Loyalty Program Optimization Relies on Getting Personal

Loyalty360 talked to Senior Consultant Manolis Bardis and Associate Director Andy Rigby of Willis Towers Watson about their views regarding how more advanced modeling techniques can help loyalty program sponsors receive more long-term bang from their investment and marketing dollars.

Loyalty programs already model member behaviors extensively, so why do you feel they can benefit from upping their game?

Bardis: As loyalty programs continue to grow in both popularity and scale, finding the balance between unlocking additional revenue opportunities and granting points that can represent a very material liability on a company’s balance sheet will be even more important than it is now.

The predominant current approach is to analyze summarized data of a program's redemption behavior. A limitation of these aggregated data modeling techniques is that they can’t tell sponsors about individual members’ earning and redeeming activity that, in turn, provides significant insights on the overall benefits and costs of a program.

What’s the way forward then, in your opinion?

Rigby: For companies looking for a richer and more detailed understanding of the financial impact of the behaviors of members, we advocate the use of decision trees and CLTV (Customer Lifetime Value) modeling. These group members into a number of clusters based on similar accrual and redemption behavior. The future performance of the program can then be forecast more accurately from this reduced number of groups.

With CLTV modeling, we can model the different states that a customer may be in, the propensity of a customer to change state, and the impact that each state has on accrual and redemption behavior. For example, one model might predict how likely the customer is to stay engaged with the product, while another might predict spending patterns while engaged. By combining these models, users can predict the overall expected revenue and liabilities for a given customer, as well as enhance their understanding of how each customer arrives at this position.

How tried and tested are these techniques?

Bardis: CLTV is certainly not a new concept. CLTV modeling has historically been used in retail, banking, insurance and other sectors to estimate profit and loss at an individual customer level throughout the time that the customer is engaged with the company.

For example, this approach has been used by retailers to determine the appropriate offers to present to customers, and by financial institutions to specify tailored rates and prices. But as I said before, their use in loyalty programs remains fairly rare, despite characteristics that make them plainly suited to this application. Combining these modelling techniques with our own well-proven LoyaltyAdvisor framework, we can provide significant insights to loyalty program managers.

What kind of loyalty program usage do you envisage?

Rigby: Fundamentally, CLTV, used alongside decision tree modeling approaches, can provide a much deeper understanding of risks and benefits. By analyzing the earning and redeeming behavior of individual members, it is possible to identify behavioral patterns that would be hidden from a more aggregated analysis. Comparing the marginal revenue and marginal cost that results from each additional point accrued, sponsors can identity the most profitable members of a program.

How does this capability translate into optimizing program investment and returns?

Bardis: A greater understanding of member behavior can allow sponsor companies to more accurately estimate how members’ accrual, redemption and engagement would change as the terms of the loyalty program change. So, this approach would allow a more accurate prediction of how customer engagement and redemption may change as the expiration rules are varied, or predict the uplift in engagement if the awarded points are increased for a given transaction type.

Ultimately, this greater understanding can be leveraged to optimize the loyalty program, offering program terms and features that best meet the needs of the most valued customers.

You mentioned potential marketing benefits as well. What are some of these?

Rigby: They all relate to more precise targeting of marketing and communications, based on having greater insights into the relative performance of various segments of the membership population.

Marketing departments can typically target advertisements to specific demographics, but most won’t have access to information on how actual or proposed loyalty program changes will impact members’ earning and redeeming behavior. With an analysis conducted at the individual member level it becomes simple to segment these behaviors and identify better performing segments to target.

Are there any dangers from being a first mover in the use of CLTV and decision tree models?

Bardis: Very few, if any, that I can see. The techniques are well proven in retail. And those loyalty programs that adopt them will potentially steal a march on competitors that continue to rely principally on more basic aggregation methods.

Recent Content