“LityxIQ provides an extremely efficient and cost-effective platform for building and deploying dozens of models.”
– Paul Maiste, President & CEO, Lityx
Our client commissioned 25 affinity models to assist in improving the results of their direct mail campaigns. The models were to be based on survey data with approximately 35,000 observations. The organization is quite large, and executes marketing campaigns to approximately 5,000,000 individuals (about 20% of the database) with a frequency of 24 times per year. Random mailings across the database performed at a 1.25% response rate with an average value of about $13.50 per responder, and a fully burdened mailing cost of $0.40 per piece. With this volume and overall cost of the direct mail program, there is quite a bit of room for improving efficiency through analytics.
The challenge goes beyond improving marketing efficiency. Building and maintaining 25 predictive models is no easy task. The client originally used an outsourced vendor to build the models using manual techniques and “classic” tools. We were tasked with replicating the process in LityxIQ in order to improve time-to market and reduce modeling costs.
Background: The original models were built by an outside vendor using standard tools and techniques on a standard hardware platform. In this effort, each model took approximately 40 hours to construct and validate. The performance showed a top decile lift of 202 and a lift of 170 over the top two deciles. This breaks down to a cost per lift point of $1,118 in the top decile and $1,629 in the top two deciles (based on an outsourced cost of $110/hr).
Automating the Process: To improve this process for our client, all 25 models were built in the automated PredictIQ solution using just 60 hours of labor for data preparation and model review, and approximately 5 business days of automated modeling and processing time. The performance estimates had a top decile lift of 207 and a top two deciles lift of 175. Cost per lift point was just $85 and $121, respectively. The PredictIQ models provided a higher average lift than the original models for the top deciles, as well as across the entire file. Additionally, our approach provided a significant development cost savings; the cost per lift point and the total development costs were lower by over 90%. This showed the power and cost/benefit analysis of PredictIQ, even when compared to manual efforts by experienced modelers.
Deployment: Deployment of the models across a series of campaigns over 12 months was projected to generate an additional 76,000 responses and total incremental value of $1.72 million. Time-to-market with fresh models was reduced by over 90% using PredictIQ.
90% savings in development costs for series of 25 predictive models
Improved model lift using PredictIQ in less than 10% of the time
Projected incremental 76,000 responses and $1.72 million customer value