How to optimize your loyalty program through data and technology: PwC

The world of loyalty programs is constantly evolving. In recent years, we have seen numerous disruptions in this space, including program restructuring (e.g., mergers, spin-offs), innovations in loyalty offerings, significant accounting changes (Navigating the new revenue recognition standards), and an ongoing global pandemic. Program managers are tasked with designing a program that can connect customers with their brand by driving desired outcomes to increase product demand and customer retention. Robust financial tools are also required to help measure the program profitability and to achieve accurate financial reporting to stakeholders in a changing environment.

The new reality

Leading companies are successfully increasing their share of wallet on existing customers while at the same time growing revenues through the development and targeting of innovative offerings which can connect new customers with the company brand. The ability to extract knowledge and understand your customers can provide companies with a huge competitive advantage in the marketplace. Improvements in technology and data analytics support more informed and proactive decision making by helping harness the power of enormous data sets and more powerful tools, models and techniques. Increasingly, companies are seeking to partner with loyalty data analytics specialists in order to meet companies’ business objectives.

More sophisticated models can utilize more data than ever to help drive customer insights. Automation tools make the processing of data quicker and less prone to human error. Program changes are frequently used to manage the liability and leading programs use predictive models to understand customer behavior associated with program changes as they rely on more powerful statistical techniques.

Meanwhile, the increased complexity and granularity required by the current revenue recognition standard (ASC 606) is aided by more detailed data and technology.

In this paper, we will discuss advanced techniques which program managers can use to harness this data and boost their decision making process, including data visualization and its relevance in predictive models, program liability management, accounting considerations, and actuarial breakage (i.e., the proportion of earnings that program members will not redeem) methodologies. We will then discuss several aspects of the recent explosion of data of which program managers should be aware such as data availability, quality and accessibility.

Leading program managers are utilizing technology in new ways to drive program performance and outcomes. Data visualization helps managers dynamically monitor program KPIs more effectively than outdated spreadsheet tables and graphs.

Data and technology to the rescue

For years, loyalty programs have been amassing a wealth of information on their customers. From transactional data (e.g., transaction level earn, redeem and expiration activity) to demographic information (e.g., member age, gender, location), programs have accumulated a treasure trove of information that is only beginning to be used effectively.

The following are some of the key ways that data and technology have changed the way that loyalty program managers can better understand trends in customer behavior.

Data visualization

Having access to large quantities of high quality customer and program information is of little use if the information cannot be extracted and communicated in a meaningful way. Several visualization tools have become popular in recent years to develop easy to use, customized, and informative management reports and dashboards. Compared to spreadsheets, these tools offer the following features:

Tailored outputs - Program managers can design fully customized charts, exhibits and reports that provide clear answers to senior management’s questions and concerns.
Ease of interpretation - Ability to present information and findings to senior management in a visually compelling way.
Dynamic results - Program managers can use the flexibility of these tools combined with improved processing speed to view various diagnostics and drill into results on demand as new data cuts are needed or as new data becomes available.
Large data sets - Visualization tools separate themselves from traditional spreadsheets with the ability to handle high volumes of data efficiently, with limited need for additional programming expertise.

Access to state of the art dashboard tools, whether they are built in house, or by an external provider such as PwC, is an integral part of leading loyalty programs to quickly enhance understanding key program metrics in order to manage program objectives.

Cloud technology

Cloud computing, at the most basic level, is the on-demand availability of computer system resources, especially data storage (cloud storage) and computing power. Cloud computing relies on sharing of resources, typically using a "pay-as-you-go" model. This structure can allow users to scale up and down much quicker than utilizing their own computer’s hard drive for storage and computing solutions.

The introduction of new cloud-based data management platforms is opening up significant opportunities for program managers. Gone are the days where an external program’s data vendor would mail a quarterly package of information to its client. Today’s managers want near-instant access to their data. Tools are more readily available in the cloud without fear of overspending, allowing managers to test new capabilities without large financial commitments. Ease of access to information across multiple users can also allow programs to gather the information needed to answer stakeholders’ questions quickly and accurately, which raises program managers’ credibility across their organization.

