Acquire More Customers and Improve Profitability

Challenge: a leading mortgage lender had unprofitable prospecting direct mail programs in acquiring mortgage loan borrowers. Direct mail is the equivalent of cold calling as it is targeted at people who may well have no interest, and hence has a very high rejection rate. It is thus important to develop an appropriate targeting strategy so that the cost of mailing is more than offset by the profit from the leads that it generates.

Objective: improve direct mail efficiency and acquire more customers profitably.

Approach: Scitics developed a direct mail response prediction model by evaluating hundreds of data attributes from three national credit bureaus: Equifax, Experian & TransUnion, and then identifying influential attributes of responders and non-responders. The predictive model ranked more than 20 million prospects into 10 groups of approximately equal size, i.e., deciles, based on their likelihood of response. The mortgage lender targeted the top four deciles, containing 40% of the most responsive home owners.

Result: the mailing response rate almost doubled, increased from 0.28% to 0.55%, and the profit increased to $3 million, up 186% from the prior best mailing campaign.

Marketing Mix Modeling for Optimizing Media Spend

Challenge: a major retail chain spends millions of dollars every year on various marketing and advertising programs via different channels including TV, radio, circular, direct mail, email, paid search, and online display. It is top of mind for marketers to understand the effectiveness and ROI of each program. They hired a top strategic management consulting firm to perform a marketing mix modeling analysis, but the lack of data granularity raised doubt on accuracy and the results were not actionable.

Objective: accurately measure the effectiveness of each media in terms of its contribution to sales volume and ROI, and adopt the learnings to adjust marketing strategies and to optimize the marketing plan.

Approach: Scitics partnered with the retail chain’s direct marketing agency to create a series of customer and prospect statistical models designed to determine the impact of each media type and spend amount on sales. These models evaluated historical customer level sales and media spend data and in so doing created modeling algorithms that serve to quantify the sales impact of each media and allocate sales either to no-media (i.e., base sales) or to one or more media channels.

Result: the granular models were implemented with weekly updates to show attributed sales by media channel, and marketers took actions accordingly to generate the greatest sales growth and maximize profits.

Sustain Membership Renewal and Empower Members

Challenge: a nationwide non-profit professional organization experienced decreased membership renewal rates. They provide professional publications, educational, networking events, and strive to stay as the most relevant force and voice shaping their members’ professional fields.

Objective: identify root causes of member attrition and provide actionable recommendations on member retention strategies.

Approach: Scitics took a member-centric approach and examined holistic member data points including members’ professional interests, industry, age, years of experience, publication subscriptions, source of membership fee, events orders and attendances. An immediate finding showed new members had a much lower renewal rate. Further drill down analysis revealed new members were not fully educated on the membership benefits and a disconnection existed between member interests and event topics. On the existing member segment, key distinctive characteristics were discovered to predict likelihood of renewal. In addition, a time series analysis provided insights on the timing of renewal.

Result: the organization acted upon the analysis findings with a better understanding of their members’ needs. They improved new member follow up strategy and made necessary changes in event programming. They were in a better position enabling members to have a successful long journey with the member community.

Guide Patients with Data on A Path to Wellness

Challenge: a leading healthcare service provider to underserved communities saw increased hospitalizations and insurance claim rates from patients with chronic conditions.

Objective: identify risk factors to proactively guide patient’s care plan and engage patients with the right information to better manage their existing health issues.

Approach: we looked at the patient database and identified the top three chronic conditions in the underserved communities: diabetes, hypertension and cardiovascular disease. We then developed a prediction model for each and applied a scoring algorithm, which categorized patients as high, medium or low risks based on lab and vital sign values. That information was shared with nurses and case managers to proactively guide each patient’s care plan and help patients work on self-management goals. The patients classified as high risks were scheduled to see their providers on a monthly basis. With the interactive data exploration and the ability to make sense of complex data, the healthcare provider was able to connect the patients to the ideal specialists, counselors and advocates based on their individual needs. The patients were also engaged and empowered to overcome or better manage their health and wellness challenges.

Result: since the healthcare service provider adopted data analytics in their patient care and support, hospitalizations were reduced by 40%.

Use Predictive Analytics to Detect Fraud

Challenge: mortgage fraud was a pervasive problem and costed a private mortgage insurance company millions of dollars. The number of fraud cases involving insurance policies increased more than two-fold, as a result, claim rate increased, and so did the work load of the loss mitigation department.

Objective: develop a mortgage fraud prediction model to save money and time.

Approach: using predictive analytics and machine learning, we developed a fraud prediction model and identified over a million dollars in fraudulent and questionable insurance claims. The model classified 10% of mortgage insurance claims as high risk of fraud. By investigating the high risk group, the loss mitigation department was able to identify the large majority of fraudulent loans.

Result: the fraud prediction model helped the mortgage insurance company save money, time and resources. It also allowed loan officers to make an informed decision and reject high risk applications to prevent future losses at the time of origination.