| A white paper on how companies should analyse | | | | might make them stop buying from you. It also |
| customer data to gain better business intelligence | | | | provides a scientific method to monitor your |
| and how they can use that knowledge. In an | | | | business performance. When deciding to mine |
| increasingly competitive world, using your client | | | | information from a database, one is faced with a |
| database smartly, to gain a better understanding | | | | wide number of available techniques. Some of the |
| of your number one asset - your customers - | | | | more popular data mining methods are described |
| can make or break the success of your | | | | below: Statistical models |
| company. Most companies use databases to store | | | | Basic statistical measurements - such as means, |
| information about their current customers, | | | | variances, and correlation coefficients - are useful |
| previous customers, business partners, and | | | | in the early stages of data analysis to gain an |
| potential customers. The challenge lies in finding a | | | | overall view of the structure of the data. By |
| way to harness the useful information contained | | | | revealing simple inter-relations within the data, |
| within these high volume databases in order to | | | | statistical modelling can show which in-depth |
| produce intelligent business solutions. Business | | | | technique is likely to bring further information |
| intelligence (BI) refers to the process for | | | | relevant to your interests. Clustering |
| increasing the competitive advantage of a | | | | Clustering is a technique that aggregates data |
| company by intelligent use of available data in | | | | according to a pre-determined set of |
| decision-making. Business intelligence consists of | | | | characteristics. It can be used to differentiate |
| sourcing the data, filtering out unimportant | | | | groups of customers that behave similarly on |
| information, analysing the data, assessing the | | | | certain factors, for example it can classify |
| situation, developing solutions, analysing risks and | | | | customer behaviours according to credit |
| then supporting the decisions made. This white | | | | worthiness, income, age or any other factor of |
| paper describes the business intelligence process, | | | | interest. CHAID Analysis |
| some elementary methods of data mining, and | | | | CHAID, which stands for Chi-square Automatic |
| how you can use business intelligence in your | | | | Interaction Detection, can be seen as the |
| company. Database Enhancement The first step | | | | opposite of clustering, in the sense that the |
| towards gaining business intelligence is to start | | | | CHAID analysis starts with the overall database, |
| with a 'clean' database. Incomplete and inaccurate | | | | and then splits it according to the most important |
| data invariably translate into incorrect | | | | variable until it achieves homogeneous sub-groups |
| management decisions. Duplicate data is also a | | | | that cannot be split any further. A major |
| problem as it can wrongly weigh management | | | | advantage of this technique is that the results can |
| decisions to one side. Whilst a good quality | | | | be presented as an easy-to-read classification |
| database does not automatically lead to intelligent | | | | tree; each split in the tree being accredited to a |
| management decision-making, it is a pre-requisite | | | | single variable (e.g. credit worthiness, income, age, |
| for all types of analysis that attempt to elicit | | | | etc). Propensity models |
| intelligent management. We could draw an analogy | | | | Propensity models - also known as predictive |
| with cooking, where starting with the right | | | | models - have proven to be very valuable in |
| ingredients does not guarantee you will bake a | | | | predicting which customers are most likely to |
| good cake, but there is very little chance you will | | | | purchase a certain product based on a set of |
| bake a good cake if you start with the wrong set | | | | current customers. The results of such a model |
| of ingredients. One of the primary reasons | | | | can be directly used to develop more |
| companies do not fully realise the potential | | | | appropriately targeted marketing campaigns. |
| competitive advantages they can gain from their | | | | Other recognised techniques to extract |
| own databases is the lack of proper integration of | | | | information from datasets are database |
| datasets across departments. Even though all the | | | | segmentation, neural networking, and wavelet |
| information might reside within the company, it | | | | analysis among others. It can be intimidating to |
| may remain elusive due to a fragmentation of the | | | | choose which method will provide the best results. |
| data across incompatible databases. Regrouping all | | | | As shown above, analysis tools can differ greatly |
| internal data into a single dataset or a series of | | | | in their approach of the problem. It is therefore |
| interconnected datasets could be the single most | | | | very important for a company to consult |
| useful step a company might take towards | | | | someone with extensive experience in data mining |
| providing a solid foundation on which quality | | | | processes before going ahead with a business |
| business intelligence can be developped. In some | | | | intelligence project. The best method to use will |
| cases, data entry errors and/or missing data can | | | | vary greatly depending on the time available to do |
| also severely impair the quality of information that | | | | the analysis, what the results will be used for, and |
| can be derived from corporate databases. Sorting | | | | the type of data that is available for the analysis. |
| these issues can range from very straightforward | | | | An important point to consider is whether your |
| fixes (e.g. matching one list against another) to | | | | analysis is guided by pre-defined questions or not. |
| more time consuming processes (e.g. contacting all | | | | Predefined points of analysis are aimed at |
| client companies to update contact details of | | | | understanding certain types of behaviours by |
| individuals working there). Ideally, all inaccuracies | | | | analysing relationships between various |
| should be weeded out of the databases. However | | | | pre-decided influencing factors. For example, a |
| limited time and monetary constraints dictate that | | | | predefined analysis of customer service Vs sales |
| you should bear in mind how this database will be | | | | would illustrate the effect of good and bad |
| used. The level of accuracy required will vary | | | | customer service on sales, and would answer |
| greatly depending on the expected use for that | | | | questions such as how important customer |
| data. Data cleansing and database integration can | | | | service is to customers and how much it |
| provide significant advantages for a company | | | | influences future sales. On the contrary, the |
| over the medium to long term. However, they | | | | objective of an open-ended analysis is to discover |
| are both extremely time-consuming activities and | | | | trends that are not anticipated by ordinary |
| can create a significant strain on internal | | | | immersion in the day-to-day business. Performing |
| resources, making them difficult for a company to | | | | an open-ended analysis internally is often impaired |
| justify. Hiring a third-party to do this job is often | | | | by the expectations brought on by individuals |
| the best solution, allowing valuable information to | | | | working within the company. The techniques used |
| be gained, without disrupting day-to-day business | | | | to analyse data are complex. In order for your |
| activities. Data Mining Analysing the information | | | | company to be able to use the results of the |
| that your company stores in connection with all | | | | data analysis, it is crucial that the results should |
| customer interactions can reveal a lot of | | | | not be clouded by the complexity of the |
| remarkable facts about the buying behaviour of | | | | calculations but are delivered in a straightforward |
| your customers, what motivates them and what | | | | manner. |