illustration of a cube with many facets featuring images of money and government buildings (Illustration by Eddie Guy) 

On a Tuesday in mid-February 2016, the state of South Carolina agreed—for the first time in its history—to pay for fewer preterm births.

The state had regularly funded programs in hopes of lowering preterm birth numbers. This was different. The deal would significantly expand a program that sends nurses to the homes of first-time, low-income parents, from pregnancy through the first years of parenthood. On a stage at the Capitol in Columbia, Governor Nikki Haley announced that this time, the state would pay only if the program actually fulfilled its intended purpose.

The same week, about 700 miles north, Governor Dannel Malloy announced an effort to help Connecticut families stay out of the child-welfare system, paid for—like South Carolina’s new program—on the basis of outcomes. The Connecticut program paired parents struggling with a substance-use disorder with a treatment-and-support team that would come to their homes.

Two projects we had designed in two very different states—one red and one blue—launched in one week. We could be forgiven for thinking that if it worked in Columbia and Hartford, it could work anywhere.

This was a heady time for Social Finance, the nonprofit we work for. We were founded to expand economic opportunity by improving systems and putting capital to work in new ways. A niche financial tool that we and a handful of peers worked on had, rather quickly, exploded into a public debate.

Governments spend trillions a year on social programs and rarely learn what the money buys. We were trying something different: setting them up to pay for results. Doing so meant that parties had to state clearly and in advance what they were trying to achieve, and then find out whether they did achieve it. This idea—a contract in which government pays not for a service (e.g., prenatal home visits) but for an outcome (e.g., fewer preterm births)—has a name. Several names, actually, which is part of the problem: Pay for Success. Outcomes-based finance. Social impact bonds.

The instrument that made these experiments possible arrived a few years earlier, in 2010, when our colleagues at Social Finance UK launched the world’s first social impact bond, or SIB, at a prison in Peterborough, England. Investors funded reentry services for men serving short sentences; the government promised to repay them, with a return, but only if reoffending fell. People described the concept as elegant; people rarely describe anything to do with government as elegant, so this was refreshing.

SIBs crossed the Atlantic quickly. The New York Times and Washington Post covered them earnestly. The nation’s largest philanthropies seeded pilots, drawn in by the potential to simultaneously build evidence and public capacity. We were invited to speak on Wall Street, at the White House, and at the Vatican. Governor Deval Patrick of Massachusetts championed a Social Innovation Financing Trust Fund; Goldman Sachs wrote checks; policy think tank Urban Institute authored articles and guides. Harvard University opened a SIB lab. Business and policy schools wrote cases. Statehouses from California to Delaware passed new legislation to help jump-start a financial instrument that, in most of those states, did not yet exist. In 2018, a Republican Congress and a Republican president put $100 million behind outcomes funding in the Social Impact Partnerships to Pay for Results Act, a bipartisan bill first developed under the Obama administration. Our organization structured many of the major American SIBs.

Beneath the hype, on the ground, things that we had genuinely not seen before were happening. Nonprofits were getting multiyear contracts that covered what the work actually cost, which meant their executive directors could spend time running programs instead of raising money for them. Elected officials started using SIB-related terms like “counterfactual” and “effect size.” Three departments in Ventura County got together and worked through a reasonable compromise on how to price a single avoided jail night.

Even some of the hard parts represented good practices we had rarely seen in government. The social sector argues constantly about how much money to spend on housing or job training or home visiting. It almost never works backward from the result, deciding first what a good outcome is worth. Doing so changes how to think about allocating resources, away from what’s cheapest and toward what gets the most impact per dollar. Of course, not everything that matters can be measured. We grappled with how to define success when it didn’t live in an administrative database. But the grappling was part of the point: Instead of hoping a program worked, we put stakes in the ground around some of the most important indicators, knowing they couldn’t tell us everything.

These conversations brought together people who should have been collaborating all along but never had: a budget analyst, a probation officer, the public defender, a foundation program officer, the executive director of a small nonprofit, a banker, an academic evaluator, a community advocate. For a few years, we felt as if we had a real path to more of the government we wanted.

Then the enthusiasm evaporated.

