Work in progress. A research tool built by the Founders Pledge climate team, now open to researchers elsewhere. Start here

Grantmaking under high uncertainty

A framework and a tool for comparing grants when the evidence cannot settle which is better.

Open the Portal How it works

Built by the Founders Pledge climate research team for our own grantmaking. Written for researchers and grantmakers; free to use.

Why does it matter?

Most of what we care about cannot be measured like a bednet

Compare it to the gold standard. Direct-delivery global health has randomised evidence, and you know fairly closely how that evidence applies. Model uncertainty is low.

Now think about climate. Building a neglected energy technology directly is not very cost-effective. Advocate for a government to deploy it instead and you probably do better — but no one knows how much better. Advocate for innovation policy rather than for the technology, and you have added another uncertainty on top. It becomes an almost meaningless question whether something meets the bar. And that is essentially every field we care about, apart from this one.

Direct delivery a ladder of multipliers
Cash, rich country GiveDirectly Against Malaria New Incentives ×200
Policy advocacy two modellers, same structure
Modeller A Modeller B Build it directly Advocate for it Innovation policy

The approach

Grantmaking methodology for high-uncertainty contexts EA Global London 2026 ↗

Model everything, all at once

So we stopped asking whether one grant meets the bar. We keep the uncertainty instead of collapsing it: every input is a distribution, and the whole set of grants is simulated together rather than one at a time.

We then compare portfolios rather than grants. Each is scored across all ten thousand worlds, so the one worth choosing is the one that holds up everywhere — not the one that wins most often.

Step one

Say what you believe

Describe the grants and the uncertainties that separate them. Each grant leverages its own subset.

Grant AGrant BGrant C
Step two

Simulate the world

Draw every uncertainty ten thousand times. Each grant's cost-effectiveness becomes a distribution, not a number.

Step three

Find what holds up

Identify the portfolios that perform best across all ten thousand worlds at once.

ABC wins most often holds up best

Which uncertainties are worth your research time

Because the model knows which uncertainties changed the allocation, it also knows which ones are worth researching. Some assumptions you spend a long time on never change a decision; others turn out to decide it. That makes this as much a way of choosing what to research as what to fund.

Which uncertainties change the decision illustrative
Advocacy effectiveness, EU Policy reversal risk, US Innovation system size Delay to decarbonisation Grantee strength Baseline success rate

Worked examples

Illustrative case studies

These are teaching examples, deliberately simplified — not records of grants we have made.

See all five worked through ›

Malaria Three grants in sub-Saharan Africa, two direct and one working through another charity. See the walkthrough AI safety talent Two routes into one field, separated mainly by how long each takes to pay off. See the walkthrough Global catastrophic risk Prioritising across cause areas, where the common unit has to be built first. See the walkthrough Healthcare workforce training Multi-year budgets, and holding money back for opportunities not yet found. See the walkthrough Primary healthcare workforce Rebuilding a model's architecture, replacing three formulas with one shared multiplier. See the walkthrough

Use this in your cause area

We have not solved this. The Impact Model is a research project of the Founders Pledge climate team, and if you make grants under deep uncertainty we would like to hear from you.

Open the Portal Get in touch