Current situation analysis

One of the very first steps of EDA is simply taking a closer look at the data in use and identify if there were any particular patterns that could potentially be of use when it comes to potential improvements of the currently employed policy. This was done with the help of various visualization and prediction models. Clients were divided into those who did accept the term deposit offer and those who didn’t in order to see what were the most eye-catching differences between these two groups. Comparison of these two groups in terms of their general shares (as a result of preliminary data analysis) and analysis of subscriptions (as a result of implemented machine learning) is presented below.

Age

Mean age of both subscribers and non-subscribers was practically the same - approximately 40 years old. This demonstrates a good match of the general audience of the bank and its target audience and can be used in further campaigns. A very small difference in the mean age of these two groups, however, also means that other variables had to be found and analyzed in order to find out what made the campaign not as successful as the bank wanted it to be.

Education

The most commonly contacted group of clients only had secondary education. It was, however, clients with higher education who accepted the offer more often.

Marital status

In general, a lot of married clients were contacted during the marketing campaign. However, single and divorced clients were accepting bank’s offer more often than those who were married.

Job

The bank mostly contacted technicians, managers and blue-collar workers. Not surprisingly, those categories were where more clients who eventually agreed to subscribe to a term deposit (in terms of the number of people) could be seen. Shares of subscribers, however, were larger in case of students and retired people.

Balance

Balance level of subscribers was a bit higher than that of non-subscribers. This is, in fact, quite intuitive since it is irrational to keep big sums of money at home when one can store it in a safe place and earn dividends from it.

Variables connected with the results of the previous campaign

The bank contacted a lot of new customers during recent marketing campaign and it appears to be the case that the share of subscribers was bigger in case of those clients who also accepted the offer from the previous marketing campaigns.

The probability to accept the current offer among those clients who agreed to the previous one was 51% while the same probability for those who rejected the previous offer was 45%. Among customers with unknown response to the previous campaign the probability to reject the offer was 86%. This might be explained by both psychological characteristics of clients (some of them being simply more influenced by such marketing campaigns) and by clients who already know something about the company and using their services being more likely to accept the offer in comparison to those who don’t have any experience with company’s services. That is one of the reasons why it is so important to provide clients with good-quality service and invest in programs aimed at increasing customer loyalty.

Additionally, it was found that the share of those who subscribed to the term deposit for bigger for those clients who were contacted more often before the recent marketing campaign and soon after the end of the previous one.

The probability to accept the new offer among those clients who were contacted in less than 100 days was - 72%, among those who were contacted in 100 - 300 days - 54%. Since it can be seen that this probability is declining as time passes which is why it can be concluded that it would be better if clients were contacted not that long after the previous campaign.

Variables connected with the current marketing campaign

Those clients who ended up subscribing to the term deposit, in general, had longer conversations with the call center operators. An operator simply having more time to explain all the details and build a rapport with a client can be used as one of the probable explanations. Additionally, those who eventually subscribed to a term deposit were contacted slightly less during recent campaign, which appeared to be counter intuitive, as by contacting a client more times a company has more chances to create a stronger bond with this client or to persuade him or her that their product is what client needs.

Even though the duration of the last call highly affected the outcome variable, it can not be used for prediction since, technically, the duration of the last call can only be recorded once the call is over and the outcome is known.

Personal and housing loan, default on a loan

People without housing or personal loans tended to subscribe to term deposits more often. It has been concluded that this might be due to the fact that those people still had to pay their loans off first and didn’t have enough money to put in a deposit (and if they did, they probably wouldn’t get a loan). Moreover, it was found that that most of the contacted clients didn’t have any credit defaults which can be explained by the bank preferring more reliable clients. Additionally, clients with credit defaults tended to subscribe to term deposits less often. It is supposed that that is due to the fact that they still have to fulfill their credit obligations before accepting a new offer which requires them not to loan money, but to deposit it.

Just as it was hypothesized, the probability to accept the offer was higher for those clients who don’t have a housing loan (20%) in comparison with those who do (17%). Even though the difference might seem not significant, it can actually result in thousands of clients rejecting the offer and huge financial losses of the company.

In fact, housing loan, number of days that have passed since the client was last contacted, and his response to the offer of the previous campaign were identified as the most important for prediction of the result of the current campaign.