In the years following, an underspecified verdict hardened into the conventional wisdom that SIBs had, simply, failed. New deals in the Brookings database peaked in 2020, at dozens a year, and dwindled to single digits within five years. The conferences thinned. Google the phrase “social impact bond” today, and much of what you’ll find has the slightly dusty feel of a trend piece about Google Glass.

What happened? This story fills the gap between the hype of a new financial tool in the wake of the Great Recession and its slow transmutation into something more boring yet more powerful.

The financial instrument was, ironically, never the point. SIBs were a loud and peculiar response to something much larger and much older: a long push to make public money answer for results, rather than activity. The deals thinned out, but the push didn’t; it accelerated, albeit more quietly, within the parts of government no one writes headlines about. It also spread beyond government, to employers, philanthropies, investors, and students. And today, as AI and new technologies transform the world, the ground is again shifting, opening new opportunities to focus on outcomes across sectors of the economy.

Learning, the Expensive Way

In a typical government contract, a nonprofit runs a program, submits the receipts, and gets paid (or perhaps underpaid) by government for the cost.

A SIB works differently. An outcomes buyer—usually, we’re talking about a government agency—defines a set of outcomes it will pay for, and what each is worth. It hires an independent evaluator to specify which metrics to use and the right counterfactual against which to judge effectiveness: What would have happened to the same people if the program had not existed? Outcomes buyers and providers negotiate a contract. The nonprofit delivers, the independent evaluator determines the results, and then the outcomes buyer pays proportionately to how effective the program is.

For nonprofits, this approach can be appealing because it lets them demonstrate their true value and often gives them flexibility in how to achieve good outcomes. But it also poses two problems. The first is cash flow: The median nonprofit has less than three months of operating reserves, so getting paid only after it achieves positive outcomes isn’t feasible. The second is risk: Even a small underperformance might force them out of business.

We’ve spoken to thousands of people about outcomes funding, and one of the questions we get most about any given project is “Did it work?” We never really know how to answer. In fact, we’re not even certain that the people asking know exactly what they’re trying to ascertain.

Part of the reason SIBs got so much attention at first is that we found a creative way to solve those problems. Investors routinely finance working-capital gaps, and they are skilled at taking on risk. In a SIB, investors put up the money for these deals. Such investors include donor-advised funds, foundations (philanthropies such as Arnold Ventures or The Rockefeller Foundation), community-development financial institutions (such as Reinvestment Fund or Nonprofit Finance Fund), and other impact-focused investors (such as BNP Paribas or Northern Trust). If the programs are effective and outcomes buyers pay, the investors get their money back and potentially make a return; if not, they absorb the loss.

On paper, it looked simple and elegant. Pop the hood on any SIB, though, and you find a V12 engine on a lawn mower.

Take, for example, Veterans CARE (Coordinated Approach to Recovery and Employment), a SIB we launched in October 2018 with the Department of Veterans Affairs, the Commonwealth of Massachusetts, the City of Boston, and partners in New York. The plan was to serve 480 veterans with service-connected post-traumatic stress disorder (PTSD) across four VA medical centers, helping them find and keep competitive jobs through a stunningly well-evidenced model called Individual Placement and Support.

It took four full years of design and wrangling to get it off the ground. The contracting mechanism it required was a postmodern work of art. It had four outcomes buyers, five philanthropic investors, an evaluator, a fiscal agent, a project manager, and an organization delivering services. The project’s legal structure won an award from the Grunin Center at NYU Law. When it finally launched, the human results were everything the research promised. Veterans in the program found jobs in banking, healthcare, education, and government. One landed a position at the White House, another with Major League Baseball, another at Memorial Sloan Kettering. One booked an acting role on a prime-time television show. PTSD symptom ratings, tracked over 18 months, stabilized or improved for nearly half of participants; employment was not a competing priority with mental health treatment but part of it.

Yet the project nearly collapsed. Two things went wrong, one slowly and one quickly.

