Business
4 Banking Analytics Use Cases and What Data You Need
The example use cases for the early adoption phase of the data team
The use case is solid proof of any concepts. You can easily recognize the benefit of the concept throughout the pre-defined success metrics. When the number goes up or down, you can compare it to the traditional one. It is so easy and comfortable for everyone to believe in its value throughout the number.
In my working life, when the management team has to invest or spend the money on anything, they will request a process called POC (Proof of concept). It is like you need to implement the small business area task and prove that it is really worked before scaling up the concept. The management team always wants to see the result of POC first before deciding to take any big actions.
However, It is hard to think of the go-to use cases if you never have experience in that field. How do you know which kind of analysis will bring a lot of revenue or conversion rate? What are the best practice analytics use cases that I can bring up to the table to maximize the rate of successful implementation? That is what we are gonna take a look at it today.
Domains
In banking, several departments contributed to the total revenue of the company. For example, the product team who creatively create a new product with many interesting features. The marketing team that makes an attractive promotion to acquire the new customer or boost the baseline revenue. The frontline and sale team who eagerly reach out to the customer to close the deal.
Analytics can help those teams to achieve better performance as the following example use cases.
Sales team
Propensity model
The go-to data science model for the sales team is the propensity model.
The propensity model is the model that tries to predict the customer who has an intention or likelihood to buy the specific product. Any product ranged from the investment fund, insurance product, loan, and any products.
The benefit of it is to scope down the size of the customer to be reached. Let’s say if we have a base customer of around 10 million customers. If the sales team needs to reach all of them in the 1 month campaign period, they can't achieve it due to limited resources.
Analytics help select the targeted customer who has a high potential to take the product so that the sales team can spend time and effort convincing those people who are interested. Besides, the model can be upgraded from the propensity model to be an uplift model if you need a more effective targeted customer.
This saves a lot of time and increases the revenue and conversion rate of the campaign drastically. I have a chance to develop several propensity models when I worked in a banking company. It is a go-to method when you do not have any implemented analytics use cases in the company.
The benefit of it can be measured by doing an A/B testing and compared the revenue generated or conversion rate between the groups. From my experience, when we test the model performance with the traditional rule-based. It is proven that the model always beats the traditional rule-based almost all of the time.
The data: you can make a propensity model based on the buy/sell transaction. You can have the binary classification and use the target as a 1 to take the product and 0 for not-taking the product.
Spending power prediction
After we get the targeted customers from the propensity model, we can improve the lead's quality by adding useful dimensions such as each customer's spending power.
The spending power is how much the size of the product customer will take. For instance, if the customer wants to buy life insurance, what is the proper ticket size that the sales should offer to them. The target of the model here is the amount of money spending on the product for each customer. The features can be anything you have, but the most important one is the financial behavior features.
The benefit of this analytics use case is that you improve the revenue metric directly by selecting only those who have a chance to spend a big amount of money on our product. Usually, I try to develop the propensity model first, and then I will convince them to continuously improve the quality of the leads by adding the spending power prediction later.
The sales team will be much happier with this model result because they have an incentive for its sold amount. If we provide them with the big-spending power leads, they will also get a higher incentive. It seems like a win-win situation for both data and sales teams.
The data: you need to have the size of the product you sold for each customer and use it as a target variable. You can use any regression model to predict the spending power on the target amount of money.
Product recommendation
The opportunity to reach the customer is limited. In banking, if any campaign has touched the customer during a specific month. That customer will be put on the excluded list for 6 months. This is to preserve customer freshness. You know by yourself that you do not want the sale to call you 3 times a week for selling the same product. That is why we have to limit the number of customers touching.
Regarding the above reason, selling more than 1 product might be a good idea for the salesperson because it increases the offering's possibility.
Analytics can help to rank each customer's preferable product and list them to the sales team to make a smooth sale script for several products proactively. The technique that can be used here is the multiclass classification model or ranking model. It depends on the data you have, whether you only have the transaction data or ranking data.
The scope of these use cases can be used with inbound marketing as well. When the customer comes to the branch for doing any financial transaction, the branch officer can propose the ranking product to the customer to increase the chance of the product offering.
The benefit of these analytics use cases can be measured by comparing the recommended product ranking and popular items or the policy-based offered product. We can measure the taking up rate and the revenue of each group.
One thing to be careful of here is that the model predicts a customer's probability of taking the product. But it is not taking into account that the customer will take this product at the proposed time. It only gives you the top 3 products that customers might take.
The data: product recommendation is an upgrade of the propensity model by making a multiclass classification based on all the targeted product instead of 1/0 for each product.
Marketing team
Segmentation
Segmentation is the basic analytics use case for business intelligence. It makes clearer visibility for the view of the customer. The marketing team can benefit from the in-depth behavior of each customer segment. Also, they can track the behavior changes of their customer after the time pass.
The segmentation can be derived from several methods and many views of the data dimension. The basic dimensions could be demographic, usage activity, geolocation, psychology, etc. we can combine every dimension of segmentation to make a complex multi-dimensional segmentation. For example, you can derive the first jobber segmentation from the demographics and combine with the high usage customer from the usage activity dimension.
In terms of implementation, you can make a segmentation based on the heuristic approached, like the rule-based from the exploratory data analysis and some basic statistics.
On the other hand, you can do a segmentation based on the modeling technique such as K-Mean, Hierarchical segmentation, etc. There is a lot of option for modeling out there. You need to review and select the proper method carefully. Knowing that what is Pros and Cons of each technique so that the analytics result will be much more reliable for making a decision.
The result could be a list of customers with the naming of their persona, or you can put it in the dashboard for the marketing team to have a grasp of their customer persona over the time attached to the number of performance metrics. The marketing team can think of the next action they have to provide to the customer to maintain or boost the performance metrics.
For these use cases, it is hard to measure the success of the analytics use cases because there are no metrics for judging whether this is a good customer group. It depends on the consensus between the analytics team and the end-user. But If you can derive the segment with some of the revenue related features, you can do the AB testing to compare the sales revenue and uplift between each group and see how effective the segmentation is.
The data: any data you have at the customer/membership level can make a segmentation based on each dimension and feature. The idea is unlimited here.
Conclusion
I rank the most analytics use cases that can be implemented with a high success rate based on my experience here. Those use cases can be applied to other industries because we have the same foundation of business operation, which is making the revenue.
There are a lot of analytics use cases left that I have not mentioned here yet. Please feel free to share your go-to method about this. The more we share, the more we can learn from each other and improve our best practice to the next level.
