Solving a paradox: refining targeting yet expanding the universe

Data led media planning represents a huge proportion of this financial services client acquisition investment. These media are consistently under pressure to continue to perform in this cluttered environment. Our challenge was to influence performance through our unique modelling techniques.

To meet this challenge, we enhanced our existing door drop model with a frequency capping tool. Our bespoke modelling tool Paragon was used to create propensity models, allowing us to segment the UK into ten groups categorised by responsiveness, conversion and profitability. These models consistently drive high quality leads whilst also allowing us to identify new areas of opportunity to increase sales growth.

Data Strategy
Paragon offers a multi variant approach able to manipulate various data streams to isolate the key predictor. An Exploratory Data Analysis (EDA) was conducted using actual customer data to evaluate the relationship between response, conversion and profitability within each individual postcode sector targeted. Alongside this, other key lifestyle variables held within Paragon were also analysed to understand the demographic make-up of audiences within each segment to drive selections and target new areas. The EDA endorsed and identified which variables were most predictive of performance and drove the ten uniquely different model segments. The model is then applied across the UK with each segment representing between 2% and 7% of the 26 million UK households.

From this analysis we were able to segment the model into the frequency of contact sub sections and outline which cells would deliver against targets.

It was clear that significant efficiencies could be made by cutting out the weaker segments however the analysis also opened up additional potential when considering that the top deciles were consistent in performance with increased frequency of contact.

Results
Three months worth of results were aggregated together to ensure robust test cell size volumes to validate the effectiveness of the model.

We took the cumulative average of each decile to validate which frequency segments would be dropped, as segments which drive a lower CPS would support those that were slightly higher. This would ensure that we maintained volume efficiencies for the overall campaign.

The top two deciles maintained performance throughout the frequency of contact subsections. The greatest gain in performance (as predicted from the match back analysis) was seen in the lower deciles where restricting contact brought the CPS for the overall campaign within the target.

Overall the enhanced model brought the average Cost per Sale down by 22.4%

We've also targeted deciles 1 & 2 with a further two distributions within a 12 month cycle, which has increased door drop volume potential by 14% whilst still delivering the target CPS.