Visualizing Bank Marketing Data with Tableau

We are going to review data from a bank's direct marketing campaign and see if we can glean any insights into what factors predict a successful campaign.  In the end, we hope to find some actionable insights we can use in the next marketing campaign.  We will look at 4 of the most interesting graphs produced with the help of Tableau.

 

Graph 1: Education Level

Explanation: The blue bars show # of individuals contacted, grouped by level of education.  The red line shows the percentage of individuals that said “yes”.  

Lessons Learned: We can see a pretty clear increase in the percentage of “yes” responses as the level of education increases.

Actions: The next thing we need to do is determine the cause of this.  Are our product offerings overly complicated, so that only the educated are attracted to them?  Is it just a matter of the more educated individuals having more money and have a use for our products.  Or is it something else entirely?  Digging deeper into this can tell us if we should tailor products to the less educated population, or if we should more strongly target the more educated.

 

 

Graph 2: Number of Attempts

Explanation: The bars again show the # of individuals contacted, this time in logarithmic scale, grouped by the number of times they were contacted during this campaign.  The line again shows the percentage of individuals that said “yes”, in logarithmic scale.

Lessons Learned: We can see that the number of individuals we contact one additional time falls very quickly (remember, logarithmic scale).  Yet, the rate of “yes” responses we receive doesn’t fall all that quickly.  We can then take the raw data into Excel and play with it a little.  We can see that, according to our current data, if we had continued to contact every “no” response again we would have almost tripled the number of “yes” responses we received.  Obviously this would be very unlikely to be the results we would see, and we would probably get a lot of negative PR from calling individuals so many times.  But it does make the point of “why aren’t we making the 2nd contact? And the 3rd? etc”.  We contacted 17,000 people once, but only 10,000 twice, and only 5,000 a 3rd time.  We did this, even though the percentage of “yes” responses only dropped from 13% to 11.5% to 10.75%.

Actions: Find out why we are not contacting individuals with more frequency.  There may be a very good reason we are not, such as legal restrictions or in order to avoid bad PR.  But we are sacrificing a lot by not contacting more of the individuals with more frequency, so we need to understand the pros and cons in order to make better decisions.

 

Graph 3: Age

Explanation: The bars represent the total number of individuals contacted by age, with the thick bars at the bottom representing the number of them that responded with a “yes” and the thin bars representing the number that responded “no”.  The area under the line represent the percentage of “yes” responses for each age.

Lessons Learned: The bars show the distribution of age for the individuals we contacted.  We can see the distribution has a positive skew with a peak at 31.  The amount of data for each age over 60 was fairly small, but once we combined them into a “60+” category we can see that it is a significant segment and that it has a very high “yes” response rate.  The same goes for the 17-25 segment.  We can see that for most of the market, a 10-15% “yes” response is normal, the 40-55 age group is much worse with about a 7-10% “yes” response, and the 17-25 group and the 60+ group have a 20% and 40% “yes” rate, respectively.  It is important to note that these last 2 groups are not of insignificant size, combined they are nearly 3,000 individuals.

Actions: It has become clear that we need to focus as much of our resources as we can on the 60+ and 17-25 segment, with the 40-55 age group being the least important.  We need to think about why these groups are so much more likely to respond “yes”.  Perhaps it is unwise to ignore the 40-55 year old market now if they are likely to sign up with us in the coming years.

 

Graph 4: Past Outcomes

Explanation: Poutcome represents whether each individual contacted had previously been contacted before, and what the outcome of that attempt was.  The “nonexistent” category means the individual had never been contacted, the “failure” category means he/she had been contacted but we had failed to sell them on the product or service, and “success” means the individual had been contacted before and purchased our product or service.  The orange part of the bar is the number of “yes” responses by the individuals in each category, and the blue part of the bar is the number of “no responses”.  The blue line represents the percentage of “yes” responses for each category.

Lessons Learned: The large majority of individuals had not been contacted before.  This segment of the population had the lowest “yes” response of all of them.  Interestingly, the “failure” segment was significantly more receptive to the offers.  This could be because if we contacted them before it was because they are good leads, but the previous product was not right for them.  The most interesting thing we learn from this graph is that the “success” segment has a very high percent of “yes” responses, with over 65%.  The “success” segment accounted for 19% of our total “yes” responses, even though the segment only accounted for 3% of our total targets.

Actions: Now we can see that we need to market to people that we have sold to in the past as much as possible because they are over 7 time more likely to buy from us than unknown people.  


Conclusion 

This analysis gave us many insights, as well as further questions to consider, that we can use when preparing the next marketing campaign.  We know that this campaign was more successful when marketing to individuals with a higher level of education.  We need to look further into the reasoning for this before considering whether this will apply to the next campaign.  We have learned that contacting potential customers multiple times seems to be a very good strategy, but, again, we need to look further into this and test this insight carefully.  We have learned what age groups are best to target.  And we have also learned how effective the previous campaign is at indicating the success response from an individual.  This analysis has proven successful and will be very beneficial to the success of future campaigns.