Making Relationship-based Pricing a Reality in Financial Services Most financial institutions tend to provide exclusive offers to their key and valued customers. These services do not follow a pre-determined yardstick as they depend on the negotiation skills of both customers and financial advisors. Quite often, a savvy customer would visit a bank’s branches to check out various pricing deals that she can receive, and in instances such as mortgages or short-term deposits, negotiation skills play a significant role in anchoring the final rate of interest or terms of agreement.

Customer Lifetime Value (CLV) is the primary driver for relationship-based pricing. Although the one with the best negotiation skills eventually wins, the other party is left with a feeling of being shortchanged. This is further aggravated by competing institutions providing a “last

Figure 1: Optimized methodology for CLV Step 1

Segment the customer base

Step 2

Derive CLV for each segment

Step 3

Identify product bundles

Step 4

Determine price elasticity

Step 5

Optimize product bundle allocation

Step 6

Define value based targeting strategy

Step 7

Deploy rule engine

ditch” effort to retain the customer by offering the lowest possible price, even at the cost of heavy risks and profitability to the bank. At the same time, it is an accepted market reality that the same price cannot be applied to all customers. Therein comes the concept of dynamic pricing of financial products, which is fast gaining higher priority in business planning. This article attempts to draw a statistical approach to Relationship-based Pricing (RBP) as there are no models on which a banker can arrive at the customized price for a deal. We desist from discussing specific statistical algorithms for usage as the optimum algorithm would be determined on the nature of the data used for the analysis. It maintains the discussion at an activity level and expects the reader to leverage appropriate statistical talent to develop specific models for each activity.

Customer Segmentation For a focused approach to customer segmentation, the bank should arrive at customer groups or clusters based on homogeneity and common characteristics. The homogeneity may be based on various dimensions such as demography, product holding, transactions, interactions, among others. Customers not belonging to any of the identified segments need to be ignored for the rest of the exercise. Figure 2 shows a typical output of an unsupervised clustering exercise on a customer database. The segment profiler for the above segments is shown in Figure 3. This helps in understanding the variables or dimensions defining a particular segment. Each segment has a different set of significant dimensions in varying order of importance. It also shows the

Customer Lifetime Value Customer Lifetime Value (CLV) is the primary driver for Relationship-based Pricing. Naturally, a financial institution should maximize the CLV through a flexible bouquet of products with specific pricing attached to each product. How does one arrive at a specific figure for CLV for each customer and maximize the total CLV for the entire customer base for an institution? Figure 1 outlines an optimized methodology for CLV calculations. Let us look at each one of these steps in greater detail.

Figure 2: Output of an unsupervised clustering exercise on a customer database Segment# (% of base) 10 5%

9 16%

1 18%

8 11%

2 2%

7 4%

9 9% 6 7% 5 13%

4 15%

21

Promontory Issue 3_For Web.indd 21

4/26/2010 10:34:33 AM

Identifying Product Bundles

Figure 3: Segment profiler for segments in Figure 2

distribution of the dimension within the segment as against the customer base. For example, the Balance Total dimension is the most prominent variable for definition of Segment 9. The customers in this segment have lower balances as compared to distribution in the customer base. Similarly, considering the other dimensions, some sort of inference can be arrived at for each segment. Here, Segment 9 probably consists of customer with lower than average total balances and with lower profitability than the rest of the customer base. Deriving CLV for each Segment The definition of CLV is taken as the discounted cash flow of the customer’s

future product usage or purchases less the cost for servicing the products and the customer. The following equation helps in defining the future CLV of the customer (Refer Figure 4).

The probability for the period is a composite of the following: 1. The probability of churn or early termination of product subscription 2. The probability of additional products that can be cross-sold 3. The probability of the willingness of the customer to spend more and upgrade to new solutions

Once the current and future value for each product is arrived at, the next step will be to understand the association between various products. This also needs to be clubbed with possible cross-influences between various products as well as within the same product. For example, it has been observed that the probability of pre-mature termination of a term deposit decreases as the number of term deposit accounts held by the customer increases. Similarly, the probability of default on a personal loan repayment decreases if the customer also holds a term deposit, whereas in the case of a housing loan, the term deposit does not hold any correlation.

Once the current and future value for each product is arrived at, the next step will be to understand the association between various products. The products held by each customer base or segment are subject to an association analysis, which provides an understanding of the interlinking between multiple products in the portfolio of the institution. Figure 5 displays

Figure 4: Deriving CLV for each Segment Net present Value of the Relationship CLV

Revenue of Product i Ri

Cost of Product i Ki

Maintenance - Retention L - Customer services S

Cost of Acquisition I

22

Promontory Issue 3_For Web.indd 22

Probability for period t p

Discount rate d

Making Relationship Based Pricing a Reality in Financial Services

4/26/2010 10:34:34 AM

The next step is to determine the price at which a customer may purchase a product. In econometric terms, the price elasticity of each product is arrived at for each segment and for each product bundle. This is essential since the existence of certain products in the bundle impacts the price of other products. For example, the existence of a housing loan can lead to a successful sale of a lower rate for a recurring deposit compared to the rate offered to other customers.

Figure 5: Results of a link analysis

the results of the link analysis and the respective associations between multiple product combinations. The existence of a connector line between products indicates the association between them, thickness of the line, and strength of the association. Another version of this analysis is the

or products that the customer will buy over a lifetime. Determining Price Elasticity At this stage, the customers have been split into homogenous clusters. For each cluster the expected value from each product is calculated, and, the product bundles that each segment is likely to subscribe to are identified.