The slow issue was enrollment. The model assumed veterans would flow steadily from VA clinical teams into the program. They didn’t—not because the demand wasn’t there, but because referral relationships are built person by person. One employment specialist told us it took her six months of showing up to clinical team meetings, case by case, before the mental health staff trusted her enough to send veterans her way. Six months is an eternity in a project that’s holding investors’ money and whose costs—staff salaries, site infrastructure—are largely fixed, regardless of how many veterans walk through the door.

By early 2020, Cohort 1 stood at 282 veterans, against a target of 480. Their outcomes were strong—58 percent employed, average earnings over $15,000 in a population whose average stretch of unemployment had been more than three years. But the volume wasn’t there, and in a pay-per-outcome structure, fewer participants meant fewer payments against the same fixed costs.

The fast issue was the coronavirus, which by March 2020 was everywhere. Specialists whose craft was building face-to-face relationships went remote. Employers froze hiring, then started layoffs. Veterans in fragile recovery sheltered in place. Cohort 2’s average earnings fell 65 percent. It was nobody’s fault and everybody’s problem.

By summer, the project’s management committee, which we sat on, faced three options: 1) Continue as structured and watch the lenders absorb losses on a project whose underlying program was demonstrably working. 2) Wind down services and abandon enrolled veterans midrecovery. 3) Restructure. Use accumulated outcome payments to pay back the lenders, and shift the program to direct funding from the VA and the Commonwealth.

We restructured. The lenders were made whole; the veterans kept their specialists. Massachusetts, having watched the employment outcomes hold up through a pandemic, extended the program. It runs to this day, with the state’s Executive Office of Veterans Services as sponsor, long outliving its financing.

Over the years, we’ve spoken to thousands of people about outcomes funding, and one of the questions we get most about any given project is “Did it work?” We never really know how to answer. In fact, we’re not even certain that the people asking know exactly what they’re trying to ascertain.

The first American SIB, financed by Goldman Sachs, paid for cognitive behavioral therapy for teenagers jailed on Rikers Island. When it failed to reduce recidivism and was shut down in 2015, two of the deal’s architects coauthored an op-ed that grapples with the question. The program failed, they wrote, but the contracting structure didn’t. It had allowed the city to test a promising idea at no cost to taxpayers, measure it carefully, and—when it didn’t work—stop. Compare that with the ordinary fate of an ineffective government program, which is immortality.

So: Did Veterans CARE work? The intervention did well. The data systems held up. The partnerships survived a pandemic. The financial structure—the part with the investors and risk transfer—had to be dismantled so that everything else could survive. When stress arrived, the thing that made headlines, that we modeled so carefully, and that we spent long nights arguing about fell away. Government and philanthropy absorbed the risk.

A Financial Answer to a Governance Problem

Around the time of the Veterans CARE unwinding, people in our organization—one of the few structuring these mechanisms—were having doubts. Ten years after the first SIB, in Peterborough, the market just wasn’t taking off. Everything felt hard; deals didn’t replicate; we couldn’t manage down the complexity or accelerate the pace. The model was elegant in concept and excruciating in practice.

In June of that year, when the pandemic was three months old and most of our team was working from home, Social Finance’s leadership convened a meeting we had been dreading. The premise was simple and somewhat masochistic: Take every SIB we had launched and grade it, project by project.

As Veterans CARE suggested, what we found was something more complicated than abject failure. Every deal forced into existence data-sharing agreements and dashboards that outlived the project. Many had built better evidence and measurement hygiene. Most had meaningfully expanded important programs with high-quality providers.

But we underestimated the pressures and complexities of government. Competing interests and lack of capacity made the intricate, almost delicate realization of a SIB impractical. We asked for newfound accountability and attribution, genuinely novel procurement authorities, impact predictability, and financial precision. Meanwhile, governments rarely process their payments to nonprofits on time. We were building for the wrong decade, for a government of the future.

Cubes floating in the air with photos of social issue areas like housing, education, cash assistance (Illustration by Eddie Guy) 

Our government partners, whom we interviewed candidly, told us what they valued. Political cover. Measurement rigor. A shared vocabulary about outcomes. Multiyear funding stability across leadership turnover.

