I wrote a draft on this a long time ago, thought I can publish it during draft amnesty, realized that it's pretty out of date and ended up spending a lot of time improving it. I'm confident about the questions and problems I raise here. I'm much less confident about what should be done about them or specific examples I provide.
These are my personal opinions, it doesn't reflect the opinions of any organizations I'm affiliated with. Thanks to Ben Anderson, Ceren Karabulut and Bahadır Şirin for feedback. 

Two-sentence summary: Health progress at scale comes from stronger health systems, and there are probably cost-effective interventions to strengthen health systems in LMICs. EA has engaged little, and late, with finding what those interventions are in my opinion.

Actual Summary

EA has a strong track record of finding, funding, and founding cost-effective delivery NGOs, and has saved hundreds of thousands of lives in the process. It has a weak track record of conducting research, founding organizations, or funding programs aimed at effectively strengthening health systems that fail to serve people who need to be saved by philanthropists.

The historical record suggests that large-scale health gains usually from stronger health systems (or innovation) rather than externally funded delivery NGOs (same with poverty alleviation). Recent history also offers well-documented examples of NGOs helping governments improve health systems, indicating tractable pathways for supporting health systems strengthening efforts.

Although EA interest in HSS has grown post-USAID, being late to government-facing work is especially costly, as it requires, or benefits from years of relationship-building, institutional trust, and local knowledge.

Outline:

1. Effective charity and effective development are related but distinct questions. The first asks how to deploy philanthropic resources to help people as cost-effectively as possible. The second asks how countries build functioning health systems. Both matter. EA has built world-class infrastructure for the first and engaged minimally with the second. Unlike the growth debate, the health version is more tractable: there are concrete examples of external actors successfully strengthening health systems.
2. EA's global health portfolio converged on a specific approach, for understandable reasons. Vertical, delivery-focused, grant-dependent, measurability-first. The approach was appropriate to the problem GiveWell was solving. But the community overindexed on it while underexploring alternatives.
3. The USAID cuts revealed the distinction in practice. EA organizations responded quickly, including funding systems-level work like technical support to Ministries of Health. But this was exactly the kind of work that hadn't been in the portfolio before, and could have been done better if it wasn’t previously neglected.
4. HSS interventions are likely harder to measure, and there is need for better frameworks to measure and prioritize HSS interventions well. $70,000 generating evidence that convinces a government to fund a national program is a different kind of impact than $70,000 buying a fixed number of commodities. I believe some HSS interventions may score well with existing frameworks, but there is likely to be a lot of value in investing in new frameworks and analysis before jumping to conclusions about best HSS interventions.
5. Recent diversification is welcome, but came late and remains narrow. Essentially, I highlight reasons for why this post would have been more valuable to write two years ago! As there is more interest from EA actors. GiveWell, AIM, Rethink Priorities, Coefficient Giving has entered the space. But I believe these came late, and many of these efforts still partially reflect limited engagement with the “how to build functional health systems” question.
6. Historical, structural, and sociological factors explain why this gap persists. Limited funder diversity, declining intellectual engagement with GHD, the community's demographic composition, and the absence of GHD-focused community infrastructure have all contributed. These compound each other.

My recommendation isn’t that EA should pivot wholesale or put most of it's resources ito health systems work, as that doesn't seem to be the comparative advantage of many EA organizations. I'm saying that frameworks, talent, organizations, and attention that could have unlocked high-impact projects haven't been sufficiently developed, leaving substantial impact on the table.

Section 0. Effective Charity and Effective Development Are Different Activities

"Is there any economy that escaped poverty (ie eliminated poverty at scale) through foreign aid or by giving handouts like chicken and cash?" 

— Yuen Yuen Ang

Yuen Yuen Ang, Ken Opalo, and Lant Pritchett’s answers to Ang's opening question above is a firm no. They argue that many (or most) aid and philanthropic projects do not meaningfully contribute to development, they answer the question of charity. Pritchett draws the distinction this way:

"The development question is: 'How can the people living in Niger come to have broad-based prosperity and high levels of wellbeing?' The charity question is: 'If some agency is going to devote a modest amount of resources to targeted programs that attempt to mitigate the worst consequences of a country's low level of development, what is the most cost-effective design of such programs?'"