Cloud resources can be integrated with popular statistical tool packages and visualization software, making it an excellent choice for loyalty program data storage where analytics is playing an ever increasing role. Additionally, companies using the cloud are able to work more effectively and reduce costs to help create a competitive advantage.

Where is this helping you with your decisions?

While visualization and cloud computing are nice on their own, they are not helpful if program managers do not know how to apply these aspects of data and technology to make decisions. Using large data sets, program managers not only have a wealth of historical information to allow for more accurate reporting and analysis, but as systems gather more detailed and reliable information, it can be used in powerful predictive models to understand things like customer lifetime value and member retention.

The following examples are key ways which in many programs and consultants are using loyalty program data:

The power of predictive models, combined with large quantities of available data and the most recent modeling and data handling technology, has evolved the way program managers can analyze customer behavior. Machine learning and artificial intelligence approaches are popular for loyalty and non-loyalty modeling applications. Several algorithms such as random forests, k-means clustering, neural networks, gradient boosting, and generalized linear models have proven very useful for the following common loyalty applications:

Customer Lifetime Value (CLV)

This refers to the economic value of a member over its lifetime. Simply explained, one can imagine the CLV as being equal to the sum of all future revenues minus all future costs. It is often thought of as the “holy grail” in loyalty customer modeling as it tells you through a ranking process which customers are the most important to try to incentivize, retain, or acquire. In order to estimate future revenue and costs, an analyst must figure out how to capture and quantify all sources of revenue (member spend) and costs (e.g., redemption costs, tier benefit costs) as well as estimate the probability of attrition at various points in the future. This is where rich historical data and powerful modeling tools can make a significant difference.

Retention Model

A retention model is a subset of the CLV model. It is used to estimate the probability of a member leaving the program over a specified time frame. Program managers find it valuable to create a retention model so that they can impact customer behavior through marketing actions or program changes to help maximize customer profitability. For example, programs may want to offer incentives targeted to members which are currently profitable but that are projected to disengage from the program in the near future (thus lowering these members’ CLV).

Customer Segmentation

Grouping your customers into homogeneous segments allows program managers to better understand their target audience. Are members in a program heavily skewed towards “one and done” members who use the program once, only to leave shortly thereafter? In the hospitality and travel industry, is it heavily weighted towards suburban families traveling for leisure? High income business travelers? Using widely available statistical analysis packages, program managers can better understand the broad types of customers in their loyalty programs, and program features and benefits can be better designed to target certain clusters of members.

Member Scoring

A member scoring model provides a ranking across a particular metric. Examples include past profitability, relative likelihood of participating in a given promotion, or a ranking of the chance that a member will reach a particular elite tier next year. While scoring members based on past activity is relatively easy, building models to predict future scoring is a much more difficult task requiring statistical and modeling expertise. An accurate scoring model is one important tool to improve the efficiency of targeted marketing plans to proactively drive a desired outcome.

Fraud Analytics

These analytics are designed to detect improper transactions. With the rise in popularity and scale of loyalty programs, points have become a target for fraudulent activity. These models mine data surrounding prior fraudulent activity to flag irregular and potentially fraudulent transactions.

Liability Modeling

Generally speaking, these models identify attributes or behaviors that explain how many points (or what portion of points) a given member ultimately will redeem or break. Relying on a combination of modeler experience and the power of predictive analytics, a liability predictive model will identify correlations between variables which are predictive of future redemption at the member level. For example, membership tier, tenure, average earned points per year, average redemption size, and number of historical expirations can predict member tendencies to redeem or break points. While these models utilize complex statistical techniques and can accurately identify drivers of the redemption rate, they run the risk of being “black box” models that are difficult to audit and explain to a non-technical audience. Therefore, there is a cost-benefit to assess when using a predictive model for determining breakage for financial reporting purposes.

While these models are listed separately, in practice they are often used together. For example, a retention model is often used as a key component of CLV models. Customer segmentation can be overlaid on fraud analytics to better characterize the various types of accounts that are exhibiting fraudulent behavior.The unifying theme among all of these models is that they typically require lots of granular data, need statistical and programming expertise to build, and require current technology to help unlock the value of customer data to understand and predict member behaviors.