What could they do without? The complicated financing and the unwieldy timelines. “Can’t sustain SIBs with this long development timeline,” South Carolina’s Health and Human Services Director Christian Soura told us. He was the model’s friend and a genuine champion, someone who finalized and signed our deal on Christmas morning. The most telling line in the partner feedback came from Connecticut: Knowing what they knew now, having absorbed the outcomes mindset, they would have wanted to do outcomes-based contracting—but in an easier way.

What we built as a financial tool really wasn’t one. We designed SIBs to transfer financial risk from government to private capital. Doing so focused attention and imposed discipline. But much of the real risk existed outside the financial structure. When the basic mechanisms underlying the deal—data, a stable policy climate, timeliness and functionality of core systems—turned out to be unreliable, the financial structure wasn’t important. We were using a financial tool to solve a governance-and-capacity problem.

Sometimes the whole package—financing and all—can work. But SIBs run up against two major challenges.

First, these deals are often simply too demanding relative to the status quo. They presume decent historical data, measurement capacity, interagency partnership, and collaborative governance—all simultaneously—in a public sector that has been systematically starved of exactly those capacities.

Second, the model requires a shared understanding of risk that rarely exists. Most everyone—voters, officials, nonprofits themselves—believes the programs basically work the way we expect them to. Those of us marinated in the evaluation literature know that social interventions are wildly uncertain. There’s a subculture of evaluators, policy wonks, and academics who live chastened by Peter Rossi’s Iron Law, the gist of which is that, on average, large-scale social programs are found to have no net impact. Recent empirical work backs it up. That epistemological gulf is breathtaking and can poison this work. The criminologist John Roman once wrote in a blog post that governments don’t misprice risk so much as decline to think about it at all—making the SIB a sophisticated answer to a question its customer never asked.

A contract is a meeting of minds; ours were contracts between parties pricing different realities. At the end of our grading exercise, we realized that whether the model had “failed” depended on what you thought it was in the first place. As a mass-market replacement for how government buys social services, yes, plainly—the volume never came, the deals never got streamlined and standardized, the complexities were never ironed out. As a device that forced a generation of governments, nonprofits, and funders to learn what outcomes-based work actually requires, it succeeded, we would argue.

Still, enthusiasm was being replaced quickly by exhaustion. Already, the Social Impact Bond Lab—launched by Harvard economist Jeff Liebman in parallel with Social Finance’s own founding—had renamed itself the Government Performance Lab. When the people who built the SIB Lab take “SIB” out of the name, the field is telling you something.

Bricks, Not Cathedrals

Since 2020, those of us in the small niche that built SIBs, both at Social Finance and beyond, have tried to bring the philosophy of outcomes funding to new spaces. For some organizations, that has meant trying SIBs in new contexts—turning to environmental restoration, or to expanding the ambitions of hospitals as they treat complex health conditions. At Social Finance, that has meant rededicating ourselves not only to strengthening the basic infrastructure of good government but also to applying principles of the outcomes movement to education, workforce training, and impact investing writ large. It has meant working with businesses to use outcomes thinking to fund training and retention, working with philanthropy to create models that balance accountability with flexibility, working with investors to ensure their impact goals are really met. The thinking from our first decade, brought to new spaces, has found real traction.

The SIB was a cathedral: bespoke, ornate, and years in the making. When it comes to good government, we are now laying bricks. If the value of the SIB is discipline—define the outcome, measure honestly, fix what fails—then that discipline becomes the product, and the capital stack is ancillary. You don’t need a nine-figure number on a term sheet to write a contract that says measure it and act on what you find. You need a government willing to do what it takes to get better. But could we actually achieve greater discipline without the forcing function of private investors?

That’s what we set out to test. Where is outcomes thinking useful—and what is the range of outcomes strategies that work?

That brings us to the ducat test. While the American War of Independence was in full swing, Enlightenment philosopher Immanuel Kant was writing about ideas relevant to outcomes funding. In his Critique of Pure Reason, he proposed that our convictions are often held too strongly. “It frequently happens that a man … appears to be under no apprehension as to the possibility of his being in error,” he wrote. But put some skin in the game, and things change. “The offer of a bet startles him, and makes him pause,” Kant continued. “Sometimes it turns out that his persuasion may be valued at a ducat, but not at ten. For … he immediately becomes aware of the possibility of his being mistaken—a possibility which has hitherto escaped his observation.”