Overall probability of a customer not subscribing to the term deposit is 72%. What needs to be mentioned is that because of a small number of clients who actually accepted the offer (in comparison to those who didn’t) constructed prediction models could correctly predict around 96-97% of non-subscribers, but only 33-34% of subscribers. In order to improve the quality of response prediction models more data needs to be collected and analyzed.

Proposed policy improvements

As a result of the analysis of the current situation, customer characteristics and campaign-related variables that could potentially predict if a client will or will not accept the offer, the following policy improvement suggestions could be derived.

Because of how crucial it has become to implement the newest technology in order to remain competitive on the market, it is suggested to design and implement an app to enable the bank to collect information about its clients and their transactions more efficiently and also make it easier for clients to use various bank services. By simply analyzing the data collected by the app it would be possible to reach such goals as: Increase customer acquisition, Increase revenue per customer, Decrease customer acquisition cost, Reduce customer churn, etc. (Krishnan, 2020).

Variables to be collected and analyzed by the app:

• Age
• Job
• Marital Status
• Balance
• Personal loan
• Housing loan
• Default on a loan

Age, Job and Marital status were included as socio-demographic variables that are usually associated with persistent lifestyle and habits (Campbell et al., 2004) and could even be collected by third-party agencies, while remaining variables were included as those connected to the client’s profile in terms of his/her reliability and profitability. This data would be used to customize deposit plans for clients with particular characteristics to satisfy their needs (by changing interest rates, time periods, etc.), to promote these offers to them using push-notifications, assess clients in terms of reliability etc. For instance, it could be hypothesized that younger clients and clients with no stable job would prefer deposit plans that don’t require big sums of money and enable them to add money to the deposit after the initial sum was placed in the bank. Older clients with a stable job, on the contrary, would prefer to deposit a bigger amount of money. Connections between these variables in such a case can also be taken into account: for example, even if two clients have the same balance one of them might have a more stable job and be more reliable.

Based on the results of preliminary data analysis, a total of eight special offers were designed. Age, Job, Marital status of the client and the months of the deposit offer availability were mainly used as a basis for those.

Older people tended to have bigger balances and more of older clients had no default on their loans, while the situation was completely different for younger clients. It is rather intuitive that younger clients simply don’t have that much money to sign a big deposit contract which is why the plan offered to such clients would require them to only deposit a small sum of money and then simply add up to it as they earn money by working part-time. One of the special offers that suits this is - “Gap Year” which is created to help young people to store the money they have, add up to it throughout a whole year and even earn some interest to continue their education.

The general idea is that by creating and implementing the app it would be possible to increase customer satisfaction and, in turn, positive response rate through deposit plan customization and more convenient way of contacting (push notifications). It is expected that customers would be a lot more satisfied with customized offers because they are tailored to their needs, while the positive effect of a change in the way customers are contacted is backed up with research on the topic of how much less costly and much more efficient push-notifications and messengers are in comparison to direct calls.

To be precise, it has been calculated that:

• The probability of a customer to be highly satisfied with the service drops to 34% if the app is not implemented and rises back to 73% if that is not the case.

• That probability of a customer being highly satisfied with the service would reach almost 80% if both customization and push notifications were implemented and drop to: 44% is a case of only push notifications being used, to 69% if the campaign would be done by directly calling clients.

In general, if the app is implemented, the probability of a customer to be highly satisfied with the service raises to 73% and the probability of customers to accept an offer raises to 78%.

When CLV is calculated marketing expenses for clients are also taken into account (Kumar, 2006) which is why the suggested improvement policy would have a positive effect on the general CLV of bank clients due to the reduced costs (direct calls are more expensive than push notifications) and increased efficiency (positive response rate).

Resources

Campbell, Dennis and Frei, Frances. (2004). The persistence of customer profitability: Empirical evidence and implications from a financial services firm. Journal of Service Research 7(2), 107–123.

Haenlein, Michael & Kaplan, Andreas & Beeser, Anemone. (2007). A Model to Determine Customer Lifetime Value in a Retail Banking Context. European Management Journal. 25. 221-234.

Krishnan, K. (2020). Banking industry applications and usage. Building Big Data Applications. Academic Press. 127-144

Kumar, V. (2006). CLV. Journal of Relationship Marketing. 5:2-3, 7-35.