I don't fully endorse Pritchett's framing, I probably value improving the effectiveness of charity more than he does. But I wholeheartedly endorse the distinction itself, even though it makes me uncomfortable as someone who works closer to the charity side. This parallels Hillebrandt and Halstead's "Growth and the case against randomista development “, the/used to be the most upvoted post on the EAForum, also made this distinction, so this is not the first time this distinction is being made on the EAForum. I won’t get into growth that much besides saying it’s important and probably more EAs interested in GHD should consider working on growth. I’ll primarily focus on global health in this post, as that’s what I know more about, and also because it’s more underdiscussed. However, I still want to say a few things about how good charity work is optimizing for something slightly or significantly different than what good development work should optimize for.

The answers to questions of development and charity are both valuable. But the two questions differ in how confidently and precisely you can answer them. The development question involves long causal chains, slow feedback loops, and deep uncertainty about what works where. How to promote economic growth, how to improve civil service, or improve public service provision are questions whose answers are harder to test, verify, as well as implement.

Compared to the development question, the charity question is easier to answer. Charity question lends itself to clearer metrics, shorter causal chains, and more robust attribution. You can more confidently and quickly see whether your charitable program (or aid program) works or not. You can run an RCT to find out about the impact of bednets or deworming, or cash transfers, more easily compared to pro-growth policy, public finance reform or civil service restructuring.   It’s harder to run RCTs on effective institutional reform though not impossible (some examples are here).  

But let’s say you want to go further. You’re not interested in just impact or ensuring the effectiveness of social programs. You want to cross-compare social programs of any sort, identify problems and solutions that would give you the most bang-per-buck. If you want to rigorously and confidently understand the most cost-effective interventions, and then donate/get people to donate to them, then this probably leads you to something like the following question:

The effective charity question: Given limited philanthropic dollars, how can we help people as cost-effectively as possible, in a way we can meaningfully measure, attribute, and compare across interventions?

This question is a good question, particularly within the context of the problem EAs were solving: identifying, launching and funding charities that are demonstrably and measurably extremely cost-effective, and lack of an ecosystem that does that. But that’s even more different than the development question. Answering the question of “how to save and improve lives” is good, and it enables you to save a lot of lives. But what’s even better (and admittedly harder) is countries having strong enough health systems where people’s access to basic medicine isn’t wholly dependent on the good will of donors abroad.

Section 1: Global Health Reflects the Distinction

“Is there any country that built a functional health system, that ensured access to healthcare for wide swathes of the population, through foreign aid or foreign NGOs?” A question I believe not enough EAs are discussing.

I’ll restate the questions for health:

The effective health charity question: How can we improve people's health as cost-effectively as possible, in a way we can meaningfully measure? EA has answered this well: bed nets, SMC, vitamin A, deworming. This work has saved hundreds of thousands of lives.

The effective health systems strengthening question: How can we effectively help countries build health systems that make lasting progress against their most pressing health challenges?

Now, this is related to economic growth too, but there is a lot of variance. Countries at similar income levels can have wildly different health outcomes. Rwanda and Sierra Leone have similar GDP per capita, but child mortality is three times higher in Sierra Leone. Haryana has higher GDP per capita than Kerala, but child mortality four times higher.

Why? Well, that is a question that you can discuss with very few people at EA conferences, along with the following questions: How did Thailand achieve universal health coverage at relatively low cost? How did Rwanda built a functional community health worker program that dramatically improved various health outcomes? How did China dramatically reduce the price of drugs through pooled procurement? Why does Ghana have higher health insurance coverage?

(My impression is that) Countries that achieved good health outcomes at a low cost, didn't achieve progress primarily through a proliferation of grant-dependent organizations tackling every problem one by one. And compared the general growth argument, there is an abundance of examples of external actors successfully helping countries strengthen their health systems. To reiterate, the policies and interventions mostly responsible for most or significant number of all large-scale health gains fall outside what the effective charity question would guide you towards. Even if it’s hard to measure, hard to identify what works, it’s at least worth trying how it can be measured, and what seems to work best, which is not something the EA community has properly tried, at least publicly, except:

Rethink Priorities' recent HSS report.  found promising cost-effectiveness estimates for several HSS interventions (IMCI supervision at ~8,500x and community scorecards at ~3,100x in Open Phil's framework). Though it again focuses on the more measurable end of HSS interventions as Nick Laing notes.

