Versium, which earlier this year launched its Predictive FraudScore solution (covered here) today releases its Predictive GivingScore solution, designed to help charitable institutions and political organizations better predict who is likely to donate, be a repeat donator, or make the more significant contribution. PredictiveGiving Score is the latest of the company’s predictive Score products, which also include churn, social influencer and shopper scoring – and it’s by no means the last.
It was built with Microsoft Azure Machine Learning, a managed cloud service for building predictive analytics solutions publicly unveiled just a short time ago. CEO Chris Matty says that platform is an aid to Versium in rapidly building its new score solutions. (Just shy of ten Versium scoring products are currently in use or in development.) Azure ML, Matty notes, contains dozens of machine learning algorithms and mathematical computation models it leverages to easily and effectively experiment, create and tune models to get the highest accuracy in predictive scoring solutions.
“Once we have a score built it just takes little tuning. But when we are building a new score we need to look at some different models and see what works better,” he says. “We want to move quickly by evaluating the different models, and we can visualize very easily the process of building the predictive model.”
Versium also is making it possible to include, via API, its LifeData within Microsoft’s machine learning environment. LifeData -- which also contributes to the GivingScore solution to better analyze an organization’s existing donor base and identify additional generous donors -- is now up to about 7 billion email addresses among its billions of data points on businesses and consumers that covers everything from purchase interests to education levels, and upon which Versium draws in predicting behaviors. Since The Semantic Web Blog last reported on Versium in February, it has added to LifeData between 400 and 500 columns of additional data – energy consumption, political affiliation, and so on.
Matty believes there are many opportunities to patch Versium’s LifeData into other solutions that want to build non-competitive predictive analytics solutions on top of their computer models. “We see a lot of folks looking to bring in external data and our data is a powerful predictor,” he says. “The real interesting thing is that we see literally upwards of 30 percent or more greater prediction accuracy in computer modeling when you include our LifeData….It’s a natural fit for us to work with others.”
Historical LifeData attributes used to train Predictive GivingScore encompassed things like donation frequencies and levels, high-value donors, non-response donors, and so on, to help charities and other organizaitons narrow in to target for contributions donors who have similar characteristics to those who have donated. The solution was driven, Matty says, by the fact that charities’ marketing efforts tend to not be very efficient and to realize very low returns -- .1 percent, in the case of one organization it worked with.
“We say that by leveraging LifeData we can determine who has a higher propensity to react or respond in a certain way,” he says. “We do the same thing with other enterprises on other user-propensity type models. Within LifeData [for Predictive GivingScore] we found phenomenal data signals as it relates to characteristics around people who tend to be more likely to donate and to do it more frequently and at higher levels.” The Millionair Club Charity, an organization that addresses homelessness and Treehouse, a nonprofit organization serving youth in foster care, are both currently using Predictive GivingScore.