Zipline Obstacle course 2.”

Social impact bonds—also known as “pay-for-success” contracts, since they are not true bonds, but rather performance-based contracts—aim to help governments experiment with new ways of providing social services. The basic concept is this: Private investors provide the original funding, and the government only pays if the program achieves targeted outcomes. This lowers the cost of testing new policy ideas. In exchange for this ability to test new policy ideas that may or may not work, the government pays the private investors a premium if the experimental program reaches or exceeds the performance targets. In short, private investors finance experimentation and innovation.

The government naturally funds R&D on its own. But while government funding for science R&D is $176.8 billion in the FY 2018 approved budget, politicians are highly reluctant to finance R&D to improve public-sector operations. As social impact bond investor Bridges Fund Management explains in a 2016 report, traditional government fee-for-service contracts favor proven interventions, and providers have neither the flexibility to experiment nor the incentive to exceed expected results.1 As another example of public sector risk aversion, a 2015 article from the Kauffman Foundation reports that in the area of public procurement, “of the top 25 firms in the United States winning government contracts, the youngest was founded in 1969.”

The Solyndra scandal during the Obama administration illustrates some of the drivers of public sector risk aversion. Solyndra, you may recall, was a startup solar panel manufacturer that received a $535 million government guarantee in 2009 and went bankrupt in 2011. In sum, the Department of Energy’s Loan Programs Office made $30 billion of loans to 30 companies with varying success as it sought to develop the renewable energy industry as a whole.2

Overall, the program succeeded beyond expectations. In Slate, Daniel Gross writes, “From nothing before the loan guarantees, utility-scale solar now…is a genuine growth industry, financed largely by private funds.” But the public mainly remembers the failure of Solyndra, while few credit the federal government with being a venture capitalist that catalyzed the solar industry.

Given the political costs of risking public capital, offloading the risk of public sector innovation via social impact bonds is highly attractive. According to Bridges Fund Management, a social impact bond gives providers far greater flexibility to adapt their programs to achieve desired outcomes.3

To date, social impact bonds have generated inconclusive results. We should expect more mishaps and crashes—like Solyndra—that hone the design of financial tools that optimize public-private partnerships on social problems. Still, the potential of public sector R&D to improve outcomes is great. This article examines how this potential could be realized.

Stumbling out of the Starting Gate

Limited Success. The world’s first social impact bond, launched in March 2010 by Social Finance and the UK Ministry of Justice, aimed to reduce reoffending, or recidivism, among prisoners released from Peterborough prison. Primarily foundations and charitable trusts invested £5 million to provide intensive support to 3,000 prisoners over seven years. The greater the decline in recidivism, the more investors could receive, up to £8 million.4

The Peterborough program reduced recidivism by nine percent—1.5 percent above the Ministry of Justice target. The 17 investors in the social impact bond received a 3-percent return, as well as return of capital.5 But there are caveats. As is common practice to date, the government was not willing to offer much of a premium to investors, and the maximum possible return was 13 percent. Philanthropic support from the Big Lottery Fund to the tune of £6.3 million6 was required to make this offer. Basically, the government relied on philanthropy to pay upside returns, rather than investing in R&D itself. This works, of course, but it also means the project was more like a grant-funded initiative than true investor-financed R&D.

Failure. New York City launched the first US social impact bond in 2012 to reduce recidivism among 3,000 male prisoners at Rikers Island aged 16–18 with a three-year intervention.7 The Urban Investment Group at Goldman Sachs provided a $9.6 million loan and Bloomberg Philanthropies provided a $7.2 million guarantee to fund the project. The decline in recidivism did not meet the threshold 8.5 percent target; as a result, investors lost $2.4 million (the remainder being covered by the Bloomberg guarantee).8 

Why Getting This Right Matters—Saving Billions of Dollars and Improving Millions of Lives

It would be all well and good to dismiss social impact bonds, if only the problems that the public sector needs to solve were not so pressing. For example, the direct annual cost of incarcerations in the US is $80 billion.9 But as a new study, The Economic Burden of Incarceration in the US, points out, when one includes the costs to the individuals jailed and to their families and communities, the total cost to society exceeds $1 trillion.10 Reducing reoffending, or recidivism, would have enormous societal benefits. For example, DC Central Kitchen is a nonprofit that reduces the recidivism rate of the people it trains to six percent, far lower than the national average of 45 percent.11 If programs like this existed nationwide, the government would save billions of dollars, and millions12 of former prisoners and their family members would live more rewarding lives. The need for systems change is obvious. And finding a way to facilitate government R&D that can push forward such systemic changes is critical. 

Stumbling Blocks

The biggest impediment to scaling social impact bonds is governments seeking to offload the risk of social innovation without offering sufficient return. Social impact bonds are popular with government officials insofar as they avoid upfront costs, but a common reaction is, “What do you mean the government has to pay if the intervention works?”