Some readers may say “well, systems change sounds good, but it’s hard”. I agree, but the systems change critique against EA kind of covers everything from “EA isn’t overthrowing capitalism” to “EA should be focused more on Georgian taxes”, so that’s not a term I’ll use in this post. And contrary to radical systems change, health systems strengthening is arguably fairly tractable. It’s also measurable, though probably in a way that’s different (and probably less confident) from conducting CEAs for effective charities.

2. What I Mean by "Health Systems Strengthening"

HSS is a contested term, what counts as HSS is different compared to who you ask. Not as in "Real communism hasn't been tried"-type contestation, but more along the lines of different preferences regarding what type of work should count as HSS (e.g. helping government build a capability vs. embedding technical foreign staff without considering institutionalization) For example, some people may not count certain versions of type 1 below as HSS work. I won't get into that discussion at all, and my typology below may be pretty flawed. I provide these types and examples as illustrative, and they need deeper evaluation than I provide here.

Type 1: Helping governments run disease-specific programs better. The government is the implementer, the external organization provides various types of support such as technical assistance (e.g. embedded advisors/consultants, training, data support, or protocol development) to help the government do its own program better. This type of work seems less neglected in EA, at least funders. Still somewhat vertical, but focused on building government capacity.

Examples: CHAI helping the South African government scale up ART from 800,000 to ~3 million patients while negotiating drug prices that saved the government nearly $1 billion. PATH supporting malaria vaccine rollout in five countries through readiness assessments, cold chain planning, and coordination between national immunization and malaria programs. LEEP or Concentric Policies is also an example of this.

Type 2: Transitioning NGO programs to government ownership. NGO-delivered programs designed with explicit handover plans, where the goal is demonstrating a model and then transferring it to the state. This is conceptually different from an NGO that delivers services indefinitely with its own staff and funding. 

Last Mile Health designed its community health worker program in Liberia for government handover from the start. The transition of Avahan's HIV program to the government of India and Nexleaf Analytics in Tanzania are further examples. I know less about these than I'd like.

Type 3: Helping governments build general capabilities. Here, external actors help governments develop institutional capacity that applies across health challenges, prioritization, health financing, data systems. This is the most neglected type from an EA perspective and the hardest to measure, but it's where I see the most compelling leverage.

Example: This has many different examples, and one interesting direction here is: Helping governments improve their ability to prioritize better!

EAs love prioritization. I’m unsure whether all governments love prioritization, but all governments need to prioritize, as they are making decisions about healthcare. Health technology assessment and essential package design is a way to help governments prioritize better. I’m unsure about the ultimate impact of these examples, but: In Pakistan, an estimated $1–3 million in technical assistance helped the government design and adopt a 117-intervention essential health package, aligned with Disease Control Pririoties 3 evidence, now shaping how public health spending is allocated across a population of over 200 million. In Thailand, a $1 million seed grant (not philanthropic, domestic grant) launched HITAP in 2007, and it shapes national coverage decisions for 47 million people.

Center for Global Development's iDSI network has supported LMIC governments with health technology assessments, and I view this as exactly the type of HSS work most conspicuously absent from EA's portfolio. If you believe global health improves when philanthropic dollars are better prioritized, helping governments prioritize their own (much larger) health budgets should be explored. More groundwork may be needed, mostly modelling, with limited retrospective measurement and assessment of health impact and economic benefits so far. Beyond HTA, NGOs can also help governments develop a response to important yet neglected problems, such as a local NGO in Mali helping the government develop a policy against diabetes/NCDs. Other random examples I know are Malaria Consortium helping build digital data systems for the government-run community health worker program in Mozambique, or Results4Development helping build market shaping capacity in Ethiopia.

Working with governments and getting them to own a program is difficult. But if your instinctive response to pro-HSS arguments is "it doesn't seem tractable," (it’s just probably wrong, or to be more kind) that instinct is  not strong enough to warrant ruling out HSS. Tractability depends on the specific intervention and the country. There will or could be competing priorities, elite capture, insufficient resources. Political economy and incentives of government officials will be tricky to navigate, government ownership will take time but it's not rocket science.  

