Towards the Marketing System that Thinks

Click here to start

Table of Contents

Towards the Marketing System that Thinks

My hope is that by the end of the talk that: (1) you will want to rush off and build your own Marketing Systems that Can Think (2) you will know in general how to do it (3) will see some real potential benefits

An opportunity for me to share with you some of my ideas about Intelligent Marketing Systems that no one (so far) seems to want to use!

I have nothing to sell you, just a few ideas to tell you about!

Yet most of what I talk about tonight exists in a research context and it is currently being tested in real world applications

What is a Marketing System that Thinks?

Are you soon going to be out of a Job?

So what is a Marketing System that Thinks?

How do you SPOT a DUMB system?

Two ways of becoming smarter

The SECRET of building Marketing Systems that Think lies in their targeting mechanism

Three Broadly Different Approaches to Targeting Customers

Some common attributes:

In the 1990’s S4 became R4

Do your systems pass the S4 test?

Geodemographics and Geo-Targetting is good place to start as the THINKING MECHANISMS that apply here also apply elsewhere

Geo-Targetting Technologies formerly the S4 method known as Geodemographics!

Why seek to DEVELOP geodemographic systems any further?

There are TWO major problems

The truth is that you probably cannot answer any of these questions at present!

the same can also be said of most other targeting methods

Have you ever been tricked by your Targeting Methods?OR Are you a victim of their evil and inefficient ways?

One problem is that of retaining ROBUSTNESS whilst achieving better targeting precision

Another problem is that of Narrow Mindedness

Complacency is another difficulty!

A golden Rule is often overlooked..

Building Smart Systems is not only a matter of using Smart Methods as Smart Methods need to be embedded within Smart Frameworks in order to gain maximum benefits

There are no instant FIX plug-in replacement methods

A mixture of: misplaced Faith, Ignorance of Alternatives, and Dumb Systems create Poorer Results that would otherwise have occurred

So why bother Developing Geodemographics further?

It should be possible to offer an upgrade path that builds on current systems and starts you on the road to Building Marketing Systems that Think

Some Geodemographic Trends

Some Observations

Current Developments

BUT.. we are now MUCH wiser

IMPORTANTnot to get too carried with the technology when it is the methodology that needs most attention

Neural Net based Geodemographic Systems are not vastly better than earlier ones

It seems that the world’s BEST geodemographic classification with 50 clusters developed on the UK’s fastest parallel supercomputer will nearly always be outperformed by a conventionally produced system using a PC with 75 or more clusters in it!

There is NO WAY of KNOWING the answer without EXPERIMENTATION specific to EACH application

This is one of the SECRETs of building SMART Targeting Systems

An Intelligent Geodemographic System:

Updatable Geodemographics?

Use the Best Available Classifiers

Performance is more dependent on the number of clusters used than on the quality of the classificationBUT the optimal number of clusters is application specific

Fuzzy Geodemographics was first suggested over 10 YEARS AGO but so far there is not a single system anywhere in the world!

A major behavioural mechanism in marketing are Neighborhood Effects

How do we build them in EXPLICITLY???

Fuzzy Geodemographic Systems

Three sources of Fuzzyness

Research has shown that fuzzyness exists in geodemographics

Fuzzy Geodemographics should now be a feasible technology

We STILL need a fundamentally different and new way of APPLYING geodemographics

Enter an Intelligent Geodemographic Targeting Machine (IGTM)

Definition of BEST

IGTMcomes with a limited assurance that it will be difficult or impossible to obtain a better geodemographic Targeting result

Other Geo-Targeting

Are you being REALLY Geographically DUMB right now?

Geographical Information Systems puts the “Geo” back into geodemographics

Using a Geographical Analysis Machine (GAM) to find clusters of responders

Another Example for a different mailing

What? You do not already do this??

If you wanted to, could you do it?

Argggggggggg! How silly!!!There are almost certainly patterns in your data that you could use to improve your geo-targetting if you kept the historic data

If you want to build SMART marketing systems you MUST KEEP your historic data

KEEP your historic data

Cannot afford the storage costs?

Database Selections

You may havebeen tricked!

Response Behaviour Modeling

Trends in Response Modeling

Then Modeling Inflation set in

Data Warehousing another fashion

The problems with Data Warehouses are TWO-FOLD:(1) the Data(2) the Data Mining

If YOU don’t evolve the DATA USING technologies then having access to more and more data, more bandwidth, and down-sized hardware WILL NOT HELP at all

In the IT age a computerized bureaucracy has computer based management systems covering most areas of modern life data are being created and stored many times faster than it can be processed to produce something useful!

At present probably 90%+ of all marketing databases are not being exposed to state of the art analysis and modeling technologies

The situation is rapidly becoming WORSE..

So you need GOOD Data Mining tools

The problems at present are:

Meanwhile getting back to Response Modeling

A Response Model

Yes most models can be regarded as Black Boxs

The secret is to create a Marketing System that allows Smart Modellers to adapt to almost real-time patterns of response

Computational Marketing

A “living” system is one which

A “living” marketing system is a Marketing System that “THINKS” for itself

How do you build one?

How do you create a Thinking System that is Alive?

One solution is to switch to a self-learning technology which can be automatically re-tuned without being re-specified

Statistical Models can likewise be Automated

How do you know whether Model A or Model B or Neural Net C is working?

How do you choose between Response Modeling and Geo-Targeting?

The intelligence comes from evaluating alternatives and picking the best AND also from having access to alternatives that are themselves smart in their behaviour as long as it can all be automated

Oh YES... BUT what are the problems? Where is the catch?Why is no one doing it?

Ah BUT.. does not it mean that you have to choose a COMPLETE system?

Financial Services Product Case Study

It is a “Find the Best” sort of Problem!

ITM system automatically selects the best SINGLE Targeting system to use

Next step is to develop an Optimal Combiner that seeks an optimization of targeting performance by selecting the best bits from the available whole segmentation systems

This SOUNDS good!

So What is WRONG with this Traditional Approach?

This is REALLY dumb

An Intelligent Computational Marketing Approach

This is much more intelligent!

Want something to Read?

I know it can be done but DO YOU?

Author: Stan Openshaw, University of Leeds

Email: stan@geog.leeds.ac.uk

Home Page: http://www.geog.leeds.ac.uk/staff/s.openshaw/