ロイヤルティプログラム：重要な資産

自分がどのくらいの数のロイヤルティプログラムに入っているかご存じですか？最近では、一層多くのプログラムが、利用当初から加入を求めているようです。エアラインの利用者プログラムや、クレジットカード、またホテルチェーンから、Amazon Primeなど最近人気のオンラインプラットフォームのメンバーシッププログラムまで、様々なものが挙げられます。自分自身が貯めたマイルやポイントと引き換えた無料のフライトやホテルの宿泊に大喜びする一方で、多くの企業にとってなぜロイヤルティプログラムが重要な資産になっているのか考えたことがありますか？

事業の競争が激しくなっているため、より多くの企業が、ブランド認知度を拡大し、忠実な顧客基盤からの販売拡大を願う自社のロイヤルティプログラムを始めています。パーフェクトな状態であれば、ロイヤルティプログラムは、顧客およびブランド双方にウィンウィンなソリューションとなります。情報通の消費者なら、サービスのプロバイダーを選ぶ際にメンバーの追加的利点も考慮するでしょう。一方、ブランド側は、忠実な顧客からはリピート販売、そしてマーケティングや新契約コストの削減が見込まれます。図1は、顧客がプログラムのメンバーである場合のロイヤルティへの影響を示しています。

Can you recall how many loyalty programs you have joined? Nowadays, it seems like more and more programs are requiring a membership to start with, from the traditional airline frequent flyer programs, credit cards, and hotels chains to the latest popular online platforms such as Amazon Prime membership programs. While being over the moon about free flights or hotel stays redeemed with your miles and points, have you ever contemplated why loyalty programs are becoming the jewels in the crown for many companies?

As business competition intensifies, an increasing number of companies have started their own loyalty programs, hoping to enhance brand awareness and boost sales from a loyal customer base. In the perfect situation, the loyalty programs present a win-win solution to both customers and brands. Savvy customers would take the additional member benefits into account when choosing a service provider. On the other hand, brands would enjoy repeated sales from loyal customers and reduce marketing and/or acquisition costs. The chart in Figure 1 illustrates the influence of loyalty when customers are members of the program.

Figure 1: Influence of Loyalty Programs on customers ^{1, 2}

Different from short-term marketing campaigns, loyalty programs are continuous incentive programs with risk exposures characterized by their long-term nature and their reliance on uncertainties in future events. Members would accumulate points from purchase activities that can be used to redeem goods or services in the future. These redeemable points are represented as loyalty program liability and could be significantly higher than the company’s operating profit; therefore, its valuation is crucial but challenging to most companies.

For example, the chart in Figure 2 shows that the loyalty program liability has been higher than the operating income for all three major U.S. airlines over the past three years, with American Airlines seeing the largest differences. Any changes to the estimates of loyalty program liabilities would have significant impact on the operating income. At year-end (YE) 2020, these airlines all booked operating loss due to COVID-19, while their loyalty program liabilities kept growing steadily as consumers tended not to redeem points when the economy came to an abrupt halt during this pandemic. Similar phenomena can be observed in China. Although the frequent flyer program liability of Air China slightly decreased to RMB 3.1 billion at YE 2020 (RMB 3.5 billion at YE 2019), that drop hardly offsets the decline in the operating profit to negative RMB 18.5 billion from a positive RMB 9.2 billion at YE 2019. ³

Figure 2: Loyalty Program Liability vs. Operating Income (Loss)
for Major U.S. Airlines ⁴

Loyalty program liability has a lot of similarities to the traditional insurance liability. In the following sections, we discuss further the basic accounting treatment, the estimation of liabilities using actuarial methods and advanced modeling techniques, and what additional values actuaries could add to the loyalty program management.

Basic accounting treatment

In the past, companies could recognize the whole transaction price as the revenue at the time of purchase, while setting up a liability to cover the cost of future point redemption. However, later development in accounting standards such as International Accounting Standard (IAS) International Financial Reporting Interpretations Committee (IFRIC) 13⁵ requires companies to split the revenue related to the sale of goods or services and the value of loyalty points, and the latter needs to be recognized as deferred revenue (i.e., loyalty program liability). International Financial Reporting Standard (IFRS) 15 and U.S. GAAP Accounting Standards Committee (ASC) 606 further require an entity to allocate the transaction price to performance obligations on a relative standalone selling price basis.

To continue with our airline loyalty program example, suppose an airline rewards its members with 1 point for every $10 spent. Each point is redeemable for a $1 discount on a future purchase. During a reporting period, it sells $100 million of tickets and estimates 90% of points will be ultimately redeemed. Based on ASC 606, Example 52, the transaction price of $100 million will be allocated as below:

• Standalone selling price of tickets = $100 million
• Standalone selling price of points = ($100M / $10) x $1 x 90% = $9 million
• Total standalone selling price of tickets and points = $100M + $9M = $109 million
• Revenue related to tickets (recognized immediately) = $100M x ($100M / $109M) = $91.7 million
• Revenue related to points (deferred) = $100M x ($9M / $109M) = $8.3 million

Loyalty program liability estimation

Simply put, the loyalty program liability represents the amount being reserved to pay for future reward points and can be measured using the following formula:

Loyalty Program Liability = Cost Per Point x Outstanding Points x (1 – Ultimate Breakage Rate)

Because not all outstanding points will be ultimately redeemed, the percentage of these unutilized points is the ultimate breakage rate (UBR);⁶ the complement of UBR by definition is the ultimate redemption rate (URR). In our airline example, the URR is estimated as 90% and the corresponding UBR is 10%. Cost per point is the expected cost of each point to be redeemed and may vary subject to different redemption options or changes in program design. Outstanding points are the points issued but not yet redeemed or forfeited and are known as of the evaluation date.

