How to scale personalized targeting

Marketers increasingly want to provide personalized messaging to customers. However, personalizing CRM at scale can be challenging. A key ingredient to successful personalization is to ensure the strategic goal of the campaign aligns with the targeting method used.

By layering the following forms of targeting, marketers can deliver relevant and tailored content at scale. Below are 3 forms of targeting that comprise a robust “targeting framework” (note that this framework was created for email marketing, but can be leveraged across any form of Below the Line or direct marketing).

1. Rules-based targeting
2. Propensity-based targeting
3. Event-based targeting

1. Rules-Based Targeting

This targeting leverages traditional CRM segments to target a large base of customers. This is ideal for standard, business as usual (BAU) emails; however, even simple segmentation will outperform “spray and pray”, generalized targeting. For example, a fashion ecommerce business can announce their Fall line to all customers, but tailor the creative & copy based on the categories a user has previously purchased.

• Type of Segmentation. Business rule or clustered segments that exist within the business’ CRM system. Marketers should target 4–10 unique segments when using this method.
• How to Operationalize. Create template creatives & copy (e.g. email, direct mail) that are used on a regular basis, but tailor the content and/or offer for each target segment. Most email service providers (ESPs) allow segments to be built within their platform, or automatically upload segments from a database or FTP.
• Share of Sends. This type of targeting should generally cover 40% to 80% of sends for a week of campaigns. Of course, this varies greatly based on the extent that the business leverages the other forms of targeting.

2. Propensity-based targeting

“Propensity models” are regression models that determine the likelihood that a particular event will occur. Examples include the likelihood to make a purchase, become a high value customer or share content from our site. This form of targeting generally lies within the domain of a Data Scientist. However — with 3rd party tools, technically-savvy Business Analysts or consultants — businesses can reap the benefits propensity targeting without the overhead of a Data Scientist.

• Type of Segmentation. This targeting is based on customer scoring — a process that uses regressions to predict the likelihood of a customer doing something (logistic regressions) or the degree to which they do it (linear regressions). Once the regression models are complete, Data Scientists will take the coefficients from the models to “score” customers on a recurring basis.
• How to Operationalize. Propensity targeting should be thought of as “situational targeting” — that is, it should be triggered when a customer is on the verge of a lifecycle or sustained behavioral change. Propensity modeling outputs a score for each user — ether a probability (e.g. attrition risk) or a value (e.g. future LTV). These scores are uploaded to the ESP, and only users within a pre-specified range of scores will receive the email. Generally, these emails contain some sort of promotion or discount, since the goal is to incentivize a change (or reversion) in customer behavior.
• Share of Sends. Again, this will be dependent on testing and strategic forms of evaluation. However, targeting 10% to 20% of the user base is a reasonable baseline. This form of targeting should have a higher priority than business rules based targeting.

3. Event-based Targeting

Also known as “real-time targeting”, this tactic has become popular in e-commerce and other online businesses. It is triggered when a customer takes a certain action on site: makes a purchase, abandons a basket, registers an account. This is an extremely effective tactic, however it requires some technical sophistication to execute.

• Type of Segmentation. This process isn’t segmentation per se; we are tracking users’ actions in real-time and targeting them based on these actions. Therefore, marketers must decide what actions they will include, as well as what communication will be delivered once that action is taken. Developers (or 3rd party packages such as Criteo) can flag these events and deliver the communications in real-time.
• How to Operationalize. This process combines the rule-based nature of rules based targeting with the action-oriented nature of propensity targeting. However, the way we execute this type of targeting (from a technical/data perspective) is very different from the two previous forms of targeting. Specifically, JavaScript tracking gathers on-site behavior and APIs communicate the action to the ESP that automatically initiates the targeting. In general, it is easier to execute this type of targeting through an API rather than ingesting & processing the data in a database.
• Share of Sends. Because this targeting is generally fully automated (due to the need for immediate communication), it is difficult to specify what volume of users will be targeted through this channel. It also depends greatly on the number of events against which we target. However, 1% to 15% is a rough range for the share of the user base to be targeted. The key is to manage the frequency of these (and all) forms of communication.

Final Thoughts

• Prioritization. Because these forms of targeting are going to be used in concert with one another, prioritization is very important. For instance, when a customer is included in lists for all three forms of targeting, we should prioritize event or propensity based targeting over rules based. This is because we expect the more granular forms of content to outperform generic content.
• Testing. When implementing these targeting tactics, the BI or Analytics team should test their efficiency. This will have a substantial impact on the extent to and frequency with which we use each tactic.
• Execution First. Modeling and data manipulation isn’t the most challenging part of email execution. It is creating a streamlined executional framework that (a) works every time, (b) that is a good experience for the customer and (c) that can be analyzed & optimized. Always prioritize execution and UX over analytics.