Preemptive Answer to a Potential Question: Well, this sounds all good, but isn’t HSS saturated?

Health is also a saturated field, but that doesn't mean there aren't impactful interventions worth pursuing. HSS is incredibly broad, and at a high-level, you may deem it intractable or saturated, and you'd be correct in saying it's an area with a lot of funding. You may also say the same about nutrition, but that doesn't have anything to do with whether Fortify Health or Taimaka should exist. It's important to think more granularly about how neglected a certain thing is, and for HSS being saturated at a high-level doesn't mean there are neglected opportunities. Saturation at a high-level also provides an opportunity: You can probably create impact (that would count as impact from the EA framework), using non-EA funding. I’ll provide a suggestion on this in the last section.

3. EA's Portfolio Converged on a Specific Approach, For Understandable Reasons

There are Playpumps and Scared Straights, and there are Helen Keller Internationals and Against Malaria Foundations. There is a need to differentiate very bad charities, mediocre charities, very good charities, and best charities. There was very a scarcity of resources to help donors, small and large, do that differentiation back in the day.

My impression is that, beyond maximizing impact, core EA institutions such as GiveWell and Giving What We Can emerged partially to provide individuals a way to maximize impact in a rather specific way: building platforms, research and an overall ecosystem that promoted rigorous measurement and cost-effectiveness-oriented thinking, that can provide donors an understanding of where to donate. And provide you an opportunity to save lives, or helped you fulfill your moral obligation effectively, depending on your outlook.

To partially repeat what I said in the first two sections: If you are able to say "If you donate this much, you are very likely to have this much impact because we have this RCT and/or we can measure the charity’s impact fairly rigorously", that’s more conducive for building a credible platform or ecosystem that cares a lot about cost-effectiveness and effective use of funds. Donors that care about measurable or verifiable impact, and/or cost-effectiveness would probably be less excited about "We funded an NGO that has a 25% chance of convincing the government to do something we probably can't measure, but it can tremendously impactful!". As a result, the evaluation infrastructure optimized (or infrastructure itself was optimized to better capture) for interventions that are vertical (e.g. disease-specific), delivery-focused (e.g. an NGO does it rather than the government, through or in parallel to government channels), measurable through RCTs, short-causal-chain, and grant-dependent (e.g. attribution is easier).  Or, the evaluation and funding infrastructure optimized for the effective health charity question, rather than the question of development, with insufficient thinking about what the ecosystem-level portfolio should look like.

The approach made sense for a new movement/team that tried to find out how they can save the most lives with their money: GiveWell’s founders were investment bankers who didn’t have platform they can trust to understand where they can donate to, EA’s founding figures were philosophers who realized that they can save lives with their donations by investing in measurable cost-effective charities. Both of these actors helped save a lot of lives, and their approach is very reasonable within the context of what they were aiming to achieve, and they seem to have achieved that fairly well (e.g. helped a lot of people that really needed help). 

They were also able to update fairly well: GiveWell didn't start as a global health organization. Its 2008 recommendations included US programs like the Nurse-Family Partnership and KIPP. The convergence on global health wasn't (or doesn't seem to be) a founding assumption, it was the result of applying a specific methodology and seeing where it led. Then, they also realized that there were some more speculative grants worth making, and GiveWell Labs/Open Philanthropy spinned out of GiveWell, and Toby Ord published the Precipice. There was cause area expansion, but limited (and slow!) expansion within global health and development.  Additionally, EA ecosystem also ventured into careers beyond donations, and I think the career ecosystem also partially reflected the distinction: It focused more on the charity question rather than the development question.

But, now there is even more divergence from the vertical delivery approach as I'll further describe in section 5. As of late 2025, Health Systems Strengthening is also a focus area for GiveWell, and many other organizations have been focusing on harder-to-measure type of work. Still I think the development or HSS questions remains insuffuciently explored, and that it’s hard to dispute that the EA ecosystem overoptimized for measurable, vertical delivery interventions and underinvested in health systems strengthening interventions.

I also think it wasn’t optimal that the ecosystem decided to take HSS seriously (excluding a few grants) after the largest aid agency on Earth got shut down and many programs and systems collapsed.

