I wouldnt say there is skepticism that there are interventions with high ROI only with ROI higher than the currently known interventions. Medicare pays $45 for a two dose course of a COVID vaccine. Compare this to the $7 cost of a bednet and you have to ask, is a COVID vaccine more than 9x as likely to save a life as a bednet. Plus the search cost of the RCTs has to be recouped. You are saying work on the demand side, but i think the evidence from many interventions shows the best way to increase uptake is to meet people where they are and provide the product/service for free.

I think the thesis is right (EA should fund RCT) but we actually shouldnt believe RCT will find interventions better than those currently being funded.

I will also restrict my analysis to global health/development as you did. A simple model of cost effectiveness is as follows:

cost effectiveness = (severity of problem instance * probability of solution from one instance of intervening) / cost of one instance.

I think this is straightforward but ill try to justify here. Cost effectiveness is by definition effectiveness over cost. Effectiveness is the expected value of the intervention. The expected value is the good done per success * probability of success. Therefore, if we define problem severity as "amount of good done by solving the problem" - which i think is intuitive - the model holds.

Looking at this model we can see that to find interventions more cost effective than the ones currently being funded the intervention would have to do at least one of the following:

1. Solve a problem more severe than the problems currently funded

2. Have a higher success probability than the interventions currently funded

3. Be cheaper than the problems currently funded

1 seems pretty unlikely. The best currently funded interventions (lets just use the givewell top charities list as an example) address the following problems:

Child mortality - malaria, vitamin A deficiency , vaccines

Blindness - Deworming

Poverty - Givedirectly - ill leave this one aside because its the benchmark

Youll be hard pressed to find problems where a single instance of solution does more good than preventing a single death or a single instance of blindness. The only thing I can think of would be infectious disease intervention where one person not being infected means that multiple lives are saved at the margin by preventing transmission. Importantly, malaria and vaccines already have this quality so the only possible improvement would be changing to more prevalent infectious disease, but the only very deadly/prevalent infectious disease not being heavily intervened on above is HIV/AIDS which right now is very expensive to prevent or treat.

2 is also very tough. The basic formula for success probability in the currently funded interventions is

1. Find a common ailment
2. Prevent everyone from getting the ailment using basically 100% effective prevention on everyone

This is how anthelmintics, vaccines, vitamin A supplementation, and malaria prevention work.

I can only see two ways to improve this formula, treating more common ailments (the problem is there arent any) or by more effective targeting. For example if we only gave bednets to the particular people who would have gotten deadly malaria without them. The problem here is even if predicting who will get malaria, which children are vitamin A deficient, will get pneumonia, or will have schistomiatosis is possible, the data collection for making predictions would probably be more expensive than the gains from targeting.

That leaves 3, but the popular interventions are just so cheap, dollars per instance.

But I started by saying I agree. The reason why is the current interventions are basically fully funded. Givewell is holding money because they cant spend faster than they receive donations. We dont need to find interventions that are better than the standards, we need to find new effective interventions, period. RCT is the way to do this, and no organization is funding RCT with the explicit goal of finding cost effective interventions that can be implemented by granters. The main funders of RCT are governments funding the academy with the goal of advancing knowledge -which of course is extremely important - this leads to RCT which are often undertaken in support of theory (see microfinance) or funding of RCT which are based in the host country rather than in developing countries where problems are more severe and interventions are cheaper. An organization that exclusively funds RCT with an EA focus would be filling a huge need in the EA community.