Customer Lifetime Value

Solution

Customer Lifetime Value (CLV) represents the net present value of a future profit or revenue from a specific customer. Instead of measuring the customer’s past value, CLV focuses on their future potential that has a significant impact on business decision-making, for example in the following areas:

• Targeting marketing and retention campaigns;
• Determining the level of customer services;
• Design of new potential products/business plans;
• Planning and forecasts of the entire business; and
• Portfolio valuation.

Benefits

• Using CLV to target campaigns generates higher revenues.
• CLV has the potential to decrease the customer churn rate and allows to retain customers with a high future value.
• CLV improves the overall profitability of the organisation and return on investment of marketing campaigns.
• CLV can function as the primary indicator of business performance.
  

Method

At the beginning we define the purpose of the CLV model and the target. We discuss with the users what length of prediction horizon is the best, if we are more interested in the future revenue of a customer or the future profit (revenue net of costs). We are interested in the availability of reliable data on historical revenues with respect to the client and time unit as well as in historical costs with respect to the client and time unit. We need to know what instrumental variables should be included in the CLV models for the purposes of a what-if analysis.

In the second stage we start working with data that needs to be profiled, cleaned and accepted. We face the task of creating customer microsegmentation with the input of instrumental variables and predictors that will explain the present value. In this area we have achieved good results with regression trees with a high level of expert input. The right number of segments ranges from 10 to 100 based on the size of the data, so they really are microsegments developed specifically for CLV and not rough segments suitable, for example, for a service model. There are two segments with a special role: inflow (new customers) and churn (customers who have left). The quality of the microsegmentation determines the quality of the CLV model as a whole.

In the third stage we work with the matrix of transition between microsegments. In order to estimate the transition matrix, the whole client basis has to be retrospectively scored by the microsegmentation model for a historical period of time defined in advance. The probability of transition is estimated by comparing a customer microsegment in two consecutive periods and calculating the average for the entire historical period. The probability of transition to the churn microsegment represents the churn rate, which is a significant factor of customer value. The distribution of customers by microsegment is projected to the future using linear algebra and the theory of homogeneous Markov chains and their future value is discounted in the present. The setting of the prediction horizon and the discount rate are the two inputs from the client that determine how far into the future the CLV model looks. In other words, what importance the CLV model attributes to revenues from the distant future in comparison with revenues from the near future. The output is the allocation of discounted future values via the microsegment to the customer level.

In the fourth stage, CLV is visualised and interpreted using dashboards. At the same time a production scoring job is implemented that will calculate CLV changes and historise CLV development in time using fresh data. Now is the right time to use CLV for business decisions. For example, to build a business case on cross-sell, retention or collection. Using a CLV model we assessed the successfulness of the cancellation of little-used tariffs, the introduction of a new type of current account or the valuation of customer portfolio for the purposes of mergers and acquisitions.