4. The USAID Cuts Revealed the Distinction in Practice

For a long time, the global health community has been discussing localization, government ownership and various other questions. Some argued that NGOs’ building parallel systems were indirectly slowing or blocking sustainable progress, raising concerns with how important programs are solely and unsustainably funded by external actors. Some people argued that government engagement is pretty hard (which is true), many governments lack the technical and fiscal ability to maintain these programs, and how NGOs doing their own thing could be more efficient in certain cases.

In early 2025, USAID funding cuts destroyed many life-saving programs. Drug supply chains, technical assistance, health information systems, functions operating for decades, suddenly faced collapse, potentially resulting in the death of around a million people.

One could argue how this shows that their camps were totally right about the delivery NGO-central models were unsustainable and risky, and localization was, and is the way to go.

At the same time, in many countries governments were fairly slow to fill in the gaps in the aftermath of the aid cuts. So one could argue that this inability of governments to act swiftly shows why localization is hard, and why delivery NGOs are so prevalent (and needed), and why government ownership is easier said than achieved.

I would pull the rope sideways and argue: Yes, government engagement is hard, but designing programs in a way that doesn’t perish when your donor changes their mind is also not very easy. And when your donor does that, it's not unlikely that all the functions and programs you’ve built may perish pretty quickly. To be more direct, of course a country’s critical health system functions shouldn’t be completely reliant on external funding and external human resources, and even if you can’t think of a solution, this reliance is a significant problem. I probably disagree with EAs on what this point implies: Some people just should be thinking about how to build functional health systems (Type 3 HSS work), even if it won’t yield measurably cost-effective interventions immediately. This is just work that needs to be done, some/a lot of people are already doing it, and it’s better if it’s done rigorously.

When many programs perished due to aid cuts, EA organizations responded fairly rapidly. GiveWell made ~$53 million in grants. GiveWell (and I believe Open Philanthropy/CoGi) co-funded CHAI Technical Support Units in six countries (though you can read a critique here), and these units helped the government navigate the uncertainty caused by aid cuts. Founders Pledge and The Life You Can Save deployed a lot of resources fairly swiftly. I’m unsure whether it was an EA project, but Project Resource Optimization was also very valuable. Overall EA’s response to USAID cuts weren’t inaction, and seems like funders moved pretty swiftly.

But it happened reactively. The intellectual work, analysis and prioritization of HSS interventions, could have been happening for years. I may be totally wrong, but my low-confidence guess is that, when the crisis hit, there wasn't deep expertise on how to think about HSS interventions in general, or a framework for evaluating which systems functions were most critical, and probably a very limited pipeline of organizations to support.

And being late to systems work is especially costly, it requires years of relationship-building and local knowledge (e.g. country-specific tacit knowledge about how the system works), more so than the delivery work. However, GiveWell or Open Philanthropy doesn’t aim to do every single high-impact global health project, it’s an organization that has it’s own assumptions, history and scope. But those assumptions and history behind GiveWell aren’t that different from other EA organizations’ assumptions and history, and as a result the equilibrium had and still has a pretty big gap, a gap of meaningful engagement with the development questions, and insufficient testing about how NGOs can contribute to development beyond rather than effective charity. This lack of meaningful engagement seems problematic to me as:

Cost: Could the support provided by large iNGOs have been provided more cheaply? If large iNGOs deliver at higher cost than lean GiveWell or AIM charities, why assume cheaper models for systems support aren't attainable?

What works: What solutions or interventions can help governments build capabilities so that their various components of their health system won’t totally collapse once aid money goes away?  I don’t think this question is a novel contribution from me, but I wouldn’t be surprised if there was little research on this up until USAID cuts. This is also not a question where applying the regular cost-effectiveness frameworks may not yield the best answers.

4.1: Methodology Interlude: Cost-Effectiveness of Unlocking Government Programs

Asking how cost-effective a direct intervention is different from asking how cost-effective it is to influence how governments spend (including spending on cost-effective things). Neil Buddy Shah (then-CEO of IDinsight) illustrated this at EA Global 2018. IDinsight designed a rapid trial for $70,000 showing that "Mama Kits" increased facility deliveries in Zambia by 47%. The government then scaled the program with its own funding.

