service is the most appropriate to each customer and how ready they are to buy. This segmentation means that these customers can then be targeted with relevant messaging.

How can modelling and scoring be used?

Customer modelling can be used to improve targeting in a number of different situations:

  • A common objective is to target direct marketing at those most likely to take up our offer and buy our product or service (“responders”). This could be to identify new prospects (who are not yet customers) or to generate additional revenue from existing customers.

  • In other situations, modelling and scoring can be used to predict how likely a new customer is to have a high lifetime value. If we can predict who these people are then we can ensure that they receive special preference and highly tailored marketing communications.

  • Alternatively, we may also want to predict customers who are likely to lapse, so that we can contact them before this happens with an incentive to remain a customer.

In these situations (and in many others!) a model is used to score a database to identify the customers who should be included in the marketing activity (the target group).

How do you create a Model?

In all customer modelling situations you work with a customer database, where a subset of customers have been identified as the “target market”. In the examples above, this could be responders to a previous campaign, existing high value customers, or people who have already lapsed.

The modelling process compares the target group to the whole prospect database to identify the characteristics that differentiate them. This might show, for example, that the target group tend to be of a certain age, income band or are users of another product.

image01_2.png

How do you use a Model to create a Score?

The end result of the customer modelling process is a way to generate a score for all people in the database. People are given a high score if they have similar characteristics to those typical of the target group.

image02_2.png

The building of a model identifies key characteristics of the target group. The model is created by analysing data where responses are known and creating a formula or set of rules. This model is applied to data containing the same predictor variables but where the response is unknown to produce a set of propensity scores. These scores can be used to select other customers who look most like the responders in the original target group.

The 4 Basic Modelling Steps

The main steps in modelling are outlined below.

  1. Identify the business question

The first step in customer modelling is to specify the business question that we are aiming to model (also called the response variable, independent variable or target variable). This could be as simple as identifying the records that are flagged as your customers or might be far more complex e.g. based on product upgrades or lifetime value criteria.

This article will consider modelling customers/non-customers and refer to “prospects” as those customers identified by the model as having characteristics similar to existing/good customers. Remember, however, that this is just an example of a larger class of analyses. Similarly references to consumers could be extended to businesses when modelling in a Business-to-Business (B2B) environment.

  1. Explore and prepare data

To build a model we need information that is available on both customers and non-customer records. Customers could be businesses in a B2B environment, or could be individual consumers. Potential predictor variables for businesses include ones such as, geography, company size, business classification, turnover, etc. or for individual consumers these potential predictor variables might be age, income, household size, location, etc. Together with a response indicator (e.g. Y for customers, N for non-customers) this forms the “training data”.

Some thought is required in choosing which variables to use. In some cases you may need to derive new, more relevant variables. For example, you may need to calculate the total spend of customers in the last year, or calculate the ratio of this year’s spend to last year’s spend.

  1. Build the model

This information is fed into the model building process and used to build a formula or set of rules, which will be applied to identify likely customers. There are a wide variety of model building techniques, but at the simplest level, the model formula or rules are a way of describing the key characteristics of your customers which distinguish them from a background consumer or business universe. The model can then predict likely prospects by seeing if they share these key characteristics.

Model building involves validating alternative models by applying them to data-sets in which we already know the customer status and measure how successfully the model classifies known customers. A useful model will give predominantly high scores to customers known to be in the target group. To avoid testing the model directly on the data used to build it, it is common to reserve a proportion of the data as a holdout sample to be used for testing.

  1. Score Database and Select Prospects

Once a model has been built it is applied to the database to give all customers a score. Generally, those with higher scores will be selected (i.e. the most likely to be responders or high value customers).

Some thought is required in choosing the cut-off point and deciding how many of the high scoring customers to select. This decision could be made based on a financial trade-off of the cost of mailing relative to the expected revenue, or other business factors. Customer modelling and scoring also enables you to gain deeper understandings into how your customers interact with your business throughout their buying journey.

Make the most out of your data, download How to Track Customers as They Take A Journey With Your Business now

How to Track Customers as They Take a Journey with Your Business

Ray Kirk

Consultant Developer

Ray is leading the development of AI techniques within Apteco. He is extending and automating the statistical techniques within FastStats® which he developed in his early days at Apteco. This “Apteco Intelligence” will then drive campaign optimisation in PeopleStage and deliver insight in the Orbit online platform.

Subscribe to our blog and get all the latest data analysis and campaign automation news.