Want to understand what
differentiates one group of customers from another? Or responders from
non-responders? Or buyers of product 'Y' from buyers of product 'X'...
Profiling, Scoring and Predictive Modelling
Need a better way to profile your customers and add propensity scores to your database? One of the age old problems when undertaking statistical analysis, segmentation and in particular predictive or response modelling, is being able to transfer the model results to a marketing database for improved targeting. Another issue is that of data preparation - especially where data is scattered across an organisation and its resources are scarce. Fortunately we have an ideal solution. Read on.
What is scoring?
For those of you new to the concept of using scores on which to segment and select individuals for targeting, the following serves as an overview.
Most people have at some time applied for an insurance policy or credit of some kind - a loan, mortgage, finance for a new car, etc. - and during the application process have been asked questions such as age, length of residency, how many credit cards, health history, etc. For each of the answers given, data will have been entered into a computer program that assigns a number (or score) to each piece of information. This is usually termed a score card. The way the numbers are then added together will depend on the formulae used, and this will have been derived from what is termed a predictive model.
The important thing to remember is that whilst each separate piece of information may in its own right not be seen as important or indeed predictive and might otherwise be ignored, only when these items are transformed and combined using the formulae produced by a predictive model does the resulting information and its score becomes significant and predictive.
A predictive model would be usually created by a statistician and used to compare the characteristics (or profile) of individuals who are known to represent, for example, a good vs. bad risk. This produces a formulae and computes a value (or score) for each individual. Depending on how high or low the final score is, the lender or insurer can determine the most appropriate offer for that individual and be fairly confident in the degree of risk.
In direct marketing, to find the most likely responders for your offer could well involve making hundreds of separate or combined counts and selections using only the information that you think is important - i.e. possibly missing some key predictors. However, by using a scoring process the selection task can be simplified as each individual can be allocated a single score that takes account of all available information - automatically selected by its worth - and computed based on the individuals' propensity to take the kind of action you desire.
NOTE. There will never be a single model or formulae that meets all your targeting or profiling needs. A separate model and scoring formulae will need to created to optimise the targeting and response to each different offer you make, or the result you desire. And remember, customer needs' change over time as do the activities of your competitors and thousands of other factors including your data. What works well today may not work so well in the future so the art is to continually test, learn and refine to get the most from your marketing spend!
Profiles
Developing
models, profiles and scores
typically require extensive
analytical or consultancy projects but Tech4T can provide a solution
where
Typical applications for
predictive modelling
-
Selection of look-alike prospects (i.e. prospects who match the profile of selected customers). This is the most common approach and is a natural progression from profiling
-
Using scoring to grade prospects to prioritise mailing, e-mailing, marketing and telemarketing
-
Using different analysis selections, marketing priorities can be applied based on risk - financial, churn, propensity to respond, purchase, upgrade, etc.
-
Using scoring to grade and select best prospects from third party lists
-
Profiling responders against the base of all those marketed to in a test campaign and score prospects to prioritise who to select for the roll-out campaign
The solution - KbaseT (powered by FastStats Weblink) and the New FastStats Discoverer - where predictive modelling provided is easier, faster and arguably provides more insight than some of the alternative modelling approaches, plus the models compare well - especially when the number of variables and categories is high and the data is sparse.
What’s more, the data, selections and model segments are all visible, making it easier to progress, report, avoid mistakes and action marketing based on the model. And to provide the ultimate analytical tool kit we can complement KbaseT with SPSS, AnswerTree, 4Thought and other powerful analytical tools. For more details or to set up a demonstration - optionally with your own data - click the gold blob below.
Data for Enrichment
If your data is lacking in additional information on which to profile, Tech4T can usually help. Please call +44 (0) 1733 890790 or use our call-back form to let us know what you need, how many records you have on your database and whether these are mainly business or consumer addresses. We will then make some suggestions. Also click here for additional information on data enhancement.
