digital strategy, data & analytics
The power of using primary and secondary data for geotargeting optimization
Our digital marketing team conducted a deep analysis of internal and external data to identify the highest ROAS generating areas and improve geotargeting strategies.
Who doesn't know that marketing is only effective if you target the right audience?
Working with the financial sector partner, our team had a chance to conduct a full analysis project aimed at improving the company's geotargeting strategies. By using all available data, including both years' worth of internal data and readily available demographic data, we have been able to identify such trends, which would later allow us to create an entirely new and effective marketing strategy.
The main goal was to identify the areas (up to zip level) which have previously proven to generate the highest ROAS. The project's biggest challenge was bringing together all data at hand and picking out the factors which would later be the basis for evaluating state performance. The latter was going to be done using a single index number comprising all the factors affecting the ROAS.
Having access to millions of customer records and second-hand data, we have planned to put it all in use and be open-minded about finding new unforeseen results. We have chosen to avoid as much data loss as possible while merging and data cleaning. Our main plan was to aggregate all internal and external data up to zip/area level and keep all variables for the correlation analysis. After the correlation analysis, we would have been able to identify the most affecting factors, which would later have a weighted impact on the final index.
Step 1 - Data preparation, cleaning, and aggregation.
To be able to measure the performance of a single zip area, we have created a single view report by merging the multisource data. Those who have ever worked with data from different sources know well how challenging and time-consuming it can be to finish this process. Luckily, the client already had the data in the GBQ database, so we just had to bring all data into single formatting and aggregation.
Step 2 - Correlation analysis
After having all data ready for analysis, it was time to explore and test existing hypotheses for factor selection. The primary assumption was that states with higher standards of living would produce higher ROAS for the company. To prove the assumption, we have made use of external demographic data. Correlation analysis has been conducted between demographic factors and client quality scores to get more detailed insights into the following:
- Do higher income rates actually provide us with higher-quality leads?
- What other demographic variables could affect the close rates and lead quality?
Step 3 - Index Creation
After identifying the highly correlated variables with higher ROAS, we were ready to merge them into a single index numeric value. It could allow us to rate the zip areas and create a top list that would serve the purpose of further geotargeting strategy building.
To build the index, we have first brought down all the features to a common scale without distorting the differences in the range of the values. After the standardization process, we used the weighted linear combination method to evaluate the zips.
Step 4 - Building geotargeting plan for Yahoo (Gemini)
The analysis was further filtered to only the Yahoo channel as this was the first testing point. Having the complete zip evaluation, we have created control and test campaigns to start the testing.
The results were both predictable and surprising to some extent. As assumed by the marketing team, the lead cohorts with higher income DO usually provide higher ROAS. The exciting and unexpected part of the research was discovering new features which would help us evaluate the zip areas.
After finishing the testing, the results were worth the whole research behind it. We have generated leads with much higher close rates and quality scores which eventually provided us with higher ROAS and total revenue.