Loyalty programs thrive when they can drive profitable customer engagement with their product. Program managers frequently adjust the terms and conditions or modify the structure of the loyalty program to incentivize customer behavior. The goal of a program manager is usually to strike a balance between rewarding the right members with generous benefits while managing costs to a level that allows the program to meet its profitability objectives.

Visualization tools and dashboards can be used in combination with scenario testing to illustrate the potential impact of such program changes on members behavior and, ultimately, on the program’s financial performance. Scenario testing generally involves a deep dive into members’ historical behavior followed by the testing of assumptions regarding the impact of various changes. Using robust statistical analysis of historical data can allow you to replace judgmental assumptions with data-supported ones.

Making adjustments to a program is commonly referred to as “pulling levers” and is generally used to alter the economics of the program. Program levers often impact the cost of redemption or the estimated breakage rate of the program, as listed below:

Cost to redeem each point (cost per point)	Breakage
Change in the number of points required to redeem for an award (and/or moving to dynamic redemptions)	In cart redemption capability
Change in the award reimbursement structure (franchise model)	Change in the inactivity or date stamping policy to expire points
Introduction of blackout periods for redemption	Addition of lower point threshold redemption options
Introduction of new redemption types (e.g., gift cards, partial redemptions to offset cash payments, charity)	Introduction of co-branded credit card

Other changes which might indirectly impact either the cost per point or breakage include changing the requirements for attaining elite membership levels, differentiating the marketing or push notifications for different redemption options, or moving to a revenue based earning model (e.g., points issued are based directly on spend). Again, to the extent that program changes impact the cost per point or breakage rate estimate, underlying member behavior might change as well. For example, increasing point redemption requirements may cause members to seek alternative redemptions and slow down the speed of redemption which in turn can increase the deferred revenue accrual.

Ultimately, program managers should understand both how member behavior might change as a result of pulling levers and how to quantify the expected financial impacts of such actions under the applicable accounting guidance. With a large, high quality data set, program managers are well prepared to perform analyses to estimate how program changes will impact their financial results.

While the adoption of the ASC 606 revenue recognition standard beginning in 2018 did not affect the underlying economics of managing a loyalty program, it increased the complexity of estimating financial statement impacts of program changes and simultaneously increased the level of financial transparency due to additional reporting requirements to stakeholders. For example, program managers find that pulling levers may have financial statement impacts that differ from those that would have resulted under the previous standards (e.g., triggering changes to the standalone selling price (SSP), breakage rate assumption, or timing of future point redemptions) and explaining the impacts to stakeholders is more challenging under the current accounting guidelines. The new revenue recognition standards not only change the accounting, but also include new disclosure requirements such as year over year movement in the loyalty liability, assumptions and estimates related to revenue recognition for the loyalty liability, and disaggregated analysis (e.g., for co-branded credit cards) of revenue recognition. In light of these challenges, we have observed that finance teams and program managers will likely need more granular data and are now seeking more robust tools to understand the estimated impact of pulling the various levers, in both direction and magnitude.

Understanding the financial statement impact of changes in estimate as well as pulling levers – impacting breakage, timing of recognition, and for future transactions, the SSP of issued loyalty points for future revenue deferrals – can be critical in quantifying and articulating the drivers and effects of true-ups to stakeholders. Additional data, more robust analysis, and visualization tools can all play a role in this communication process.

The valuation of loyalty program liabilities involves assessing whether members will redeem points along with the timing of those redemptions. Specifically, estimating loyalty program liabilities, often as a deferred revenue, involves projecting the probability, timing, and amount of award redemptions. As programs grow over time, acquire different partners, and more granular data is available, the size of inception to date data sets can become quite significant. What might have been able to be transferred over a USB drive 15 years ago has now turned into gigabytes - or even terabytes - of transactional data available for actuarial breakage methodologies. Depending on the program, the sheer size of data can require specialized technology just to transfer data to a breakage analysis environment.

What should you know about data?

Armed with some of the most important uses of data, loyalty program managers should also be familiar with practical aspects of data. Before jumping into building a predictive model, updating an accounting model, or a new breakage methodology, program managers should consider the types of data available and the overall quality of that data.

