The Wow Factory Blog

It depends on the value of the improvement that can be achieved by applying
predictive analysis techniques, on the how well customer purchase behaviour
for what the company sells/offers can be predicted and on the data available
to drive any analysis work undertaken.

The availability of data - in particular the availability of well populated,
multiple variables (of releavence) is absolutely critical in creating models
that will work. If the right data for modeling is available, predictive
analysis can have a real value as it will help the client to better identify
who best to contact, with what offers and can help improve the timing of
offers made to potential / existing customers.

Knowing if or when a customer is most likely to re-purchase or buy
additional products and using this to drive timely, relevant communication
to the customer should increase sales and bring added benefits (such as
discouraging customers from looking elsewhere). However, modeling should
always be combined with common (business) sense; for example if a model
tells you to offer a new combi-boiler to a particular group of customers,
but it's the middle of July, you might be better waiting to communicate the
offer a few months later.


Is it necessary to understand something about the sophisticated modelling
techniques that are used, like logistic regression and CHAID? Do they need
at least some knowledge of statistics?

Some knowledge of statistics is useful and some level of understanding as to
what the software that you are using is actually doing is key - for
complicated analysis it is therefore almost always best to get help from a
specialist. More important than understanding the statistics or the software
itself however is a solid understanding of the data you are using. And most
important of all is how to combine understanding of the statistics, the
software and the data to best interpret the results of the analysis - this
is why best results in our experience are achieved when the specialist and
the client work closely together.

For example, analysis results can sometimes be self-fulfilling - i.e. you
might find the results of analytical work simply reporting back activity
that you have undertaken in the past - thus for example if a company heavily
promoted a specific product at a specific time in the past and the analysis
shows that this product should be promoted again, are the results really
showing this or are they simply showing up the fact that the product was
heavily promoted in the past. What's been happening in the 'real world'
always needs to be considered alongside the analytics, as some things may
not always be as remarkable or important as analysis results alone might
show.


Can any marketer create useful models on their desktop? Or are they better
advised to leave it to the experts?

Some useful analysis can be undertaken using readily available desktop
software, providing the user knows how to prepare the data for analysis
(which is usually around 80% of the actual 'modelling work' involved).
Where an expert will really show their worth is in the preparation and
selection of the data, using the right technique(s) for the particular
project and in determining how initial analysis results are best adapted or
re-run to generate the most valuable learnings. Simply relying on desktop
software alone to give you the answers first time can result in expensive
and painful mistakes being made. As most marketers are not analysis
specialists, it is generally a good idea to have an expert do the actual
data preparation and analysis work, but to do the interpretation of the
results together.

This article is posted by Simon Clubley on behalf of blog contributer Renee Hopmans.

For more information on Customer Modelling and Predictive Analytics please contact Simon Clubley at The Wow Factory. 01233 713852 | wow@thewowfactory.co.uk