The first test for where outcomes thinking is useful asks whether real, meaningful risk exists in this program. If it’s a vaccine, focusing on and paying for outcomes adds cost without information; just buy the service and monitor it. On the other side of the spectrum, if nobody would stake their lunch money because it’s too unclear, you have a research question, and you just need more information.

Outcomes thinking is most useful in between, where the risk is genuine and the odds are arguable. Argument is half the point; it forces leaders to think hard about the realities of achieving impact and the weak points in the delivery pipeline. We have watched that conversation change how an agency behaves long after anyone could remember the deal it produced. It helps narrow, just a bit, the distance between how risky social programs actually are and how risky everyone assumes them to be.

That middle ground is about to get crowded. Governments are being pitched AI tutors, benefits-enrollment chatbots, and algorithmic case-management tools at a remarkable clip, often by vendors who appear, as Kant put it, to be under no apprehension as to the possibility of being in error. Most of these tools are plausible; few are proven, and fewer still have been tested with the people public agencies actually serve. It’s hard to think of a better candidate for the ducat test.

Turning Up the Dial

People occasionally ask what advice we would give to a newly appointed official with an issue that passes the ducat test. We’d tell them to find a starting place and move incrementally along what we’ve come to imagine as a dial. Each notch on the dial buys more accountability—and demands more of the system.

1. Deepen measurement. For many governments, the first objective is to decide exactly what to measure and to measure it well. Everything else rests on it, and almost nobody does it well. It takes clarity about your true objectives and discernment about what matters most. It requires accessing, sometimes integrating, and often improving the data needed to measure those goals. And then it means considering a range of technical measurement and evaluation tools and choosing the right one for the job—deciding where intensive, highly rigorous evaluation models are important, versus where simpler measures are good enough for the decision at hand.

Fine-tuning measurement is hard but doable. It demands data collection, but it offers visibility into what’s working and what isn’t. If this step sounds too aspirational to be practicable, choose one program and three metrics and ask a local university to be your accountability partner.

Excellent measurement helps set up better management. In Texas, for example, we helped the state’s Department of Family and Protective Services to create a scorecard that defined three key outcomes measures and set up a system to routinely report progress back to providers, along with each provider’s scores relative to other, similar programs. Data in this Texas department are no longer a one-way funnel of information from provider to government; a steady stream of feedback that helps organizations test and improve is now in place.

2. Manage actively. The biggest jump in outcomes orientation concerns contract governance. Most contracts solve for compliance: They tell providers what to do and make sure they do it. That presumes a level of certainty about the world—about people’s needs and the efficacy of solutions—that is unrealistic. When AI tools available to a caseworker or a job coach change faster than a procurement cycle, a contract that specifies activities locks providers into old methods. A cost-reimbursement contract punishes providers finding less expensive ways to do the work, since lower costs mean lower payments. Pay for results instead, and the provider that uses new tools to serve more people, and better, gets to keep the gain and reinvest it.

Outcomes-oriented management is a muscle dedicated to using good measurement to adapt programs to the reality of a changing world. Good management involves regular experiments, candid reflection, and ongoing program optimization, ideally by an empowered governance body helping adapt it to the contextual truths on the ground.

This notch on the dial demands governance bandwidth and timely data. In return, it offers adaptiveness—fixing things while they’re still fixable.

3. Add stakes. Creating a direct connection between meaningful outcomes and contract payment shines a spotlight on priorities. It drives real accountability, since all parties have to agree on a mechanism that determines precisely what happened—and, in doing so, it focuses everyone’s attention.

In Missouri, for example, we worked with the Children’s Trust Fund to add stakes to its early-childhood home-visiting contracts, which—as in South Carolina—send caseworkers to the homes of families with young kids. We defined a set of priority outcomes—with a particular focus on incentivizing prenatal enrollment—and offered modest bonus payments (up to 10 percent of core funding) to providers who achieved them. No investors are involved, but it’s enough money that providers take careful note and test new ways to improve their results.