The reframe: How cost-effective was that $70,000 of evidence generation? The question shifts from "is this intervention cost-effective compared to bed nets?" to "how much government health spending did this philanthropic dollar shape?", and of course, to what extent, and what was the impact of the government expenditure. It’s hard to measure…

"There are clear benefits that such a small amount of money, whether it's for research or advocacy, has the potential to influence many multiples of government spending. However, from the point of view of an effective altruist that really wants to nail down, 'Is my money demonstrably improving lives?' There are clear challenges. Attribution is very tenuous... I think this is one of the areas where effective altruists need to be comfortable or grow more comfortable with uncertainty." — Neil Buddy Shah

Attribution is genuinely difficult. But the potential multiplier is large enough to deserve serious analytical attention. And beyond potential multiplier, I believe there is another value in HSS that I haven’t emphasized in this post, as I’m trying to turn this draft into something more palatable for the EA Forum:

Does success for EA looks like 100s of extremely cost-effective NGOs proliferating across South Asia and Sub-Saharan Africa distributing various health commodities until 2100, or governments (or local NGOs) having some capacity to make sure broad portions of their population aren’t dependent on funders far away for not dying from very preventable conditions?  
I’d certainly prefer it if my government had a good lead policy without needing LEEP to fly-in, and LEEP is actually successfully getting certain governments to eliminate the need for them. (Disclaimer: I used to work for LEEP, so I may be biased)

5. Recent Diversification Is Good, But Came Late and Remains Narrow

Now, I wish I posted this when I initially drafted it (though it was terrible at the time). As stated above, EA has and is already venturing outside of its traditional scope. There is now more health policy (and global health R&D work, which I’m not getting into), and that is great. Founders Pledge also seems to be engaging with harder-to-measure interventions. LEEP also seems to have updated EA community’s thoughts on the tractability of policy work. Rethink Priorities is producing various reports touching on the issues I’ve raised in this post, which indicates strong interest. AIM is looking to HSS and included sugary drink taxes in its current round, and has previously incubated Concentric Policies that advocate for excise taxes.

Now, one final point about the importance of delays: GiveWell made its first grant in lead paint ban back in 2017. It only made one other 250k$ grant up until 2021. The issue here isn’t GiveWell, or certainly not individuals behind the grant decisions: GiveWell’s framework and scope probably wasn’t a good fit for lead paint work, and to be honest I find it somewhat surprising that GiveWell made a grant for lead paint bans in 2017 despite it being highly uncertain (there was even less data on lead back then). GiveWell’s grants were a catalyst for helping build lead exposure as a mainstream issue. GiveWell is, and probably should be ambiguity-averse given it’s mission, and “The grant we made enabled an NGO to reduce lead paint share in the market, and if it wasn’t for them, lead paint share would have been reduced… probably much later?” is different from the rationale GiveWell can provide for scaling up SMC in Niger or improving immunization rates in Nigeria or India. 

Answers to the health system strengthening question will also be harder to answer, and rationales for investing in certain answers will probably look different. If EAs can be ambiguity-tolerant for lead exposure (or in cause areas besides health), why not explore harder to measure but extremely high leverage opportunities in HSS?

As the system is primarily optimized for measurable interventions that would make good answers to the charity question, the organizations reflect that. Or vice versa. And not all good ideas worth investing in are being captured by existing organizations. As… existing organizations can’t do everything. GiveWell may be an excellent research institutions that channels funding into life-saving opportunities, but it's mandate isn't solve everything relating to global health. The issue is you had/have limited work done by organizations or individuals on the development or health systems strengthening question. The deeper issue isn't any single organization's choices, and I don't think that blaming any single individual, funder or NGO for this gap makes sense. The deeper issue is insufficient thinking about what the ecosystem’s portfolio should look like (e.g. explore-exploit tradeoff), and overall absence of infrastructure that could channel funding and attention toward high-impact global health work (such as more speculative HSS work) that doesn't fit the dominant model.

6. Why This Gap Persists

Some people who may be sympathetic to my arguments may respond: "We're a small community, everything is neglected. These sound good, but there are many other things that are not being done as we’re fairly small. As you say, there are a few organizations, but we can’t expect them to do everything" That was a reasonable argument in 2010. It is weak in 2026 for an almost 15+ year old movement that has hundreds of millions of dollars flowing into global health and development, that still markets Global Health and Development when it’s presenting its identity to outsiders.  

