Predictive Modeling and Analytics

Among the top philanthropy buzzwords are “data” and “data scientist”. Certainly the phrases “predictive modeling” and “data analytics” have made their way into the fundraising lexicon. Predictive modeling can have a significant impact as part of a data-driven annual fund in helping to take some of the unknown variables out of the equation. To help explain modeling and analytics further, we spoke with WealthEngine’s Vice President of Analytics, Cong Qian.

Q. How would you describe Predictive Modeling or Predictive Analytics?

Cong: Essentially, predictive modeling is trying to summarize what you have accumulated from the past and apply that knowledge to predict future events. In other words, assuming future events will be relatively similar to the past, then let’s see what the future could be, so if this happened then these things are also likely to happen.

Q. What is the process for creating or building a predictive model?

Cong: Usually, a model is trying to answer a specific question, such as “Which of my donors are most likely to give more than $250?” or “Which of my non-donors are most likely to make an annual fund gift?” Statisticians then try to identify the most predictive data elements that can be grouped into a formula to answer the question. At WealthEngine we leverage not only our client’s important data such as giving history, demographic information, etc., but also data from over 60 sources that we are able to find through our wealth screening process, including Propensity To GiveTM codes (P2G), Estimated Giving Capacity, lifestyle and business information and other additional demographic information. All of these data points are evaluated to determine their relationship to each other and their impact upon the desired behavior.

We then build the model using the appropriate statistical methodology with the data set known as the “training sample”; we also validate the model, to ensure its accuracy and consistency against a subset of the data tested.

Q: How, then, is a model helpful in the context of an annual fund?

Cong: Once the model has been built, the data pool is divided into segments based upon their statistical likelihood to perform the desired behavior. These segments, or “deciles”, are ranked from highest to lowest in terms of likelihood. So, for example, the people in an Annual Fund Model Decile 1 would be most likely to make an annual fund gift at the established gift threshold. The higher the model score, the more likely they are to perform the way we hope they will – make a gift to the annual fund. Figures 1 and 2 represent the percentages of model likelihood in each decile.

Figure 1: Percent of major donors in each model decile compared to average or random sampling (represented by gray line)

Figure 2: Cumulative gains of major donors captured by decile compared to random distribution (represented by green line)

Q. How can this type of information help an organization’s annual fund?

Cong: Modeling is designed to take some of the guess work out of fundraising and therefore target resources more efficiently. By focusing the annual fund on the prospects with a higher likelihood, the response to the solicitation should improve and the return on investment (ROI) should be stronger as well.

Q. Are there other types of modeling?

Cong: There are. The typical models we’ve discussed ask the data a specific question and then predict outcomes based upon the contributing variables. Other types of models are more of an overall analysis of data behavior or a more customized segmentation methodology. At WealthEngine we utilize what we call a Donor 360° model to analyze donor behavior with cluster analysis and group similar types of donors into clusters. We also create what’s called a Look-Alike Model, which we can use to gain insight about the best donors. We measure and evaluate an array of variables from giving history, wealth and capacity ratings, demographic and life styles data, etc., and then cross-reference significant variables onto other donors or prospects to identify potentials that are likely to behave as the best donors for fundraising outreach or acquisition. Modeling can be very flexible and extremely useful in the process of data mining.