|
Data mining
Some vendors of business intelligence tools use the term
data mining loosely to mean any kind of data analysis, but
most offer only basic techniques such as CHAID.
In a strict technical sense, data mining is at the most sophisticated
end of the analysis spectrum (see business
intelligence).
The objective of data mining is to find useful patterns
and relationships hidden in large data sets, with no prior
hypothesis. The term covers a variety of mathematical algorithms,
many of which will already be familiar to those with OR experience.
Typical applications are credit scoring, fraud detection and
market segmentation.
The best-known approaches are genetic algorithms and neural
networks. Both use rule induction techniques to learn
from a sample of historic data. The rules derived can then
be applied to new data to predict missing values, or to estimate
the likelihood of future behaviour patterns e.g. fraudulent
transactions.
Perhaps the biggest weakness of these and other mining techniques
is that they offer no explanation of the results generated.
The most advanced mining tools have facilities for testing
alternative models to find the best fit. These may include
automated multi-variate statistical analysis procedures such
as CHAID, cluster analysis, factor analysis and stepwise regression.
Thus there is some overlap in the techniques used for data
mining and traditional statistical analysis.
Organisations that have tried data mining report two very
useful findings:
- data mining usually requires large sample sizes (very
large to detect rare behaviour);
- success is more likely where a domain expert and technique
expert collaborate.
A data warehouse containing detailed behavioural data is
therefore needed to make the investment in data mining tools
worthwhile.
For a more information on data mining, refer to Kurt
Thearling’s website
Customer relationship
management
The central idea behind CRM is to help the organisation focus
on its customers, and to recognise and deal with them as individuals.
The aim is often to identify and keep the most profitable
customers (based on lifetime value) or to facilitate cross
selling.
To do this well, an organisation must integrate systems and
processes for contact management, sales pipeline management,
sales order processing, post-sales/technical support, market
segmentation, direct mailing, campaign management, customer
service and sales performance analysis.
As if this were not sufficiently challenging, early adopters
of CRM technology have found that customers may use different
channels at different times. They therefore need to capture
and integrate data from all possible communication channels
(known as touchpoints),
including telephone, letter, fax, face to face, e-mail and
website and to do this in real time.
Despite the number of recent mergers and acquisitions, few
CRM tool vendors offer well integrated, best of breed solutions
covering all these functions and channels.
A customer data warehouse is now considered to be a pre-requisite
for effective CRM. Indeed, most of the business benefits are
likely to come from the CRM initiative, not the warehouse
supporting it. Thus CRM may provide the reason for building
a data warehouse, or at least a place to look for demonstrable
quick wins.
A CRM initiative is likely to be more complex than merely
implementing a data warehouse. There will be much wider systems
integration issues, and more radical process changes as the
organisation shifts to a more customer-centric culture.
E-business
The rapid growth of e-business is having a major impact on
data warehousing.
Web browsers are very easy to use, and require minimal maintenance.
Many organisations now have an intranet, which provides a
ready infrastructure for delivering information to users.
The web has thus become a popular platform for business intelligence
applications.
As the corporate web site becomes a sales channel in its
own right, so the web activity log becomes a new data source
for the warehouse. These so-called "clickstream data"
can generate massive data volumes and require significant
transformation before they are useful. Specialised tools for
clickstream
analysis are evolving rapidly, but on their own, such
data have limited value, because most visitors cannot be readily
identified.
However, if visitors can be persuaded to register as users,
then their clickstream data can be linked to other customer
data to yield real insights into their preferences and buying
habits. Many organisations are prepared to invest heavily
in data warehousing, CRM, data mining and e-business technologies
in the hope of gaining such potent knowledge. Those with access
to OR skills are more likely to succeed.
|
RELATIONSHIP
WITH OR
