Successful loyalty programs offer value that is relevant to program members. This value can come in many forms—discounts, free gifts, special offers and others—but the key is to ensure that it is what customers actually want. Here at Aimia, a data-driven marketing and loyalty analytics company, we’re seeing first-hand how consumers are becoming increasingly savvy about the way they engage with loyalty programs. According to the 2016 Aimia Loyalty Lens study, 42% of Canadian customers view their data as highly valuable, while 51% of consumers get annoyed when companies don’t use what they know about them to offer better, personalized products and services.
Customers expect tailored and relevant experiences and companies are turning to data and analytics to develop multi-dimensional views of customer preferences and behaviour. Ultimately, the challenge they are trying to solve is how to deliver the right message to the right customer at the right time.
Many companies are becoming adept at recording and analyzing consumer data on several levels to inform their customer communications. For example, records of past purchases are a type of transactional data that provides great insight into customer behaviour, given that buying often occurs in patterns. Companies will typically aggregate such transactional data to create meaningful input variables for predictive models. However, to truly maximize the value of transactional data, one should consider enriching our transactional data prior to any type of aggregation, analysis or predictive modeling. Such enrichment can include simple things like data cleansing and imputing missing values or more complex things like inferring new transaction-level variables based on pattern recognition. Whatever the case may be, there is an enormous amount of untapped informational value that can be leveraged by focusing on data enrichment at the most granular (i.e. transaction) level data prior to jumping to predictive modeling or more advanced analytics.
Aeroplan: A case study in enriched transactional data
Aimia owns and operates Aeroplan, Canada’s premium loyalty coalition program of more than five million members, 75 partners and more than 150 brands across multiple industries and sectors, including airlines, banks and retailers, among others. As such, Aeroplan manages a large expanse of data to ensure Aeroplan Miles are issued to and redeemed by members transacting with Aeroplan partners.
It all starts with enriching your data before jumping to analysis and modelling. Sometimes when we look at our data, we’ll notice certain gaps. For example, postal code and detailed address information is often missing or incomplete.
Without postal code or detailed address information there are many types of analysis that we would not be able to execute against. Some simple examples include: understanding our member’s preferred shopping regions or inferring where our members live, work and travel.
We can estimate the latitude and longitude for a given merchant by first mapping all transactions at that merchant to the home address of each Aeroplan member. Given that we have accurate home address information for most of the Aeroplan membership base, such an exercise would provide a graphical illustration of where a merchant might be located.
As expected, especially within certain categories such as gas stations and grocery stores, most transactions at a particular merchant are made by people that live closest to that merchant. Based on this underlying assumption our team is able to engineer a data product that converts each customer’s postal code to corresponding latitude and longitude, and then estimate the merchant location by taking an iterative weighted average of where the most transactions took place.
Executing this simple algorithm against our transactional data provides us with fairly accurate estimates for many brick and mortar retailers (especially gas stations, grocery stores, local shops/restaurants, banks) when validated against a known dataset. When you look deeper into the data, you can find that there is a correlation between the types of retailers we are able to estimate location most accurately and the distribution of the customers that have transacted there. For example, a local bakery in the suburbs has a low standard deviation or distribution in transactions compared to a fast-food retailer in downtown Toronto which attracted customers from all over the city.
This is a general trend: certain merchants/partners that are located in downtown or urban settings have large standard deviations (i.e. people from a wide array of distant postal code are all transacting at the same merchant). The same is true for regional small, specialty shops and tourist attractions.
Another area where the algorithm struggles to converge is with online retailers. For example, transactions for “Amazon Kindle” are conducted online and addresses associated with these transactions are from all over the place. This is where we find unexpected value! Although the algorithm fails to do what it was original designed for (i.e. providing an accurate estimate for a merchant’s location)—it inadvertently provides us a useful, scalable method to identify online retailers without trying to parse the merchant name for a .COM or .CA.
Lastly, we took this a step further and were able to distinguish between one-time online purchases versus repeated purchases that seemed to be pre-authorized payments—with either monthly or bimonthly recurrences.
With enriched data, we can determine the general areas in which most purchases take place. For example, if we can see that an Aeroplan member shops often in downtown Toronto, since purchasing items like food and gas are fairly consistent we can make an educated guess that they work downtown. Items rated as high or very high in standard deviation require a touch more investigation.
In an age where brands are increasingly working hard to differentiate themselves from competitors, enriched transaction data provides more visibility into spending patterns and trends to better identify opportunities to meet customer needs. In this example, being able to identify preferred shopping regions and locations enables us to deliver the right message to the right customer at the right time. As deletist consumers don’t think twice about shutting out brands that send them irrelevant messages, it’s more important than ever to use enhanced insights to deliver meaningful value to customers.