Policymakers worldwide have proposed a new contract - the 'social impact bond' (SIB) - which they claim can allay the underperformance afflicting not-for-profits, by tying the private returns of (social) investors to the success of social programs. We investigate experimentally how SIBs perform in a first-best world, where investors are rational and able to obtain hard information on not-for-profits' performance. Using a principal-agent multitasking framework, we compare SIBs to inputsbased contracts (IBs) and performance-based contracts (PBs). IBs are based on a piece-rate mechanism, PBs on a non-binding bonus mechanism, and SIBs on a mechanism that, due to the presence of an investor, offers full enforceability. Although SIBs can perfectly enforce good behaviour, they also require the principal (i.e., government) to relinquish control over the agent's (i.e., not-for-profit's) payoff to a self-regarding investor, which prevents the principal and agent from being reciprocal. In spite of these drawbacks, in our experiment SIBs outperformed IBs and PBs. We therefore conclude that, at least in our laboratory test-bed, SIBs can allay the underperformance of not-for-profits.