Is it worth appending third party data to better understand your customers?
Invariably one will be in a better position to understand what their best customers look like. This is foundational knowledge for any marketer and can then be used to inform messaging and marketing communication. Moreover, third party data will enable more precise target marketing that can dramatically improve marketing KPIs.
Third party data comes in lots of flavors. At a basic level, there are examples like age, income, gender and presence of kids; but there are potentially thousands of available variables. Some are collected from various public records, magazine subscriptions, sourced from social media and some are even modeled from survey data. Data will also provide behavioral and attitudinal information; for example, likelihood to subscribe to financial newsletters, or interest in gourmet cooking or gardening, as well as likely political affiliation.
A valid question is whether apparently interesting information on an individual or household can be correlated with a client’s very specific customer behaviors. It might seem surprising, but if correlations (data relationships) exist, which is typical, one can often find proxies for specific customer behaviors. This enables us to describe who are our best customers at a fundamental level but can also help us create predictive models to use in target marketing.
A few years ago, I built a model for a home appliance insurance company. On the face of it, it may have seemed that being risk averse would be a key predictor, so financial type variables and propensity to have Whole Life or Umbrella insurance would be great predictors. However, it turned out that customers were buying primarily for the convenience of having someone come out and fix their furnace without any hassle. Proxies like occupation, income, the presence of kids, gender and age were better predictors in this case. Another example was a bank, that was unable to use age, gender, or race for prediction because of financial regulation. These variables were excluded, and other related proxies took their place in the model. The client was surprised when “interest in gourmet cooking” was a key predictor, but one should not be surprised because it is highly correlated with income and affluence.
One must also be careful with data. Choosing a strong analyst that has access and understanding of various data sources is critical. For example, I once built a model for another client and the initial model included a variable based on “likelihood to watch Judge Judy”. Clearly, this was a proxy for other behaviors but it was a very strong predictor in this model. In this case, even though it was highly predictive, it could have turned out to not be a stable variable and therefore affect the stability of the model over time. This is because it is a little too specific and TV channels come and go, along with viewing habits. In the end, a variable measuring the amount of cable TV watched in a week was a more stable and predictive attribute to use in the final model.
Data can be very powerful and data relationships can be obvious, but can also reveal surprising relationships that turn out to be highly predictive because the data attribute is revealing some underlying characteristic. Understanding your customer has never been easier, and along with a skilled analyst and good software, one can reveal insights that really move the needle and dramatically improve marketing KPIs.