The Case for Cash: Why Direct Transfers Beat Research Grants

The overhead problem with institutional grants

A research grant is a long chain of hands. A donor funds a foundation, the foundation runs a program office, the program office issues a request for proposals, universities and labs write applications, review panels score them, and the winner hires a grants administrator before a single researcher is paid. Every link takes a cut — salaries, buildings, compliance, reporting — and every link adds delay.

None of that overhead is theft; most of it is real work. But it compounds. By the time a dollar becomes a researcher's hour, a large fraction of it has paid for the machinery of giving rather than the giving itself. Cash transfers invert this. A mobile-money transfer to a poor household costs a few percent to deliver and arrives in days, not fiscal years. GiveDirectly, the best-known direct-transfer charity, reports that roughly 85–90 cents of every donated dollar reaches the recipient (GiveDirectly). Very few institutional grant pipelines can honestly claim a comparable number once you count the full cost of administering them.

The overhead question matters even more in AI-safety funding, where a small number of large foundations and labs direct enormous sums to a small number of research institutions. When the money concentrates, so does the infrastructure needed to manage it — and the people the research is supposedly for never see a dollar of it.

What the evidence actually shows

Cash transfers are among the most studied interventions in development economics, and the findings keep pointing the same way.

• People spend the money well. J-PAL's review of cash transfer programs found no systematic evidence that recipients increase spending on alcohol or tobacco; in several studies that spending fell (J-PAL).
• Transfers don't kill work incentives. The widely cited 2016 review by Banerjee, Hanna, Kreindler and Olken of seven government cash programs found no significant negative effect on recipients' work effort (J-PAL).
• They raise the floor durably. GiveDirectly's own research collection documents gains in consumption, assets, food security and mental health, including in its long-term studies in Kenya and Uganda (GiveDirectly research).
• They're a benchmark, not a gamble. GiveWell treats unconditional cash transfers as a strong baseline intervention and evaluates other programs against them (GiveWell).

Compare that to the average institutional research grant, where the measurable outcome is often a paper. Papers matter — this essay cites several. But the causal chain from a funded paper to a changed life is long, uncertain, and rarely audited the way a cash transfer is.

The moral argument: people know their own needs

Underneath the cost-effectiveness data sits a simpler claim: the person closest to a problem usually knows more about solving it than a distant committee. A parent deciding between school fees, a tin roof and medicine is doing the same thing a grant committee does — allocating scarce money against competing needs — with far better information and far stronger incentives to get it right.

Institutional grant-making implicitly assumes that experts can identify needs better than the people who have them. Sometimes that's true: vaccines, basic science and infrastructure are genuine public goods that no household can buy alone. But a huge share of philanthropy funds things people could decide for themselves — training programs they didn't choose, products they wouldn't buy, research whose benefits are promised rather than shown. Cash transfers treat poor people as decision-makers instead of research subjects. That's not just better economics; it's more respectful.

Children make the case even more sharply. A transfer to a family with children reaches the child as food, schooling and a safer home, with no parallel bureaucracy required to "target" the child's needs. The evidence on child outcomes — health, nutrition, school attendance — is among the strongest in the cash-transfer literature.

What this means for AI-era giving

AI-safety philanthropy is unusually concentrated: a handful of funders, a handful of grantees, and a shared assumption that the best use of billions is to fund institutions that produce research and policy. Some of that work is clearly worth funding. But the sector's overhead is real, its feedback loops are slow, and its beneficiaries — the public the research is meant to protect — receive nothing directly.

A balanced portfolio would treat cash as the floor, not the leftover. Fund the research that genuinely needs institutions. Then send the rest to people, including children, through the channels that already work: mobile-money transfers with published delivery costs, benchmarked against GiveDirectly-style programs. If a research grant can't beat a direct transfer on expected good per dollar, the transfer should win by default.

C∞A∞I∞O builds donation pages and machine-readable impact summaries for mission-driven groups, and we're planning projects in housing and relief, including microhom.us. See each one and its real status on our Fund the Impact work page.

FAQ