None of the ideas I note here are novel, I’m not the first one to say “HSS is important!” on the EA Forum, I’m also not the first saying it can be really cost-effective.   The essence of what I'm saying: 

There are impactful projects that EAs tend to categorically disregard due to methodological and historical factors. Some of these impactful projects need the kind of careful reasoning, talent, funding, and infrastructure EA is well-positioned to provide.  Rather than the size of the community, incentives are usually a better way to understand why certain ideas are neglected (except they are bad ideas, and I hope my idea isn’t bad!)

Now I’ll drop the caveats and calibration, and just write bluntly about why I believe this gap still persists/exists, and lack of incentives/disincentives that’ll make it harder to tackle this gap:

High barriers to entry. Limited funder diversity, limited seed-stage funding, limited funding for certain modalities like country-specific organizations (e.g. intervention-specific organizations vs. country-specific, multi-issue organizations).

Within (and outside of) EA, the entry barrier for speculative/exploratory GHD work is high: EA Funds for meta-EA, animal welfare, and longtermism accept proposals; the GHD fund doesn't. If you aren't going through AIM, it's very hard to start a GHD organization. One could say, “Well, we know what works in global health much better compared to other cause areas, doesn’t a higher barrier for funding make more sense?”. Yes, I somewhat agree, GHD work should be more ambiguity averse compared to AI Safety or animal welfare, as there is more evidence of what works. I also somewhat disagree due to the implications of arguments I’ve laid out in this unnecessarily long post: EA GHD is too risk averse, and lack of exploratory funding is one reason for that, and because there is a lack of exploratory funding, EA GHD insufficiently invests in risky/complex but higher value projects. Now, I’m running a particularly exploratory GHD project, but still: I started a project within an existing organization which is now spinning out and I'm unsure whether we would have started it (i) If I wasn’t someone who has unfortunately high levels of tolerance towards personal risk and financial uncertainty (ii) if we’d had a more accurate understanding of the funding landscape.

Declining intellectual engagement. If a malicious AI deactivated David Nash and Nick Laing's EA Forum accounts, how many GHD posts would remain from 2025? There's insufficient discussion of promising areas, and no place to have discussions beyond a few passive Slack channels. Relatedly: some community members are cashing out retirement savings on AI timelines, but there's remarkably little public thinking on how AI progress should influence thinking about global development. (besides David Nash, and some other AI safety adjacent organizations)

The community's composition shaped its questions. EA global health has been predominantly shaped by Europeans and Americans with backgrounds in quantitative analysis, philosophy, or economics, more donors and charity evaluators than LMIC policy professionals or health governance experts. This probably resulted in less interest in or exploration of HSS within EA, or themes prevalent in non-EA GHD such as localization or government ownership. I think as EA is expanding in Africa and Asia, as EA GHD engages more with non-EA GHD ecosystem (e.g. lead exposure partnerships), with Probably Good’s work on GHD, this is currently changing, but infrastructure isn’t there. (I should note: I'm Turkish, which leaves me unable to say whether I'm European without at least three historians and two anthropologists present. And I’m a middle class person who is living in Istanbul who studied economics and analytic philosophy, so I partially reflect this profile myself, both intellectually and geographically. I’m also 24, and government engagement is a domain or function where experience can be pretty important)

Community-building dropped the ball on GHD. Animal welfare has Hive. AI safety has multiple communities and mentorship pipelines. For GHD, that infrastructure is largely nonexistent, no ongoing fellowship, no equivalent of SERI MATS or Oxford Biosecurity Group. This is not because EA is too small; it's because community-builders deprioritized GHD, and those who were interested in GHD failed to build meaningful infrastructure even though it’s fairly easy just to write an EA Forum post on GHD-relevant community-building resources. (including myself.)

These compounded. As a result, EA ecosystem’s entry to HSS was fairly late, and its entry strategy is still slower than it could be due to these factors in my very low-confidence opinion.

7. Some Potential Directions

Short answer: More people should think about how to think about this (e.g. which frameworks to use to prioritize). More people should think about figuring out what already works and can be replicated/tested/scaled further. 

