Many organisations are looking to set up a data mining capability, often
called the analytics team. Within the organisation, data mining projects
can be initiated by the business or by this analytics team. Often, for
best business engagement, a business-initiated project works best, though
business is not always equipped to understand where data mining can be
applied. It is often a mutual journey.
Data miners, by themselves, rarely have the deeper knowledge of
business that a professional from the business itself has. Yet the business
owner will often have very little knowledge of what data mining is about,
and indeed, given the hype, may well have the wrong idea. It is not
until they start getting to see some actual data mining models for their
business that they start to understand the project, the possibilities, and
a glimpse of the potential outcomes.
We will relate an actual experience over six months with six significant
meetings of the business team and the analytics team. The picture we
paint here is a little simplified and idealised but is not too far from reality.
- Meeting One The data miners sit in the corner to listen and learn.
The business team understands little about what the data miners might
be able to deliver. They discuss their current business issues and steps
being taken to improve processes. The data miners have little to offer
just yet but are on the lookout for the availability of data from which
they can learn.
- Meeting Two The data miners will now often present some obser-
vations of the data from their initial analyses. Whilst the analyses might
be well presented graphically, and are perhaps interesting, they are yet
to deliver any new insights into the business. At least the data miners
are starting to get the idea of the business, as far as the business team
is concerned.
- Meeting Three The data miners start to demonstrate some initial
modelling outcomes. The results begin to look interesting to the business
team. They are becoming engaged, asking questions, and understanding
that the data mining team has uncovered some interesting insights.
Meeting Four The data miners are the main agenda item. Their
analyses are starting to ring true. They have made some quite interest-
ing discoveries from the data that the business team (the domain and
data experts) supplied. The discoveries are nonobvious, and sometimes
intriguing. Sometimes they are also rather obvious.
- Meeting Five The models are presented for evaluation. The data
mining team has presented its evaluation of how well the models perform
and explained the context for the deployment of the models. The business
team is now keen to evaluate the model on real cases and monitor its
performance over a period of time.
- Meeting Six The models have been deployed into business and are
being run daily to match customers and products for marketing, to iden-
tify insurance claims or credit card transactions that may be fraudulent,
or taxpayers whose tax returns may require refinement. Procedures are
in place to monitor the performance of the model over time and to sound
alarm bells once the model begins to deviate from expectations.
The key to much of the data mining work described here, in addition
to the significance of communication, is the reliance and focus on data.
This leads us to identify some key principles for data mining.
(collection)