In the extreme case of no redemptions (i.e., URR = 0% or UBR = 100%), the loyalty program liability would be zero. Therefore, the vital ingredient of the liability estimate is to estimate the UBR (or URR), which requires forecasting future redemption behavior. That involves understanding and analyzing trends and patterns in consumer redemptions. Among the considerations in forecasting breakage rates:

• Data quality. Does the system correctly capture all important fields of customers’ personal and historical transaction characteristics? This will determine the exact liability estimation techniques to be adopted. If data permits, companies could even consider building complex predictive models at the individual customer level.
• At what level of granularity should the redemption behavior be analyzed? E.g., by age bands and gender, by region, by different loyalty levels (for example, platinum/gold/silver).
• While historical patterns are an important consideration, redemption behavior may be impacted by various other factors. Quite often we assume that the historical pattern will repeat in the future. Is this true for loyalty programs? Will historical redemption behavior continue in the future? This could vary because extensive advertising of the program could increase consumer awareness, or because of changes in program design, changes in the underlying customer base, seasonal effects such as the holidays when travel and retail purchases are at a peak, or the impact of external events such as COVID-19. Redemptions for airline tickets likely came to a halt except for essential travel during the pandemic.

After data analysis is completed, models need to be built to forecast future points redemption activities. As loyalty program liabilities resemble property and casualty (P&C) insurance reserves in many ways, actuarial triangular methods are commonly used.

Traditional actuarial reserving methods are performed at an aggregate level. For example, models for an airline loyalty program can be built for each member class (platinum, gold, and silver). The historical redemption data by member class is aggregated into triangle format, whereby the rows are point issuance periods (Point Issuance Period method) or member join periods (Aggregate Member Join Period method), and the columns are evaluation ages; by observing the development of points from one evaluation age to the next one, and possibly allowing for any trends, the future expected redemption development patterns can be estimated.⁷ These development patterns can then be used to project the URR.

The simple examples below illustrate the difference between the two ways triangles can be organized.

• Point Issuance Period method
  The points issued in each issuance period (row) are fixed. The cumulative points redeemed triangle shows how issued points are redeemed over time. The forecasted ultimate points to be redeemed in each row, divided by the points issued gives the URR for each Point Issuance Period, as shown in Figure 3.

  Figure 3: Point Issuance Period Method

  
• Aggregate Member Join Period method
  Each row represents a member join year and the cumulative points issued over time. Therefore, the denominator of URR will not be a fixed number as in the case of Point Issuance Period method. We also need to construct the cumulative points redeemed triangle for each Join Period. Dividing the cumulative points redeemed triangle (right-hand table in Figure 4) by the cumulative points issued triangle (left-hand table) gives the cumulative redemption rate triangle, which could then be used to estimate the URR.

  Figure 4: Aggregate Member Join Period Method

  

Comparison of the two methods:

• As discussed earlier, the denominator of the URR for the Point Issuance Period method is fixed, which is not the case for the Aggregate Member Join Period method. In practice, the Point Issuance Period method is more suited for loyalty programs in runoff state, or for programs whose account balance can only be reduced over time. The Aggregate Member Join Period method is more commonly used for programs whose account balance can increase or decrease.
• If the points have an expiry period (e.g., three years after issue), usually there will be bulk redemptions prior to expiry. The Point Issuance Period method can better capture such redemption patterns.
• On the other hand, the Aggregate Member Join Period method allows better understanding of changes in member portfolio, changes in member behavior, or changes in program designs over time.

There are other pros and cons of each method, but we will not go further here.

However, it should be noted that, although loyalty program liability valuation resembles insurance reserves in many ways, there are certain issues specific to loyalty programs that must be considered.

• Loyalty program member behavior could be far more dynamic than insurance claim development and the URR tends to increase over time in well-managed loyalty programs due to mix shift (more concentrated with high-earning and high-redeeming members).⁸ This may violate the underlying assumption of traditional actuarial methods that development patterns in the past will continue to be seen in the future.
• The probability of making redemptions tends to increase as more points are accumulated in the account. After a redemption is made, it may take some time for members to accumulate points until the next redemption is made.
• When constructing the points redemption triangle, assumptions need to be made on which points will be redeemed first. The industry standard for most loyalty program models is "first in first out" (FIFO), under which point redemptions are tied to the member’s earliest earned points that remain outstanding.⁹

A more advanced actuarial method was introduced in 2017¹⁰ that relies on the snapshot date (the cohort being tracked is outstanding points) and observation age triangle as the format for organizing the data for each member. It leverages machine learning algorithms such as decision trees and regression to model redemption patterns for loyalty programs at the individual member level, allowing for more insight into member behavior and more accurate reserve estimates.

