 | | | |
|
 |
Cas client : Renault France Automobiles |
|
| Comment générer des leads commerciaux en concessions lors du lancement d’un nouveau véhicule ? Quel dispositif mettre en place pour obtenir un faible coût d’acquisition ? (en anglais uniquement) |
|
 |
 |
|
|
| | | | |
|  |
British Telecommunications
Analyzing data and identifying the best target |
|
|
| |
| | |
|
 |
Challenge |
 |
|
|
|
To obtain the greatest value from its marketing budget, BT (British Telecommunications) needed to identify customers' propensity to purchase and calculate their likely comparative value once they became customers.
After creating accurate customer profiles, BT intended to develop new products targeted to specific customer groups. |
|
 |
|
|
The Big Idea
Analyzing data and build exploratory models for its campaign, which was aimed at small business customers. The expected results? A higher response rate to marketing campaigns, increased product revenues — and an even greater market share for the company.
|
|
|
Strategic Solution
The once-peaceful telecommunications industry has turned cutthroat. BT, a former monopoly, is a leading supplier of local, national, and international phone and data services in the United Kingdom. It also competes with other U.K. telecommunications companies. To retain its customers, gain new customers, and maximize sales, the company needed facts about exactly who was buying its products and services.
To identify these customers, the company established a customer and campaign analysis team, within its business connections division. The team's first assignment was to model customer profiles for BT's product, which provides small business customers with three telephone numbers, one standard and two digital, on a single line. The launch included a major direct mail campaign and national media coverage.
Finding hidden patterns with data mining
To mine the data sample and find underlying patterns and trends, we used a wide range of analytical methods, including clustering, neural networks, association rules, and decision trees. It also easily handled common data problems, such as outliers, missing data, and low-value data.
Analyzing data and building models
During data analysis, we identified data quality issues, became familiar with the data and data distribution, and eliminated data attributes not strongly associated with the purchase. Then we measured the predictive strength of individual data attributes in relation to the customer's propensity to buy the product. For example, two-digit district codes, a geographic indicator, were clearly linked to response and purchase data.
After the analyses, we quickly built and tested a series of experimental models using decision trees.
Identifying a targeted 'best prospects' list
The main output of our analysis was insights about the data — that's what data mining's all about—and visual representations of those insights. The deliverables to sales and marketing were lists of customers and charts showing why these were the customers they should speak to about the product.
|
|
|
Measures of Success
This project raises issues about how you can benefit from using data mining in business. The exploratory data analysis and visualization we were able to do up front enabled us to develop satisfactory customer selection criteria. Even before completing the final models, we were able to surpass our original target — and increase the campaign response rate by 100 percent.
More work remains. Next, we plan to identify customers who have the greatest profit potential and those customers who demand lots of attention but do not buy. In the future, we may also try to determine consistent patterns for customer defection, often referred to as 'churn.' |
|
|
| |
|
