Improve Model Performance with New Spatial Variables


Even when you use unique data sets, data models can grow stale. The continuous challenge becomes how do you improve model performance and keep it fresh. When our models seemed flat, we searched for new ways to increase their predictive power. Our solution was to create 4600+ spatial variables that strengthened our models and allowed us to score all of the records, even the ones we couldn’t match to our national consumer file.

Meet the Super Variables

As technology and marketing channels change, we are constantly looking for new ways to improve model performance. When a client brings us a new challenge, we love the opportunity to partner with them and build great solutions. We created a model using standard demographic and lifestyle variables for one of our clients, and the results were flat. Nothing seemed to really pop regarding the variables most analytics teams use today.

After exploring different options, we decided to throw our entire cast of 4600+ new spatial variables at the modeling. When we added our super variables to the data set, the model came alive. We achieved a 42% lift in the top 4 deciles of our new model compared to the model using standard demographics.

Freshen Up Your Model

The new spatial variables increased our model’s predictive power, and we have consistently seen positive results across all our client industries. These proprietary variables give us a unique look at data sets that other providers are not able to see. Some of our clients have their own internal data scientists, and they incorporated these super variables into their own models and are loving the difference they’ve made.

One data science team used our spatial variables to create a brand-new risk model, and it has outperformed their previous models. Adding our super variables breathed new life into their model and helped them exceed the goals established by the executive board.

Score All Your Records

Before we run a data model, we matchup the records to our national consumer file. If we are not able to match the records, we remove them from the data set in order to produce the best results. Even if the records have a name and address, they are cut because we cannot append any of our lifestyle and demographic data. As with most analytics, we want to score all of the records. We always want to analyze every customer record to fully understand the true potential of purchasing as well as likelihood of response.

We are able to score all of the data our clients provide by appending our new super variables to the records. Our team can score the records spatially and predict a finite value for each customer. The super variable results are then combined with our MicroModeling® results to deliver higher quality models and more refined data lists for our clients.

Next Steps

If you want to improve the power of your models, let’s have a conversation to determine how we can best create a test performance track for how you can append some of our 4600 super variables to your data set. Our team works with businesses who need help building a model from the ground up as well as corporate analytics teams who create their own models internally but need to enhance their performance.