Types of data: The data captured by programs vary significantly. Granular data, including transactional level earning or redemption activities, allows for statistical analysis of customer behavior and a large degree of flexibility in the selection of modeling approaches. Demographic data often play a key role in predictive models which try to understand customer behavior. The table below lists examples of data that can be relevant for various customer behavior analyses:

Transactional data	Member/demographic data	Analyst derived data
Number of points earned, redeemed or expired	Member elite tier	Percentage change in points activity over time
Date of activity	Credit card holder (yes or no)	Distribution of points by earn/redemption type
Other activity details (activity location, non-revenue related earnings)	Location	Tenure (time since enrolment at a given point in time)
Point type descriptors (earn/redemption type)	Age & gender	Flag for whether or not a member has redeemed before
Award standalone selling price or redemption cost	Income	Time since last activity

While not all data fields are useful for each analysis, having more data is often a better option than finding yourself with too little data to effectively understand your program.

Availability: Data availability varies significantly across loyalty programs. For example, we have observed that larger, more mature programs tend to have direct access to their data (i.e., “in house”) while smaller or more recently launched programs often use an external data vendor to store and manage the data. Some programs have many years of detailed historical customer level data. Other programs may only have access to aggregate data or may only have access to a limited number of years of data due to historical data retention decisions (e.g., if the program structure has changed through a merger or if information technology systems have changed).

Recent improvements in storage capacity and a better understanding of the potential value of data as an asset is pushing programs toward conserving more data for longer periods. The complexities of data management and large storage volumes have led more and more companies to use third party external vendors to manage their data. Programs also have access to cloud storage to allow users to scale up data needs with agility.

While recent data may be more representative of the current realities of a program, the benefits of historical information should be weighed against data and storage costs. Programs rely on historical information to help understand the impact of prior significant events (e.g., macroeconomic shocks, program promotions) and anticipate how future events may impact them. Historical data is also helpful to better understand the long term value of customers for the program.

While having data storage in the right environment is one aspect to availability, another equally critical aspect is having the right personnel to access that data. Anecdotally, Business Intelligence (BI) and IT teams often have many conflicting professional demands, making it difficult to secure nuanced or one-off data requests in a timely fashion. For example, for a one-off breakage analysis where data is sent to a third party consultant, getting the necessary resources who have the requisite coding skillset, knowledge of the data, and ability to transfer data can be an impediment to such analyses. Program managers should be cognizant of that constraint and consider the appropriateness of bringing BI skill sets into their teams to minimize the friction around accessing data when it is needed.

Quality: While data may be widely available, the quality and usefulness of it varies widely across programs. Ideally, there is a single, easily accessed and well maintained source of truth. In practice, we have observed that detailed data used for program analyses (e.g., breakage analysis) may not always perfectly tie to accounting data. Understanding this situation prior to launching an analysis may take up a disproportionate amount of time or effort, rendering otherwise useful information difficult to use for financial reporting analyses. Other programs may have a single data source, but internal analysts or external data vendors may not sufficiently understand the data which may lend it to be potentially misused. The program’s management team can provide significant insights into possible data issues, which is often critical in the early stages of any statistical analysis. Meaningful analysis can only begin if the data fields are understood and the data has been accurately reconciled and validated.

How PwC can help

Data, and more importantly the ability to use that data through technology to make informed decisions, is revolutionizing how loyalty programs are making impactful business decisions. Of course, data has been collected and analyzed from the first day that loyalty programs were launched decades ago, but now more than ever, opportunities exist to enhance the data mining practices and the robustness of models used by loyalty programs. New technology and visualization capabilities allow to significantly improve program management’s abilities to help make informed decisions and provide better insights to internal and external stakeholders.

PwC’s Risk Modeling Services loyalty specialists work with clients to help build intuitive actuarial models that are tailored to each client’s needs, while meeting the current accounting standards. Our team of specialists has been providing services to some of the world's leading customer loyalty programs for more than 25 years. We regularly collaborate with leading loyalty programs to provide consulting services such as designing data visualization dashboards; quantifying the impacts of prospective program changes; and designing state of the art breakage and deferred revenue models under ASC 606/IFRS 15 that help our clients react to a changing environment and clearly articulate drivers of change for financial reporting to stakeholders.

_{Editors/additional contributors:

Marc Oberholtzer, Principal, Risk Modeling Services}

Managing your loyalty program through data and technology

The new reality