Wallet share is the percentage volume of business that a customer conducts with a bank vis-à-vis the customer’s total needs. Today, customers have the ability to make informed decisions on what they want to buy. And more importantly, from whom they wish to buy. An existing relationship is not a guarantee of repeat business or an expansion of existing business. Competition comes not only from other banks but also from non-traditional sources. For example, consumers can go to P2P lenders for loans as well as retail store brand credit cards.
Wallet share maximization has inherent benefits when you consider the fact that acquiring new customers is 10x more expensive than expanding business through your existing customer base.
Identifying Wallet Size
Estimating the wallet size of customers within your current portfolio includes a combination of the knowledge about customers that is already available and extraneous factors, such as macroeconomic indicators and spending patterns of customers that exhibit similar characteristics.
For example, critical financial and demographic data available within a bank or a credit union forms the basis of estimating the potential and undeclared needs of existing customers. This data, merged with buying patterns and consumption patterns of similar customers, as obtained through say, a credit bureau, becomes intuitively more realistic.
There are several techniques that can be employed to estimate wallet size.
In this approach, we begin with an estimate of the aggregate of a market opportunity within a geographical region. We then proceed to split this estimate across individual customers using heuristics that are based on the customer characteristics within each region. For example, if the customers are companies, the overall opportunity could be divided among the companies in proportion with the number of employees spread across several geographical locations.
The estimates of wallet size are derived directly at the customer level, using heuristics or predictive models that are based on available knowledge about a customer. A common approach is to obtain actual wallet information for a random subset of customers through primary research. A model is then developed based on this data to predict the composition of the wallet for new prospects.
Unlocking the Wallet
The objective of this exercise is to enable us to unlock the customer’s wallet and increase wallet share. A key element to increasing and retaining wallet share is through continuous engagement with customers and active monitoring of customers’ behavior. By comparing behavior patterns of similar customers, we gain critical insights about current customers.
We can now identify key indicators of impending change that could be classified as triggers for a customer’s change in status. This in turn, translates into opportunities for cross-sell and/or up-sell.
Some of the key indicators of change are listed below:
- Daily Grind Transactions:
Customers typically engage in certain transactions within a specific periodicity. For example, customers pay monthly expenses like utility bills, auto loans, and rent/mortgage payments. Any change to the frequency could indicate a change in status. If the account is otherwise regular and the credit score is healthy then it indicates a positive change in the customers’ income profile or cash flow situation (i.e. an opportunity to upsell the customer).
- Blue Moon Transactions:
There are times in which recent transaction(s) are distinctly different from the frequently occurring transactions. For example, a customer has used your bank’s credit card to pay for airline tickets as opposed to regular spends on groceries. The transaction is an indicator of a new, but infrequent transaction which opens up an opportunity to offer an overdraft service or loan product to cover for an unexpected short-term expense. On the other hand, the customer could have experienced an increase in income. This in turn translates to an opportunity for your bank to offer more investment or insurance products.
- Spiker Transactions:
Customers generally tend to operate within a band of spends, usage or utilization of their facilities but any blip in this band is an opportunity for the bank to leverage its relationship and increase its wallet share with the customer. Real-time identification of this observed behavior may be used to make an offer to the individual customer. For example, a high value purchase at an automobile store could indicate a down-payment on a car or an upgrade made to an existing car, an opportunity for the bank to offer a HELOC, automobile insurance or even a personal loan against the value of the car.
Predictive Analytics and Wallet Share Maximization
Machine learning and statistical techniques are well-suited for wallet share maximization. To realize maximum wallet share, we must be able to statistically model customer behavior in real time, at the transactional level. Models must be fluid, unrestrictive and responsive to influential transactions.
In order to effectively model “on-the-fly” transactional data and maximize wallet share, we use robust methods of Bayesian statistics. Bayesian statistics is founded on a single, very simple and intuitive concept: to use the past interactively, to predict the future.
Let us suppose that a coin has ‘n’ sides, where each side represents an individual customer within your portfolio and each ‘flip’ of the coin is a transaction. We may now want to decide which one of the “n” sides or cluster, each customer belongs too.
By analyzing customer data at the transactional level, customers can be interactively ‘regrouped’. The effect would be similar to having a coin with number of sides constantly changing depending on the results of each previous flip. Capturing the results of the clustering in real time therefore gives a set of similarities and dissimilarities between customers, dynamically and in real time.
Using the dynamic clustering approach you can measure the profitability of offers made to groups of customers based on their similarities and current life-stage, as a snapshot. The results can also help you gain valuable insights, by measuring profitability and conversion rates of the offers made to customers over the life of the relationship.
While the concepts of the Bayesian model are highly intuitive, methods to perform the modeling effort are often labor intensive requiring custom user-supplied computer code and sufficient computing power. Algorithms such as the Metropolis-Hastings, Gibbs two-stage sampling and various MCMC methods are regularly employed.
By employing Bayesian methods such as the non-parametric Bayesian Dirichlet described above, wallet share is maximized by dynamically clustering customers based on transactional segmentation.
The dynamic models evolve dynamically over time thereby helping banks keep a finger on the pulse of its customer portfolios.
Wallet share maximization can be driven through the analysis of transactional data of your existing customer base. By applying predictive analytic techniques, you are able to extract meaningful insights about your customers’ behavior while identifying trigger points for opportunities to expand your customer relationship at an individual level, in real-time.