So, what does government do? Well, it tries to have its cake and eat it too, and so can end up with a small cupcake to show for it. For example, investors in social impact bonds are often limited to foundations, charitable trusts, individuals, and entities that are tolerant of low financial returns. Social impact bonds therefore can look a lot like grants—and are hardly investor-financed R&D vehicles. Many hands are employed to make the project look like an investment, but it is an odd sort of an investment where return and risk are minimized by government caps on one end and grants on the other.

Bringing the risk-adjusted returns of social impact bonds closer to market rates would be a precondition for large capital pools, like public pension funds and sovereign wealth funds, to invest. Naturally, improving returns would require public agencies to pay more when interventions are successful.

Other stumbling blocks to scaling social impact bonds are structural and, as detailed in the next section, a number are already being addressed.

First, standardization is critical to scalability, and social impact bonds lack standardization. As discussed, despite their name, social impact bonds are not bonds. From a legal perspective, social impact bonds are structured notes,13 commercial loans with contingent payouts, or multi-party contracts that are privately negotiated.14 The opaque nature of these agreements limits scalability.

Second, there are data and analytical impediments. In order to know if a program improves outcomes, you need to establish baseline rates and ensure that impact data is consistently reported.15 Analytical challenges range from variation in how outcomes are assessed to the difficulty of determining the causal relationship between the reported outcomes of the program and the social intervention.

Third, social impact bond scalability is limited by the size of government contracts. So far, in the US, social impact bonds have been issued at the local or state, rather than the federal level.

Lastly, social impact bonds have been offered only as private placements, which are limited to accredited and institutional investors.16 This private status limits liquidity, therefore increasing risk. 

Building Blocks

Committed organizations are already removing the secondary stumbling blocks or reconceiving them as building blocks.

Return profile. One firm, NPX, is working on an impact security that is a hybrid between a social impact bond and standard debt. Investors would get a blended return that includes a partial guaranteed payment stream and a royalty or reward payment, contingent on the success of the program. Investors are therefore guaranteed some return of principal and interest while still bearing enough risk to catalyze greater public-sector R&D.

The potential size of government contracts. Targeting mainstream social issues that concern large population groups of course facilitates greater impact.17 Social Finance Israel is working on a two-year social impact bond to help prevent type-2 diabetes.18 Since 422 million people globally, or roughly 8.5 percent of the world’s population, has diabetes,19 success here could have extraordinary impact.

Similarly, UBS Optimus Foundation, USAID, Population Services International, Palladium, Merck for Mothers, and a nonprofit promoted by the Indian state-owned HLL Lifecare Limited20 are working with the government of Rajasthan, a state in India, on Utkrisht Impact Bond, the world’s first development health impact bond.21 Development impact bonds are social impact bonds focused on international development.22 These development impact bonds are focused on maternal and fetal health—the only issues more universal and scalable than death and taxes! This one bond has the potential to improve care for up to 600,000 expecting mothers during delivery.23 The number of lives impacted is nearly as large as the 738,000 estimated lives impacted by the 108 other social impact bonds issued since 2010.24

Sources of support. Public sector, private foundation, and private university support continue to catalyze the development of social impact bonds. Examples of government and supranational support includes the state of Massachusetts’s $50 million authorization for pay-for-success initiatives25 and the Inter-American Development Bank’s $5.3 million facility to fund social impact bonds in Latin America and the Caribbean.26 In philanthropy, the Rockefeller Foundation has supported the launch of the Social Impact Bond Technical Assistance Lab at Harvard Kennedy School,27 which has evolved into the Government Performance Lab.28 Such support is critical to build capacity among service providers, assessors, investors, and government. 


Going forward, we can expect to see both crashes and meaningful progress that generates usable finance vehicles. Encouragingly, the Utah High Quality Preschool Program’s pay-for-success phase I results are promising.29 At the same time, it is worth noting that encouraging public sector R&D is itself a vital task. By definition, with R&D, some interventions will fail, while others will succeed. But worse is to underinvest in R&D as a whole, which, alas, the public sector has been guilty of for some time now.

The $392 million placed in social impact bonds represents a drop in the ocean, whether measured in terms of government budgets or the world’s $118 trillion capital market.30 US impact bond issuance stands at $211 million across 20 deals and is less than 0.02 percent of the $1.7 trillion in total annual nonprofit revenue and contributions.31

The field, in short, remains in its infancy. While the future is uncertain, there is reason to believe that social impact bonds may help generate increased investment in the development of improved service delivery systems going forward. What is indisputable is that the present level of public sector R&D is grossly inadequate. Figuring out how to change that should be of great interest to all of us.


  15., page 56
  17., page 67
  24., page 25.
  30. Accessed July 31, 2018.

This article has been altered from its initial publication to update note #30 at the author’s request.