FRAMEWORK : Social Mission
Apply Financial Services Analytical Modeling Process to Help Government Solve Social Problems

Would you like to see our public policymakers use “data science to create social good?”
Three Harvard Business Review authors (Kleinberg, Ludwig, Mullainathan) create “A Guide to Solving Social Problems with Machine Learning” to provide direction. In dealing with trepidation about machine learning, they ask: “How can we maximize the benefits while minimizing the harm?”
To test their concept, the authors applied machine learning to a “dataset of over one million bond court hearings.” As successful as their predictive model was in reducing risk and preventing crime, they still puzzled over what they see as the limiting nature of outcomes.
It’s here – in a broader understanding of outcomes – where there is an opportunity to be even more expansive:
- Preventing further threats to the community
- Optimizing costs for courts
- Fairer criminal justice decisions
- Enhancing chances of improved behavior
Outcomes like these that would improve public policy need to be identified, and then models constructed to predict those various outcomes. Because you must determine what outcome you’re looking to predict—and that outcome must be distinct—it’s our methodology to use a series of models.
The process of defining a model’s objective function – finding drivers to meet the objectives, identifying variables, finding data sources, building and choosing methodology, etc. – is a process that can be applied to any industry, government included.
In lending, for example, a risk model predicts risk; a response model, response; a performance model, performance, etc. Accordingly, in the public policy instance of bail, you can create multiple models for preventing threats to the community, optimizing court costs, fairer decisions, improved behavior, etc.
The authors find it “hard to imagine moving to a world in which algorithms actually make the decisions.” Rather, the authors expect algorithms to be used as “decision aids.”
This is exactly the framework that we in the financial services decision management community have been using for years. Applying data science to public policy will take our art-and-science discipline for reaching the best outcomes to help guide policymakers toward their best decisions.
For using machine learning to create public policy that treats social problems, we should indeed, as the authors advise, “pair caution with hope.” We also should partner with data experts with years of data analytics process, experience and success to help us find our way.