Actions Speak Louder Than Words in Data Science Models

Over the last year, I met a lot of new people through work and other personal endevours. As I got to know each of these people on a deeper level, what I soon found was a difference in what they said they believed versus what they did. I actually do not fault any of these people norm attribute them to be liars. Rather, I believe the issue comes from not really knowing ourselves or taking the time to reflect on our actions, something we all need more of. Whatever the case though, the same story was usually true, over time what one says becomes less important and to truly understand another person you will know them by their behavior.

What I found so interesting was the correlation of people in my personal life and how it also applied to predictive models. When building predictive models I started to notice that behavioral data outperformed demographic data as a leading indicator in my dependent variable every time. Therefore, it's not what people say they want in survey’s, or with demographic data that they identify with, but rather the actions they take that lead to consistent results. Therefore, to get faster and more accurate results in your model building process, start with behavioral data rather than demographic or survey data.

Previous articles have highlighted the need to move from demographics in marketing segmentation models, but I believe this principle of using behavioral data should be applied to more than just segmentation models.

For example, let’s say you want to predict who will be a good candidate to target in your marketing efforts for a product purchase. The easy tendency is to think of this in terms of gender, age, income, education etc. but you will achieve far better results if you look at behavioral interactions your target candidate has had with your product, company, or similar products first. For this reason, I would take 10 behavioral data points over 100 demographic variables any day when building a predictive model. If more accurate results aren’t enough, I often find that using behavioral data requires far less variables, which keeps the model simple and is an added bonus.

So whether you are developing character in your own life, wanted to judge others character, or are building predictive models, take time to reflect, analyze the behavior, and remember that actions speak louder than words.