Creating contracts with high degrees of contingency—that is, meaningful outcomes determine more than 50 percent of payments—changes the stakes further. This is the structure of the SIB: Someone is on the hook for what ultimately happens. Governments pay only for what really works, but, in return, they have to offer back the possibility of shared financial rewards and have to cede some degree of control, offering more flexibility for sophisticated, entrepreneurial nonprofits. For most governments, this is a stretch, but when it’s possible, it’s powerful.

Performance-based payment demands some contractual sophistication but, in exchange, offers providers a stake in success. More contingency requires the leadership and resourcing of a committed outcomes buyer, but it disrupts standard approaches and cultures as parties reorient their systems and management around results.

How these contracts are financed is a different question entirely. Some providers look to investors both for the working capital and to take on the risk; some ask for partial support; others can self-finance. Particularly in the years since 2020, we have often turned to philanthropies—whose primary allegiance is to improving lives, not protecting their principal—to share the risk.

That’s what happened in Anchorage, Alaska. Under the leadership of Mayors Ethan Berkowitz and Austin Quinn-Davidson, we participated in a project expanding supportive housing. Its financing was about as far from Wall Street as possible: A small amount of philanthropy from local funders jump-started the program until it started unlocking outcomes funds. As a piece of outcomes-based public finance, it is what a decade of bruises taught us to build: a simple, reliable metric; fast feedback; philanthropy where investors used to be; and a government paying for the thing it wanted.

After years of trying to walk, chew gum, and fly to the moon simultaneously, in Anchorage we had finally built something that just walked.

Where the Philosophy Went

The wide-eyed commentary that celebrated the creation of SIBs mostly ignored where they came from. We often conflate innovation with novelty, but that’s a mistake. Like many innovations, the SIB is an idea both old and new. It is rooted in—and an explicit reaction to—a century-long history of performance-based contracting that is, by any account, largely problematic. Value-based care in health, income-based repayment of student loans, pharmaceutical pricing based on patient results, prize competitions, and advance market commitments are promising experiments that approach big problems using outcomes thinking. Whenever deep variability occurs in results, buyers demand change. The big idea of outcomes-based contracting is to define what success looks like, agree on how you’ll measure it, price its value, and then pay for it if it happens.

Part of the impact is inspiration itself, rekindling the notions that brokenness is not inevitable, that innovation is worthy, and that we can yet squeeze more value out of the grand, chaotic architecture of government. That hopefulness, especially these days, is important.

All of this suggests, at the very least, that the broadly understood history of SIBs is mistaken in a fairly fundamental way. The implication that they were a fad that flared and faded ignores that they were one edge of a much broader outcomes movement—a small, intense, unusually self-aware and self-conscious edge, admittedly. But the movement, while hard to discern beneath the differences and the flash of these models, is real. It’s an economy-spanning search for ways to pay for what we actually want, rather than what we can easily count. That search predates our work, and it’s much bigger than we are.

That realization also suggested to us, beginning in 2018, that significant space for impact and new thinking existed beyond what we were pursuing with the likes of Veterans CARE.

Fittingly, another SIB pointed the way. The Massachusetts Pathways to Economic Advancement Project sought to help refugees and recent immigrants find employment by acquiring English-language skills and tied payment to participants’ earnings gains. It was one of our better projects: The data and partners came together the way they were supposed to, and the impact was substantial.

But Pathways also surfaced a niggling question. The value created by good training shows up first, and most clearly, in the wages of the person who was trained. Government captures some of that value eventually, through taxes and avoided costs. But if most of the value lies in the learner’s paycheck, why were we routing the entire contract through a state agency’s budget?

We began building a first cousin to a SIB: another form of outcomes-based financing, but with students, not governments, buying the outcomes. After some early experiments, we wrote up the idea in a white paper and called it a Pay It Forward Fund.

On October 15, 2020, when Governor Phil Murphy and the newly formed New Jersey CEO Council announced a coalition commitment to hire or train 70,000 New Jerseyans by 2030, they called for such a model to help more students succeed in training. The next year, we began working with the Higher Education Student Assistance Authority and half a dozen state agencies to design it.

