1. Understanding the model
Firstly, it’s important to understand why and how the model was built. This means partnering with your analytical resources to get a little more into the detail in order to understand how it can best be leveraged as a campaign tool. For example, a model built to develop a customer strategy will have a different implementation to one identifying a specific audience of customers. Both will play different roles in your overarching comms plan and impact the design of your campaign.
Furthermore, understanding the analysis techniques used to create the model helps shape how it is used. For example: A lapsing model could identify customers who have only lapsed from shopping or lapsed from the brand – both are actionable but have slightly different objectives, so they require two different campaign plans. Finally, the insights provided by the model might be better used to drive content, such as a recommendations model. This could be used to optimise existing campaigns instead of a new campaign.
2. Overlaying targeting and contact rules
It’s true, models do a lot of the hard work for you, but the model output alone cannot simply be used as a targeting tool for campaigns. Additional targeting and contact rules also need to be overlaid. For example, if a model predicts a customer’s next action that drives an increase in value, the feasibility of a customer achieving this needs to be considered alongside whether the message makes sense. An instance of this could be the model recommending promoting children’s clothing to a customer, but this is only feasible for a customer who has children.
Additionally, it is important to consider what type of campaign you are looking to build; will it be a daily triggered campaign or a one off solus? In the case of trigger campaigns, contact rules need to be overlaid to prevent customers continually receiving the same communication.
3. Measuring and reporting
Once the campaign is built, it is essential to establish a measurement framework for reporting back performance to the business. Where possible, test a campaign that includes the model against one that does not, as this will allow for a direct performance comparison. In addition, a campaign holdout will provide an understanding of the incremental impact the model has against the natural behaviour of your customer audience.
Finally, slicing results by model variables can add a deeper layer of insights. For example, if a model includes propensity bandings that segments audiences by their likelihood to purchase, it would be useful to understand performance by these to optimise the future campaigns and continually refine.
In summary, with well thought-through execution, predictive models are an integral part of building a suite of campaigns that supports an overarching customer strategy. The key is to remember to use them wisely by planning ahead, refining the execution and regularly measuring success.