Making Bank Data Work for Marketing

Housekeeping:
To keep this short I removed some mathematical/statistical jargon regarding why I chose to make certain assumptions. If you have any questions surrounding any of these assumptions please let me know and I’d be more than happy to discuss them and any other parts of my work.

Source Information:
The data contains ~45,000 observations of a direct marketing campaigns of a Portuguese banking institution. The marketing campaigns were based on phone calls. Often, more than one contact to the same client was required, in order to access if the product (bank term deposit) would be (or not) subscribed.

*In P. Novais et al. (Eds.), Proceedings of the European Simulation and Modelling Conference - ESM’2011, pp. 117-121, Guimarães, Portugal, October, 2011. EUROSIS.

Available at: [pdf] http://hdl.handle.net/1822/14838

Getting to Know the Customer

The mean age of a customer contacted by the bank is approximately 41 years old, with a balance of approximately 1,400 euro. If you take a look at the charts below you’ll see that he/she is most likely to be married, a blue-collar worker, has a housing loan without a personal loan, has no credit in default, and a secondary education.

Figure 1: Averages

AverageAge	AverageBalance
40.9	1362.3

Figure 2: Most Common Attributes

MarriageStatus	Job	HousingLoan	PersonalLoan	CreditDefault	Education
married	blue-collar	yes	no	no	secondary

• Renderer: SVG | Canvas
• Download

• Renderer: SVG | Canvas
• Download

• Renderer: SVG | Canvas
• Download

Customer Changes With Time?

The type of customer that we have spoken with does seem to have changed with time, albeit slightly, when we examine average age and balance. We can see this visually below. In the chart, the color of the dots change from from dark to light from 2008-2010 respectively. As balance tends to increase in the chart, we can see a slight trend towards lighter colors; the dark blue region of balances below zero emphasizes this. With age as well, we see that as we move towards the right side of the chart, we see ligher blue dots and less dark blue, which seeems to suggest an increase in age over time as well.

• Renderer: SVG | Canvas
• Download

From a very high level perspective, the most common attributes of our customers, grouped by year, don’t seem to show show much change.

Figure 3: Most Common Attributes: 2008

MarriageStatus	Job	HousingLoan	PersonalLoan	CreditDefault	Education
married	blue-collar	yes	no	no	secondary

Figure 4: Most Common Attributes: 2009

MarriageStatus	Job	HousingLoan	PersonalLoan	CreditDefault	Education
married	blue-collar	yes	no	no	secondary

Figure 5: Most Common Attributes: 2010

MarriageStatus	Job	HousingLoan	PersonalLoan	CreditDefault	Education
married	blue-collar	yes	no	no	secondary

Which Attributes Are Important for Marketing?

I fit the data from the campaigns with our banking customers into a logistic regression model. The model can test two things that we’re interested in: 1) Is a certain attribute significant (important) within our campaigns? & 2) how much does that attribute increase the likilihood that someone will say yes to our campaign? Quite a few attributes were significant, but for our purposes I’ll assume that you’d like to know the factors leading toward greater likilihood of saying “yes” to the campaign. These are:

• Balance - Customers with high balances at the bank. This seems to make sense: customers already with a large account balance probably would tend to have term deposits as they just have move money to spread around.
  
• Duration - Customers who spend a longer time interacting with the campaigns. This tends to also justify the effectiveness of the campaigns in general; the longer you’ve interated with a customer, the move likely he/she is to have a term deposit.
  
• Students - Customers who are students tend to be more likely to hold these accounts.
  
• Tertiary Education - Customers that have tertiary eduction. Unknown education was also significant, but this may be difficult for marketing to target proactively.
  
• Outcome of Previous Campaigns - If a customer didn’t exactly say no last campaign the “outcome other” category, or said yes to the previous campaign, this suggest he/she is more likly to enroll, be enrolled in a term deposit.

Noteworthy

I found it very interesting that the number of times contacted during this current campaign seemed to suggest a lower probability of having a term deposit; this seemed contradictory given that duration of time spent on last interaction was a positive predictor. However, it could simply be that customers don’t want to be feel “pressured” into choosing to enroll in a term deposit. It might be useful to investigate this topic more, to understand more how customers might want to be interacted with.