The NJ Pay It Forward Program launched in 2022 with roughly $10 million, half from the state and half from employer partners. The core product is a zero-interest loan covering tuition and fees remaining after Pell and state grants. Repayment is capped at 10 percent of discretionary income, only above a minimum income threshold, for no more than five years, after which any remaining balance is forgiven. Learners who don’t clear the threshold owe nothing. Repayments recycle into the next cohort. And it’s all matched with a small, nonrepayable living stipend.

In the years since, the state has continued to expand the fund, which has grown past $25 million and enrolled more than 500 learners. Most of the training it funds is in nursing and allied health, where the outcomes and employer demand are strongest.

This was not a SIB. The state is the up-front funder, not an outcomes buyer. The learner’s own earnings—capped, time limited, and forgivable—determine whether the money is reimbursed. It is closer to income-based repayment than to anything we built between 2012 and 2020. But so much of the philosophy is the same: diligence on training providers using Department of Labor wage records and licensing pass rates, rather than self-reported placement numbers; explicit wage thresholds agreed upon in advance; capital that recycles based on results; carefully parsed and shared risk.

That design matters more as AI begins to reshuffle which skills the labor market rewards. A learner choosing a credential today is placing a particularly uncertain bet. Because repayments track earnings, the fund works as an early-warning system: When the benefit-to-cost ratio declines for a credential, it shows up in repayments well before it shows up anywhere else.

In the past five years, Pay It Forward Funds have become a major pillar of our work. Today, we manage 13 workforce and education funds nationwide, totaling over $330 million.

We haven’t given up on SIBs, but they are a small part of our work today. For all their challenges, they offer a captivating crash course in what it means to revamp everything in service of getting better outcomes. They show us that we can see the world through the lens of outcomes, rather than get sucked in by the black hole of compliance, and that accountability can enable entrepreneurship, rather than serve as a weapon. Part of the impact is inspiration itself, rekindling the notions that brokenness is not inevitable, that innovation is worthy, and that we can yet squeeze more value out of the grand, chaotic architecture of government. That hopefulness, especially these days, is delicate and important and ineffable, and we need to keep it aflame.

Acceleration Through Aggregation

SIBs may also become easier to implement. For most of the past 15 years, we built projects one at a time, revamping procurements and integrating data systems and identifying funding and sometimes writing new legislation—almost always for a single effort. We asked each government to invent, from first principles, the act of purchasing an outcome: to decide what it wanted, what that was worth, how it would know, and who would check.

There is, somewhat obviously, a better way to do it. As with pretty much everything in this field, it has an underwhelming name: an outcomes fund. It is also the closest thing we have to a general theory of how to stop doing merely artisanal work.

An outcomes fund buys a bunch of SIBs in a standardized way. It pools outcomes buyers, and then—instead of structuring one bespoke contract at a time—it sets standards. The Government Outcomes Lab at Oxford University describes five questions every outcomes contract has to answer: what outcome, for whom, at what price, verified when, and with which partners at the table. In a one-off SIB, all five get renegotiated from scratch, whereas a fund decides some of those questions centrally. It also centralizes expertise: Jurisdictions can rely on the fund for measurement capacity or governance support or analytics.

We have started to feel something closer to optimism about SIBs. Some of what made these contracts punishingly slow and expensive is becoming less cumbersome. The artisanal work that made each deal a bespoke negotiation is the kind of labor Al now accelerates.

The United Kingdom has been at this for years, having launched a series of outcomes funds since 2013. In the summer of 2025, the new UK government announced the 1 billion British pound ($1.3 billion) Better Futures Fund.

The United States came close to launching an outcomes fund, too. In January 2024, the Advanced Research Projects Agency for Health (ARPA-H)—the government agency known for health moonshots such as surgical robots—launched the HEalth care Rewards to Achieve Improved OutcomES program. HEROES would have paid up to $15 million per community, plus a $30 million local match, for population-level reductions in opioid overdoses, cardiovascular risk, and severe maternal complications. Three design choices made it the most advanced outcomes fund anyone has attempted. Outcomes were to be measured across an entire geography—a metro area, a state—regardless of who insured the people in it. ARPA-H built the measurement itself, a central dashboard every applicant would use, so the evaluation didn’t have to be reinvented and separately financed in each contract. And the agency positioned itself as the first buyer in line, expressly to draw other buyers—insurers, health systems, philanthropies—in behind it.

