Technological advances in the last five years have made it possible for anyone with access to the Internet to use powerful data analysis programs. Unfortunately, although even cash-strapped nonprofits can now easily run models using big data and develop algorithms to understand how many clients they are affecting and in what ways, few do so. For-profit companies have already harnessed the power of client data, especially data from online buying habits. The New York Times Magazine reported in February that Target stores knew one of its female customers was pregnant before the customer’s own father knew. How? Her buying habits. Target’s algorithms tracked the woman’s purchases and matched them against its enormous consumer data set, finding a match between her habits and those of other pregnant consumers.

Having hard numbers means that nonprofits can no longer ignore the math of social change. The fight for social/environmental justice, regardless of the form the battle takes, is deeply imbued with emotion: we want to hear about the life changed, the forest preserved, the child educated, the disease eradicated, the war averted. And to be sure, whether one donates to a nonprofit, works for an organization that seeks to better the world, or volunteers for a cause, one’s time, money, sweat, and tears will almost always achieve a positive impact, however infinitesimal.

However, this new math to social change cuts through emotion and gets at a simple question: Does a particular intervention actually achieve the desired impact ? No matter how many people you serve, only a certain percentage of those served will truly benefit from the service. This realization shouldn’t be surprising, for not everyone wants to, or is able to, or is ready to better his or her own life. External factors—illness, an economic downturn, a family emergency—can prevent even the most dedicated of people from moving out of poverty. Yet when nonprofits work so hard to serve people, it is painful to realize that only a subset of clients will truly benefit from the nonprofit’s work. As we have begun to assimilate the painful math of social change, we have come up with four possible ways of looking at the numbers:

  1. We can say that, even if only 20 out of every 100 people served will truly benefit, it’s still worth it because that’s the only way of reaching those 20 people.
  2. We can look at the data and try to determine the characteristics of those 20 people and focus exclusively on them.
  3. We can assess the effectiveness of our intervention—in our case, microfinance—and consider tweaking or changing it.
  4. We can focus on broader forces that affect our clients. For instance, we can advocate for welfare reform or policies that will strengthen the middle class or improve education. Under this model, the focus isn’t on serving the poor, one person or family at a time, but rather tackling the structural barriers to ending poverty.
  5. A combination of the above.

Again, these five options may seem obvious, but there are numerous challenges. 

With option one, it’s hard enough to raise money for a nonprofit, and it’s even harder to make the case that it takes a lot of money to reach those that will actually benefit. For instance, imagine an organization that serves 1,000 people per year, at a cost of $100 per client. Let’s further assume that 20 percent, or 200 people, truly take advantage of the organization’s products or services. What this means is that the true cost of changing a life is the total cost of serving the 1,000 people ($100,000) divided by the 200 people that benefit ($500). In other words, the cost of truly changing a life is five times the cost of serving a client. In short, that’s a hard sell for funders, especially considering that funding from individuals, foundations, and other grant makers is hard to come by.

Option two is the highly compelling: instead of reaching a 1,000 people to find the 200 people you really want to reach, why not figure out how to target those 200 clients to begin with? This is where data mining comes in. We can ask ourselves, what are the characteristics of the clients that most benefit from our services? How are we reaching them? Is there a particular product or service that has the most impact? And so on.

Option three is an essential question to ask. What if tweaking our intervention can dramatically increase impact? For instance, we may find that, though we over 10 products and services, 90 percent of our impact comes from three of them. Conversely, we might find, through survey work, that if we added a health module to our financial coaching, we’d be able to meet an unmet need. Again, data analysis serves as the driver for organization-wide self-criticism and adjustment.

Option four is the one that perhaps has the most potential to affect significant change on a societal level. After all, if we look at the trends that have brought the most people out of poverty in the 20th century, it’s arguably not the work of social services agencies but of macroeconomic trends such as globalization, the information technology revolution, industrialization, and the green revolution. The problem, of course, is that it is exceedingly difficult and time consuming to affect this sort of change: after years of lobbying and advocating for a single policy change, powerful interests can still block the initiative, leaving one with nothing to show for so much effort.

Option five is, ultimately, the best solution. While we work to advocate for macro changes that have the potential to change society for the better, we must recognize that there are human beings in need of support today. And while we must always do our best to rigorously look at data, tweak our business model and theory of change, and recognize the math of social change, we have a moral obligation to continue doing all we can to serve humanity. Even if 20 percent of our clients truly benefit, that’s 20 percent more people moving out of poverty than had we not launched our organization. At Capital Good Fund, we have no doubt that as we continue to grow, not only will we get better at identifying those we really want to reach, but we will increase the percentage of clients whose lives are changed. And finally, we will have to maximize the extent to which we become aware of, and advocate for, policies, laws, and other structural changes that remove the barriers to creating a more just and equitable society.

So yes, the math of social change is difficult to stomach, and we cannot move forward in the fight for social justice without the best tools—data mining, predictive analytics, behavioral psychology, and the like. That said, not matter how advanced our tools become, social change work will always be as much poetry as it is math, and numbers can only say so much about how a human being is impacted by any intervention.

Foundations have been using big data for some time, namely REDF and the Robin Hood Foundation (See the SSIR article about calculated impact). However, these foundations mainly look at expected return on investment, rather than at predictive modeling.

Nonprofits such as UNICEF have used tools to predict donor behavior, but not to predict client behavior. According to IBM: “For instance, by using Predictive Donor Management, UNICEF, the child rights organization of the United Nations, better mapped out the donation behavior of their contributors with greater accuracy and defined clear-cut segments and profiles. Now, they’re able to better target direct mail campaigns and produce a substantially improved yield.”

Government agencies have had access to big data for quite some time, and have started to successfully use it to predict citizen behavior. The CDC, Department of Defense, and tax agencies have all used predictive analytics.

A few nonprofits have great potential to use more predictive data. Innovations for Poverty Action uses randomized control experiments to measure the impact of poverty reduction programs. However, randomized controls only work in controlled environments, and are not as useful for predictive analytics. Harlem Children’s Zone has measured the effect of 50 services and programs, providing useful feedback about program effectiveness. This effort was time intensive, and also didn’t predict which of the 50 services was most helpful. Environmental nonprofits also have started to use predictive modeling to understand the effects of climate change and other natural disasters. Again, this is slightly different than our proposed use, since the predictions are on global ecological behaviors rather than human behavior.

At first glance, it may seem that though these techniques are naturally applicable to microfinance, other fields would have a harder time taking advantage of them. A deeper look, however, indicates otherwise. A job training program, for instance, can use data mining to compare information collected at intake with the ultimate goal—clients being placed in jobs. Armed with this data, the program can better target its services and tweak its approach to increase its success rates. In other words, the fact that algorithms and lending are peas in a pod does not mean that they cannot be used in all manner of contexts, provided that there is a clear outcome being tracked and a good data set being used for comparison.