A few statistical methodologies can be used as well. Markov chain methods construct three general classes of matrix to project URR, including state, transition, and activity to examine member states, migrations between states over time, and the activities associated with each state.¹¹ Dynamic programming techniques and comparative statics analysis are applied to develop a multiperiod model to estimate loyalty program liabilities.¹²

What additional values can actuaries add?

Now that the loyalty program liabilities are estimated, the next question companies might ask is: how can actuaries add further value to our business?

Actuaries can help design and improve the program to optimize its performance and economic value, by conducting scenario testing on program changes and other cost-effectiveness analyses.

As the amount of data accumulates, actuaries could design loyalty predictive models at the individual customer level. Each customer could be assigned a “loyalty score” based on various characteristics and historical loyalty activities. The analysis gives a detailed portrait of existing customers, which could be beneficial for future marketing campaigns as well as more accurate financial reporting. For example:¹³

• Marketing in advance: Identify what are loyal and disloyal behaviors and give early “red flags” for customers who are likely to drop out soon. Companies could then consider increasing marketing exposure or targeting surprise campaigns to retain these customers.
• Loyalty vs. value: Many companies put marketing focus on customers with higher values. However, another way is to upsell to customers with lower value but higher loyalty.
• Loyalty transition: For example, some customers that are loyal to supermarket brands may be likely to be loyal in other industries too. This provides easier access to acquire a highly loyal customer base and has practical high value to conglomerates and companies looking for collaboration with other industries.

The list above is not exhaustive. Actuaries can play more roles in loyalty program valuations with skills and experience gained from insurance industry analyses. Further innovations on reserving models for loyalty programs may be applied to the insurance industry as well.

As loyalty programs continue to grow, the demand for advanced analytics will increase.

¹Bond Brand Loyalty Report (2021). The Loyalty Report 2021 Executive Summary. Retrieved September 11, 2021, from https://info.bondbrandloyalty.com/loyaltyreport-2021.

²Accenture Strategy (2017). Seeing Beyond the Customer Loyalty Illusion: It’s Time You Invest More Wisely. Retrieved September 11, 2021, from https://www.accenture.com/_acnmedia/pdf-43/accenture-strategy-gcpr-customer-loyalty.pdf.

³Air China Annual Report (2020). Retrieved September 18, 2021, from https://www.airchina.com.cn/en/investor_relations/images/financial_info_and_roadshow/2021/04/27/A9A176C26163C7C1E382D08E55E82C96.pdf.

⁴Data source: S&P Global Market Intelligence (extracted directly from the company’s documents).

⁵Deloitte. IAS IFRIC 13 – Customer Loyalty Programmes. IAS Plus. Retrieved September 15, 2021, from https://www.iasplus.com/en/standards/ifric/ifric13.

⁶In this formula, UBR is expressed as a ratio of outstanding points. It can also be expressed as a ratio of cumulative points issued or earned points and applied accordingly. The key is to maintain a common basis.

⁷Tim A. Gault, Len Llaguno, Martin Ménard (2012). Loyalty Rewards and Gift Card Programs: Basic Actuarial Estimation Techniques. Retrieved September 12, 2021, from https://www.casact.org/sites/default/files/database/forum_12sumforum_gault_llaguno_menard.pdf.

⁸KYROS. The Ultimate Guide to Breakage Estimation. Retrieved September 12, 2021, from https://www.kyrosinsights.com/ultimate-guide-to-breakage-estimation.

⁹Len Llaguno (December 3, 2018). Your 9-Step Checklist for Loyalty Program Financial Reporting. Retrieved September 12, 2021, from https://www.kyrosinsights.com/blog/2018/12/03/9-step-checklist-for-loyalty-program-financial-reporting.

¹⁰Len Llaguno, Manolis Bardis, Robert Chin, Tina Gwilliam, Julie Hagerstrand, Evan Petzoldt (Summer 2017). Reserving With Machine Learning: Applications for Loyalty Programs and Individual Insurance Claims. Retrieved September 12, 2021, from https://www.casact.org/sites/default/files/database/forum_17sforum_01-llaguno_bardis_chin_gwilliam_hagerstrand_petzoldt.pdf.

¹¹Tim A. Gault (September 2014). Reserving for Loyalty Rewards Programs: Presentation 3. Casualty Actuarial Society. Retrieved September 12, 2021, from https://cas.confex.com/cas/clrs14/webprogram/Session7321.html.

¹²So Yeon Chun, Dan A. Iancu, Nikolaos Trichakis (October 22, 2019). Loyalty Program Liabilities and Point Values. Retrieved September 12, 2021, from https://web.mit.edu/nitric/www/lp.pdf.

¹³Vaughan Chandler and Wade Tubman (2012). The Analytics of Loyalty – Qantas Frequent Flyer. Retrieved September 12, 2021, from https://www.actuaries.asn.au/Library/Events/FSF/2012/AnalyticsOfLoyalty3BChandlerTubman.pdf.