Applications closed in November 2024; a national competition selected six communities. But in the summer of 2025, days before winners were to be announced, the federal government canceled the program without explanation.

A serious American outcomes fund, staffed and funded to meet an ambitious mandate, is a tantalizing policy idea because it’s realistic and powerful. As we found out, however, it’s still just more than the government could hold on to.

The Future Isn’t That Far Ahead

But even that may be changing. In the past 12 months, we have seen a surge of interest in SIBs. These days, we at Social Finance get weekly queries from jurisdictions intrigued by the idea. This is, in part, a logical reaction to austerity: Governments want to make sure they’re getting the most from taxpayer money. Social Finance was founded in a similar era, in the wake of the Great Recession.

We have started to feel something closer to optimism about SIBs. Some of what made these contracts punishingly slow and expensive is, with new technology, becoming significantly less cumbersome. The artisanal work that made each deal a bespoke negotiation—evidence reviews, outcome pricing, cost-benefit modeling, mapping and reconciling jurisdictional requirements—is the kind of analytical labor AI now accelerates. A pricing analysis that took a team a quarter now gets drafted, stress tested, and rerun against new assumptions in days. Fourteen-month development cycles sold to officials with 24-month tenures were a fatal arithmetic; cut development to months, instead of years, and the timeline gets more realistic.

Novel contracts can also translate more easily into understandable language for different parties. Outcomes contracts have often brought together unusual partners around common interests—that’s their value. But it also introduces a constant need for translation, because the same explanations and the same math don’t apply across prisons, Medicaid insurers, and homeless shelters. In the past, our team did its best to interpret across those differences. In the future, AI will speak to each stakeholder in their native language. We already use AI regularly to explain project details in plain language as we engage community members in design, and we use it to adapt materials to meet different partners where they are; it will only get better at this.

And we can access outcomes data faster and more inexpensively. Invariably, one of the biggest challenges in outcomes contracting is getting our hands on the right data within a reasonable timeline. But things are changing. A growing movement toward statewide longitudinal data systems is in place, and linking to administrative data systems, such as wage records, health claims, and housing vouchers, is getting easier. We’ll be able to access important indicators of success earlier, as AI allows us to understand and analyze unstructured data like case manager notes and open-response surveys.

Finally, in some cases, good counterfactuals are becoming less difficult to compute. Our deal teams and our partners regularly got overwhelmed by determining attribution of impact. What would have happened to these mothers, these veterans, these families, absent the program? Algorithmically constructed comparison groups were always possible, but big-data approaches are making them stronger. Harvard University economist Raj Chetty’s team at Opportunity Insights, working with longitudinal federal wage and demographic data from the Census Bureau, now uses massive-scale earnings and employment histories to produce credible “digital twin” impact estimates at a small fraction of the cost and time of a randomized trial. Linked administrative data can now replace some of what only a multiyear randomized trial could once provide.

SIBs won’t immediately become easy, but they might move within reach for sophisticated and motivated governments. Most may still be reluctant to attempt a SIB, but what kept them from the first two notches—measuring well and managing to what they measure—was largely capacity. AI tools are unlocking some of that capacity. A county agency that could never have staffed a SIB may soon be able to build its own version of the Texas scorecard.

Maybe the first decade and a half was just the scaffolding phase, and the future we’ve been building for isn’t so very far away. After all, the medical literature says it takes 17 years for well-evidenced practice to become standard care. The first SIB launched in 2010.

Yes, the grand revolution that SIBs promised did not arrive—or at least it hasn’t yet. The work, less grandly, continues. That work is the outcomes movement, the broader philosophy behind the SIB. We would stake considerably more than a ducat that, 10 years from now, tying money to verified outcomes will be so inexpensive and ordinary that nobody bothers to give it a name. This, probably, is what most reform looks like after the press releases stop: You spend a decade learning what the simple version is, you cross out the early promises line by line, and then you build something modest enough to last.

Read more stories by Tracy Palandjian & Jake Segal.