Skip to content

‹ Back to Projects

Increasing sales of telemarketing campaigns in a bank

Predictive customer targeting helps a bank focus telemarketing campaigns on the customers most likely to purchase.

Predictive customer targeting helps a bank focus telemarketing campaigns on the customers most likely to purchase.

BankingPredictive targetingCampaign optimization

Challenge

The bank wanted to improve telemarketing effectiveness as repeated campaigns increased costs and customer rejections. Its previous campaign success rate was 1%, making better customer selection a priority.

Data

The dataset contained 1,000,000 customers and 500 features covering personal information, accounts, products, insurance and transaction activity. The target recorded whether a customer purchased in a previous campaign.

  • Personal variables: sex, age and nationality.
  • Account variables: balances and credit card debt.
  • Product variables: mortgages and car leasing.
  • Insurance variables: life and home insurance.
  • Transaction variables: cash point and website usage.

Solution

Artelnics analyzed relationships between customer features and campaign outcomes, then built an artificial neural network to rank customers by purchase probability.

Individual correlations were below 10%, indicating that targeting required multiple variables. The strongest relationships involved the number of services held, recent cash point activity and life insurance, highlighting the relevance of customer engagement.

Model

Artificial neural network for purchase propensity.

Targeting

Customers ranked by their likelihood of conversion.

Results

The analysis indicated that selecting the right customers could increase conversion from 1% to 2.5%—from one purchase per 100 contacts to one per 40 contacts.

The cumulative gain analysis showed that contacting the 30% of customers most likely to purchase could capture 80% of total sales.

A profit scenario assumed a €5 cost per contact and a €1,000 benefit per purchase. Under those assumptions, contacting the highest-ranked 35% of customers produced a projected maximum benefit of €750,000, compared with €125,000 without the model.

Illustrations

Logistic correlations between the input variables and the output variable
Logistic correlations between the input variables and the output variable
Customer targeting neural network
Customer targeting neural network
Cummulative gain chart
Cummulative gain chart
Profit chart
Profit chart