The price elasticity of each product is arrived at for each segment and for each product bundle. This step aims at arriving at the various price points and the probability of successful sale or subscription by the customer. For example, consider a personal

sequence algorithm. This activity considers not only the products being held by customers but also the sequence

Figure 6: Rule Table

in which each product was added to the bundle. This gives a good understanding of the typical products a customer will buy at various stages over a lifetime. Figure 6, called a Rule Table, also depicts the above analysis. It shows frequently occurring product combinations on the basis of the support and confidence percentages. The combinations occurring on the topmost rows give the dominant product bundles subscribed to by the customers in the segment under consideration. These are the possible bundles

23

Promontory Issue 3_For Web.indd 23

4/26/2010 10:34:34 AM

function is defined by the business. This could be any of the following or maybe even one outside those listed below:

Figure 7: Product Probability Credit Card 1. APR 8.0% 2. APR 8.5% 3. APR 9.0% 4. APR 9.5% 5. APR 10.0% 6. APR 10.5%

Housing Loan 1. Base + 25bps 2. Base + 50bps 3. Base + 75bps 4. Base + 100bps 5. Base + 105bps

1. Maximize the number of products sold 2. Maximize the expected revenue from the products sold 3. Maximize the product holding ratio for given customer segment

Arrive at probability for each rate / price in each product

4. Minimize the capital adequacy required 5. Maximize customer relationship lifetime 6. Minimize risk exposure

Savings Account 1. Free bill payment 2. Overdraft 3. Multi city checking 4. Gold Debit card 5. Silver Debit card

loan, with a certain base rate that normalizes the price across various points in time taking into account possible inflation. With past data on the price point, i.e., the interest rate, at which a customer either accepted or rejected the product, a price elasticity model is developed. This model will help to arrive at the probability of successful sale for various price points, i.e., the interest rates. Probability tagged to a particular product will vary when the product is considered a part of a different set of products (Figure 7). Optimizing Product Bundle Allocation In this step, the appropriate rate or price point to be applied to each bundle is determined. This is a typical operations research problem where the objective

Using this objective and the product bundle, the product expected value, product lifetime period, product rate card and the probability of success for each rate card option, the application of an optimization algorithm gives the optimum product bundle to be sold to Figure 8: Optimized Product Bundle Credit Card 1. APR 8.0% 2. APR 8.5% 3. APR 9.0% 4. APR 9.5% 5. APR 10.0% 6. APR 10.5%

Bundle #1 1. Credit Card APR 8.0% 2. Housing Loan Base + 75bps 3. Saving Overdraft

Housing Loan 1. Base + 25bps 2. Base + 50bps 3. Base + 75bps 4. Base + 100bps 5. Base + 150bps

Savings Account 1. Free bill payment 2. Overdraft 3. Multi city checking 4. Gold Debit card 5. Silver Debit card

24

Promontory Issue 3_For Web.indd 24

Bundle #2 1. Credit Card APR 9.5% 2. Housing Loan Base + 50bps Bundle #3 1. Credit Card APR 8.5% 2. Housing Loan Base + 50bps 3. Saving Overdraft 4. Free bill payment 5. Gold Debit card Bundle #4 1. Credit Card APR 9.0% 2. Housing Loan Base + 25bps 3. Gold Debit card 4. Overdraft 5. Multi city checking

Making Relationship Based Pricing a Reality in Financial Services

4/26/2010 10:34:35 AM

Figure 9: Customer Strategy Actual Value High

Keep

Keep/Grow

Low

Efficiencies

Grow

Low

High

each customer or customer segment. Figure 8 depicts the possible outcome. Defining a Value-Based Targeting Strategy The optimum product bundle and product pricing are now available for each customer or customer segment. Assuming this as the most probable scenario, the future value of each customer can be arrived at. This coupled with the current value gives us a 2 x 2 matrix with relation to the current value derived from the customer and the future potential value expected from the relationship. A financial institution may decide on a 3 x 3 matrix or even a 4 x 3 matrix depending on its customer management strategy. Let’s look at a 2 x 2 matrix between the current value and the future potential value of a

Potential Value

customer segment. This leads to a fourpart strategy for each customer based on the quadrant each segment is on (Refer Figure 9). The above matrix using current (actual) valuation and potential valuation provides insights into the most effective approaches for customer segments based on the incremental value available within the customer segment. Figure 10 is another way of utilizing the available valuation information to derive a customer differentiation strategy. Deploying the Rules The last step in this exercise is to operationalize the strategic information derived so far. At this stage, the information about customer segments, the product bundle to be pitched as well as

the right price to sell for each product is known. A rule engine will enable the customer facing entity to determine the appropriate product and price to close for each sale to the customer. This will help enforce the optimum allocation and go a long way in achieving the desired results. Any possible deviation from the expected result needs to be noted and fed back into the step 1 so that changing consumer preferences can be accommodated, and the model tweaked accordingly. This approach is a radical shift in the way financial institutions currently approach the customer with their basket of offerings. As such, there is a significant impact on the processes and technologies employed in the operational activities involved. However, the benefits far outweigh the efforts to be invested to enable the institution to treat each customer as an individual, and provide products and services in a personalized manner. This also enables the institution to arrive at an appropriate relationship definition and strategy.

Figure 10: Customer Differentiation High contribution

Most Valuable Customers

High unrealized potential

Most Growable Customers

Potential Value

High potential new customers

Most Growable Customers

Actual Value Neither a high current contribution nor high unrealized potential

Migrators

Negative contribution

Low Value Customers

Feroz D’Silva Feroz D’Silva had served as the head of the Customer Intelligence Practice at SAS Institute India Private Limited for over three years. He can be contacted at [email protected]

25

Promontory Issue 3_For Web.indd 25

4/26/2010 10:34:37 AM