HSS is indeed fairly saturated, so understanding what’s already happening and engaging with the field is pretty important in my opinion. Beyond HSS, the distinction between effective development and effective charity needs to be thought about more.

More specific recommendations that are absolutely low-confidence:

Frameworks before streetlighting. EA has a hammer, cost-effectiveness analysis for measurable, attributable, short-causal-chain interventions. Some HSS interventions may be nails for that hammer, but many are not. Cost-effectiveness will remain important, but it's worth asking whether the right analytical tools are being used. The HSS field has its own frameworks for thinking about health financing, service delivery, and governance, including more emphasis on reducing catastrophic out-of-pocket health expenditure or return on investment rather than $/DALY. Methodological reflection should come before methodological application. This report by R4D seems pretty relevant. Someone should probably write something about how to think about this. (e.g. compiling what exists, promising M&E directions for a particular domain of HSS)

Consider whether new risk-tolerant health incubators and evaluators would be valuable. If you even partially agree with this post, there's a lot of prioritization work to be done, and someone needs to do it. Some interventions are likely worth funding but too uncertain for GiveWell to recommend without risking its reputation as a rigorous evaluator. A newer, leaner organization could conduct that analysis at lower cost and with a higher tolerance for ambiguity, filling a gap in the ecosystem rather than competing with what already exists. AIM's recent moves toward HSS incubation are a good sign. This may partially overlap with Founders Pledge's Catalytic Impact Initiative, though I'm unsure. If the gap I'm highlighting is important, looking at AIM launching a new incubator for livelihoods (this), a new project may very well be valuable/worth testing. I think this could also be valuable for exploring AI*global health intersections (though in a careful way due to reasons Tony Senanayake outlines here.)

Generate evidence to de-risk scaling. HSS already has substantial funding globally. There are programs that appear to work, and there are examples of failed replication, and there are high-potential programs with very flawed/limited impact assessments. Rather than EA trying to pilot, and scale HSS interventions at massive scale from the beginning to the end, a potentially higher-leverage approach is generating the evidence that helps other, larger actors scale what works and avoid scaling what doesn't. More generally, a scale-centric or Renaissance Philanthropy-pilled agenda-setting approach, thinking less about doing something and more about getting influential actors to do something, seems promising, though I'm uncertain. Of course, this work would probably be even harder to measure.

Test different organizational models. The EA ecosystem overwhelmingly produces intervention-specific, internationally-managed organizations. Country-specific organizations remain almost untested. Langsikt is better-positioned to do policy work in Norway than an EA NGO based in the UK or US. Why wouldn't the same logic apply to health work in Kenya or Bangladesh? Testing this seems valuable. I'm unsure whether there are no strong arguments for not testing it. Connections, credibility, experience matters for policy work, and at least in some (or a lot of) countries, outsiders are at a clear disadvantage when it comes to government engagement.

Country-specific health organizations could take several forms: thematic (focused on maternal health, infectious diseases, or NCDs), functional (health financing and domestic resource mobilization, health information systems, or health technology assessment and prioritization of essential packages), or as implementation consultancies (helping international NGOs localize programs and build government ownership, Health Progress Hub is interested in indirectly testing a version of this through recruitment support).  

Lower the entry barrier. EA Funds accepts proposals for meta-EA, animal welfare, and longtermism, but not GHD. Seed-stage funding for exploratory GHD work is remarkably scarce. I'm unsure about whether EA Funds should open the pathway for GHD, but opening a GHD funding pathway for early-stage ideas, or high-leverage HSS (or non-HSS) ideas seem valuable. Career transition grants may also be interesting, though they should probably be targeted at more experienced professionals.

So, there are probably pilots or tests that would be extremely cheap relative to their potential upside. Some will fail, that's what exploration looks like. But alternative to non-trivially prioritizing HSS is an ecosystem that puts most of it's focus on tens or hundreds of cost-effective NGOs distributing health commodities across South Asia and Sub-Saharan Africa indefinitely, while thinking very little about building local organizational capabilities for strengthening health systems. The bets that work could help build robust health systems, or in other words: A world fewer people's survival is determined purely by the generosity of funders far away.