Research
Identifying solutions
A problemProblem. A specific issue within a cause area that reduces the well-being of a target population. worth working on usually has several ways in. For each one, we map the candidate solutionsSolution. An approach or intervention: the general word when the rung does not matter. and compare them on the same questions: what is the evidence, what has to be true for it to work, and what does a unit of progress cost.
Evidence, weighed not counted
Not all evidence is equal. We work down the hierarchy of evidence: systematic reviews and meta-analyses first, then individual trials, then observational work, then informed judgement. What we look for is convergence. A claim supported by a trial, a practitioner and an independent dataset that do not know about each other is worth more than any one of them alone.
In practice, that means starting from the best secondary sources, then talking to the people closest to the work: researchers, implementers and the experts.
The chain from money to outcome
For every solution, we map who it serves, what causes the problem, and the mechanism by which the interventionIntervention. A specific mechanism for change. changes it. Then we stress-test that chain and ask where the weakest link is. Every theory of change has one. An organisation that can name its own weakest link is usually a better bet than one that cannot imagine failing.
The rubric
Eight questions, asked of every intervention
Standardised, so that two very different interventions can be compared honestly, and so a weak answer cannot hide behind a strong one.
Evidence quality
What is the strongest evidence that this works, and does it converge?
What we look at: Hierarchy of evidence, replication, independent sources
Cost-effectiveness
How much outcome is achieved per unit of cost?
What we look at: Programme cost data, outcome measures, benchmarks
Scalability
Can the intervention maintain effectiveness at a larger scale?
What we look at: Scale-up evidence, organisational capacity data, limiting factors, operational context
Implementation feasibility
Can this intervention be delivered effectively in India?
What we look at: Operational context, NGO capacity, regulatory fit
Contextual fit
Does evidence from other contexts transfer to India?
What we look at: Context comparison, adaptation pilots and studies, generalisable components
Neglectedness
Is this underfunded relative to other ways of addressing the problem?
What we look at: Funding landscape data, sources that quantify the problem
Funding gap
Would additional funding here produce impact at the margin?
What we look at: Current funding levels, absorption capacity
Institutional readiness
Are there capable organisations to implement this?
What we look at: Track records, team and endgame assessment
The right tool for the evidence
Where the evidence is strong enough, we build back-of-the-envelope (BOTEC) cost-effectiveness analyses for the intervention: cost per unit of outcome, compared across options. Where it is thinner, we use expected-value estimates that make the uncertainty explicit.
Two caveats we hold ourselves to. Models built on weak evidence, and expected-value calculations, are almost always too optimistic. And chasing the highest number in a stack of models selects for modelling errors as much as for impact, because an outlier is either extraordinarily effective or wrong.
Currently, we draw on models created by the global evaluators named in prioritising problems to guide our theses. As we uncover new interventions in our research, we will publish the accompanying models.
Uncertainty is work, not a verdict
When a promising area has an open question, we actively investigate. We help donors make bets that systematically evaluate whether a solution is worth funding at scale. In young and neglected areas, the limiting factor is often not donors or willing implementers but primary evidence about what works, and the donor willing to fund that evidence unlocks everyone who comes after.
What they get is a field that knows more than it did, and every later grant, theirs and everyone else’s, lands better because of it. In practice, this means replication pilots that test whether a strong intervention researched somewhere else holds up, first trials of interventions with a strong theory of change and no data yet, and pilots that collect contextual information on which factors matter most.
- Map the candidate solutions for each problem and compare them on the eight-question rubric
- Weigh evidence by quality and look for independent sources that converge
- Stress-test the causal chain and name its weakest link
- Match the analysis to the evidence, and let no single number decide
- Fund the open questions in promising areas, and publish what we learn
Where this leads
Knowing what works is necessary and not sufficient. The last step is getting money to it, at the right size, through the right vehicle: maximising impact. Or step back to how we choose problems in the first place: prioritising problems. For a sense of what a rupee here buys relative to what you earn, see how rich you are globally.