Success Stories

Recent Successes

Awards, Presentations

These are some of the awards I've won.

Recommendations

Here are some recommendations I've received.

Resume

Download a copy of my resume.

Recent Successes (detail)

Credit card acquisition campaign results driven by accurate response models

In an effort to maximize results of credit card acquisition campaigns, I developed a predictive model for prospects that used a decision tree to partition the data into nodes, and then used node membership as a variable to be included with others in a logistic regression model. This hybrid logistic regression/decision tree model had KS=0.49, and was used to prioritize mailings for multiple direct mail campaigns of 1 MM - 2 MM mail pieces each. Model lift exceeded expectations as the net response rate for prospects tripled (from .06% to .17%) in 3 months. [top]

Best fitting credit card model enhances customer experience

In order to both increase fee-based revenue and ensure that customer needs were being met as well as possible, I developed a branch channel based best fitting credit card product analog model (i.e., look-alike model) for prescreened credit customers using a classification tree. This model received much positive feedback from bankers trying to find the right product for customers, and I received a PNC Spotlight Award. [top]

Predictive modeling allows bank to dramatically improve marketing results

In response to the need to develop and productionize a set of predictive models for marketing, I imagined and created an overall predictive model architecture allowing multiple models to work together in a modular fashion, and I proceeded to develop the initial 7 predictive models needed, including models for checking, savings, and early access. This resulted in $2.5 million in net revenue (520% ROI) in just the first month. [top]

Checking acquisition campaign results driven by an accurate propensity model

In an effort to maximize results of a checking acquisition campaign, I developed a predictive model that first used a decision tree to partition the data into nodes, and then used node membership as a variable to be included with others in a logistic regression model. This hybrid logistic regression/decision tree model for propensity to open checking accounts had an accuracy of 87%, and is expected to earn $14.7 million over 4 years. In 1 year, the top 3 deciles generated per household 4 times as many accounts, and 2 times the balances, as the lower 7 deciles. [top]

Best fitting checking product model drives increased revenue and enhances customer experience

Because of the need to both increase fee-based revenue and ensure that customer needs were being met as well as possible, I developed a best fitting checking product analog model (i.e., look-alike model) using multinomial logistic regression. We identified 2.5 million customers to score. This model was successfully used in several different campaigns. [top]

Money market savings model drives greater than expected balance growth

To increase our marketing ROI, on my own initiative, and with the support of my manager, I developed a money market savings model yielding $32 million in new deposits versus $21 million planned. This model was used with much success for related campaigns for over one year. [top]

Customer scoring and segmentation used to retain high balance customers

After my employer acquired a bank with $11 billion in deposits and 270,000 households, a concerted effort needed to be made to retain high deposit balance customers. I coordinated the implementation of a customer scoring algorithm and segmentation strategy, and I helped model and validate the application used to manage the resulting customer calling campaign. The campaign generated significant incremental deposit balance growth, and I prepared a detailed analysis of the results for senior management. [top]

Telemarketing analytics finds unexpected opportunities

As a result of data being in separate silos, the analytics team was unable to properly identify customers that were contacted via the 15.5 million telemarketing (TM) calls our company made annually. This led to an inability to assess the true impact of TM upon our campaigns. I worked with a cross-functional team to create a short-term and a long-term solution to this problem. In the short-term, I worked with our I.S. team to set up a regular TM data extract process. I then developed code to transform this data into a format usable by our analytics team. We were then able to find remarkable incremental response with telemarketing in certain circumstances. This information was incorporated into our regular analytics and modeling process and generated even more incremental gains. [top]