Project

Brief Description: I work for a technical support organization. We handle incoming client phone calls and emails, working with our customers to help resolve issues and concerns they have with our software. Our number one client facing metric is client satisfaction (rated on a scale of 1-5). We also utilize the Net Promoter Score (NPS – scale of 1-10) to measure client sentiment. Clients that are highly satisfied with a support organization, and have high NPS with a business are more likely to not only continue on as clients, but more likely to expand business and invest more money with a given organization.

Business Objective: What I propose is to define a model that can help to predict a client’s satisfaction with a given case. By examining this information real time, we can look to cases that are flagged as potential for negative client satisfaction and increase our attention on those cases. By doing this, we can not only improve the overall satisfaction of our clients, but we can also look to prevent case escalations, and ensure continued business with our clients.

Modeling Task: The model will examine all completed surveys for 10,000 cases and will examine them to determine potential satisfaction and/or NPS scores.

Dataset: 10,000 cases that resulted in a completed client satisfaction survey, anonymized to remove client specific details. Source will be from Salesforce.Com internal systems.

Number of Records: 10,000

Training data set: 5,000 cases

Validation data: 5,000 cases

Potential target/response/dependent variable: Transaction Overall Satisfaction (1-5)

· 1 = Very Dissatisfied

· 2 = Dissatisfied

· 3 = Neither Satisfied or Dissatisfied

· 4 = Satisfied

· 5 = Very Satisfied

Potential predictor/explanatory/independent variables: Product, Case Time to Resolve, Severity of the case (critical, high, medium, low),

Data Mining Techniques: Logistic regression

Data Mining Software: SAS Studio, SAS Enterprise Miner