Personalizing channel loyalty

First, we create a program unique to the target customers’ needs and interests, an approach also known as idiosyncratic fit in behavioral economics. This personalized tailoring helps customers feel as if they have an advantage as they pursue maximum rewards and benefits. It encourages them to keep using the program.

From there, we add a layer at a time. At the foundation, both the core-program rule structure and earning schedule should be based on customer-specific purchase characteristics. For example, rewarding customers for year-over-year progress is much more effective than rewarding for an average or one-size-fits-all target.

Next, promotions should target and trigger based on, ideally, the “next best action” for each individual customer. For example, targeting promotions based on voids in a customer’s purchase history or complementary products tied to initial purchases. Finally, content and communications should target and trigger based on customer-specific profile information.

Knowing that personalization is critical to a program’s success, it’s not surprising that many are now interested in exploring artificial intelligence to help optimize and automate elements of personalization throughout loyalty campaigns. However, because AI relies on customer data, many companies are not in a position to effectively use it. This is especially true in the channel loyalty space. The good news is there are other options that allow you to take the first steps and build a path to AI.

AI is used in many ways, so it’s important to note that the AI we’re talking about here is the application of advanced computer analysis using techniques that mimic human intelligence to interpret data, determine outcomes, learn from previous interactions, and execute actions. With the ability to scale at massive rates and quickly distill complex variables into simple, applicable formats, AI is an ideal tool for generating fast, accurate personalization.

Here’s an example of a loyalty program for restaurant owners sponsored by a food service company that shows how to integrate AI into a loyalty program. In this scenario, AI would identify the “next best action” or promotion for each individual restaurant based on consideration of a number of different variables:

• Customer-specific data including purchase history, share of wallet, size of business, purchase potential, customer lifecycle stage, and location.
  
• Product-specific data including up-sell/cross-sell matrices, complementary product relationships, seasonality and weather factors, and product margin.
  
• Data on the behavior of similar customers.

Essentially, AI would identify the ideal promotion for each customer individually — the ideal product with the ideal incentive at the ideal time — based on a virtually infinite number of variables.

Let’s look at the options along the path to AI. Targeted marketing and personalization can occur at various degrees of sophistication ranging from having next to no data, to a mid-level amount using data extraction, to deeply complex applications that begin with access to rich amounts of data. Opening the field to thousands of combinations, the choices may overwhelm any marketer.

The path to AI

Level 1: Zero-party data

In situations where there is little to no data on customers, personalization is possible by leveraging a zero-party data strategy: data that customers intentionally and proactively share. In this scenario, we provide an opportunity for customers to share their preferences early in the process. For instance, customers are offered a chance to select from a set of options or share their preferences. In our restaurant loyalty example, the food service provider could ask customers to opt-in for specific promotions or content or complete a survey related to characteristics, preferences, and interests that could be appended to the customer record in the database. The data collected then becomes the basis for future personalized marketing.

Level 2: Data extraction or retrieval

When you have a customer database to work with, data extraction is the easiest method to leverage for personalization. At the most basic level, an analyst manually identifies and selects target audiences aligned with the desired behavior. After the manual approach is proven, the process can be automated. A software engineer or analyst writes code that is run on demand to identify and select target audiences. With automation, data extraction becomes scalable. Costs are reduced. Speed to market is accelerated. Plus, the number of input (target audience)/output (desired behavior) combinations can be increased.

In our restaurant loyalty example, the food service provider would select a product to promote and then identify customers who have not yet purchased the product. The provider could also select products that complement each other and identify customers who have purchased one but not the other.

Level 3: Traditional application development

Traditional application development involves creating algorithms (sets of rules) that analyze multiple data points to identify target audiences that align with desired behaviors based on predetermined criteria. Each potential scenario and outcome is planned, developed, and accounted for within the source code. And every change or addition to the process must be manually added by a human in the form of additional data points or additional source code. Traditional application development can also involve the development of a predictive model where an analyst leverages data mining and statistical analysis to reveal patterns and trends. The insights are then used to predict future behaviors and outcomes.

In our restaurant loyalty example, the food service provider leverages algorithms and predictive modeling to greatly increase the number of input variables and output options. The outcomes then indicate the next best action for each segment or customer. It’s important to point out that the output doesn’t have to be a promotion. This same approach can be used to identify customers who are at risk of attrition, which would then trigger a win-back or retention campaign.

Some examples of inputs and outputs are:

Inputs:

• Customer transactional data
• Customer characteristics
• Customer profile information
• Customer engagement activity or lack of activity
• Similar and ideal customer characteristics
• Customer characteristics tied to retention and attrition
• Product characteristics and relationships
• Geographic factors like seasonality and weather patterns

Outputs:

• Specific behavior or series of behaviors
• Optimal promotional offer
• Optimal timing

Level 4: AI

At this point we have reached the final step in the continuum: AI. Targeting and personalization leverages the predictive branch of AI versus the generative branch. Predictive AI uses machine learning to make predictions based on historical data whereas generative AI uses machine learning to create content like text, images, and sounds from large language models. Predictive AI models find patterns, develop insights, and analyze data to predict future events. The accuracy of predictive AI depends on the quality and amount of data the model is trained on. Through trial and error, the algorithm becomes better at predicting the future.¹

Data considerations

As previously mentioned, data is at the foundation of targeting and personalization, regardless of where you are on the continuum. There are a number of factors to consider as you assess your readiness and develop your strategy:

As you move up the continuum, you can significantly increase the number, complexity, and granularity of the variables used in your targeting and personalization efforts. You can also increase the timeliness of deployment up to real time. However, data hygiene, integrity, and governance also becomes more critical because there are not as many manual checks and the potential negative impact increases exponentially.

---

Sources:

1. Smith, Robert F. “Generative AI vs. Predictive AI: Distinguishing the Difference.” (2024)