Charity Entrepreneurship / Ambitious Impact

Stated Theory of Change

CE claims that the charity sector contains massive gaps that analytical, ambitious founders can fill. The mechanism: (1) CE's research team identifies neglected, high-impact interventions across multiple cause areas, (2) the incubation program selects and trains talented generalists, matches them with co-founders and ideas, and provides seed funding, (3) the resulting charities either become field leaders (40%) or generate learning value even when they fail (20%). The meta-level claim is that creating more effective charities is one of the highest-leverage things a person can do, equivalent to $200K-$400K in donations per staff member per year.

CE explicitly limits itself to cause areas where evidence-based feedback loops exist. Joey Savoie: "I basically think I don't find a really highly uncertain, but high-value expected value calculation as compelling." This means global health, animal welfare, mental health, and pandemic preparedness -- not AI safety, not longtermist interventions, not anything where the primary argument requires speculative expected value.

Revealed Theory of Change

CE's actions are broadly consistent with its stated theory. The 50+ incubated charities span exactly the cause areas claimed. The success stories (Fortify Health, New Incentives, LEEP, SWP) demonstrate the model working as advertised. The transparent failure (MHI) demonstrates the portfolio approach functioning as designed.

The main divergence between stated and revealed theory of change is the AIM expansion (2024-2026). The rebrand and proliferation of programs (AIM Grantmaking, AIM Effective Giving, Founding to Give) suggest the real theory of change has expanded from "incubate effective charities" to "build effective altruism infrastructure broadly." The for-profit accelerator (Founding to Give) in particular suggests interest in expanding the model's reach beyond traditional charity, though it was also the flashpoint for community criticism about not doing AI safety instead.

The CEO transition from Joey Savoie (philosophical founder) to Samantha Kagel (professional manager) may shift the revealed theory of change further toward operational scaling and away from intellectual innovation. Whether this is good or bad depends on whether the core model is now mature enough to benefit more from execution excellence than from continued intellectual development.

Key Assumptions

1. The charity sector has persistent gaps that analytical founders can fill.

• Evidence for: Strong. GiveWell's priority programs list, ACE's charity ideas list, and CE's own research consistently identify neglected intervention opportunities. Fortify Health, LEEP, and SWP all addressed genuine gaps.
• Evidence against: Some CE charities in research/advocacy have struggled to scale, suggesting the gaps may be more limited than claimed for non-implementation work.
• Testable: Yes. Track whether new CE charities find genuine gaps vs. competing with existing orgs.
• If wrong: CE becomes a training program rather than a gap-filling institution -- still valuable but with lower impact.

2. Generalist founders outperform domain specialists at charity creation.

• Evidence for: CE's 40% "big success" rate, and Sarah Eustis-Guthrie's endorsement that generalists with evidence-based mindsets can build effective programs despite lack of domain expertise.
• Evidence against: MHI's failure was partly attributed to insufficient domain expertise and too much deference to experts on critical assumptions (postpartum abstinence). Sarah wished she'd "spent more time in clinics, more time understanding exactly what was going on on the ground."
• Testable: Yes. Compare outcomes of generalist vs. specialist CE founders.
• If wrong: CE should weight domain expertise more heavily in founder selection, potentially narrowing its applicant pool.

3. CE's research methodology reliably identifies cost-effective interventions.

• Evidence for: Multiple CE charities have been independently validated by GiveWell, ACE, and Founders Pledge. The multi-tool approach (CEA + WFM + informed consideration + expert views) with explicit caveats about each tool's limitations is methodologically sound.
• Evidence against: CE's own cost-effectiveness estimates are "almost always too optimistic" and can be "off by hundreds of times." MHI's cost-effectiveness case fell apart when examined closely. WFM normalization has known failure modes.
• Testable: Partially. Can compare CE's ex ante estimates to ex post evaluations by GiveWell/ACE.
• If wrong: CE would need to invest more in research depth at the cost of speed, reducing throughput.

4. The evidence-first epistemology is the right approach for doing the most good.

• Evidence for: Within measurable cause areas, CE has an impressive track record of producing charities that meet high external evaluation bars.
• Evidence against: This epistemology systematically excludes high-uncertainty, high-importance cause areas -- most notably AI safety. If AI risk is the most important problem facing humanity, then an organization that explicitly cannot address it has a ceiling on its total impact regardless of how well it executes within its chosen domain.
• Testable: Only in retrospect.
• If wrong: CE's entire framework, while internally consistent and well-executed, would be aimed at a domain that matters less than the one it ignores.

Strengths

Track record is genuinely impressive. 50+ charities, $68M+ in downstream funding, $28M+ from GiveWell alone. Fortify Health reaching 6M beneficiaries/month. New Incentives becoming a GiveWell top charity with 3,000+ employees. These are not paper claims -- they are verifiable, externally validated outcomes.

Transparent failure culture. The MHI shutdown, discussed openly on 80,000 Hours, is almost uniquely honest in the nonprofit sector. Sarah Eustis-Guthrie's candor about yellow flags, modeling errors, and the emotional cost of failure -- combined with the community celebrating the shutdown rather than punishing it -- demonstrates a genuinely healthy institutional culture around failure.

Strong leverage ratio. Estimated $5-10M in lifetime seed investment generating $68M+ in downstream funding is a 7-14x multiplier. Even accounting for counterfactual funding (some of these charities might have been funded anyway), the leverage is substantial.

Intellectual humility about methodology. CE explicitly warns that its own cost-effectiveness estimates are "almost always too optimistic" and weights CEAs at only 20-33% of overall assessment. This self-awareness about tool limitations is rare and commendable.

Model transferability. The CE model has been adapted by others (Catalyze for AI safety, 100x Impact Accelerator) -- evidence that the incubation framework has value beyond CE's specific cause area choices.

Weaknesses and Risks

Zero AI safety engagement. This is not a "weakness" from CE's perspective -- it is a deliberate philosophical choice. But for anyone whose primary concern is existential risk from AI, CE has no theory of change that connects to this problem. The incubation model could theoretically be applied to AI safety (as Catalyze is attempting), but CE itself has no plans to do this.

Governance is insular. 4/5 board trustees have direct ties to CE or CE-incubated organizations. No public COI policies. Karolina Sarek simultaneously co-founds CE, chairs a fund that grants to CE charities, and sits on boards of CE charities. This is poor governance by any external standard.

No independent meta-evaluation. CE evaluates charities for a living but has never been independently evaluated itself. The irony is acute. GWWC notes this gap explicitly. Without an independent assessment of CE's cost-effectiveness as a meta-charity, funders are relying on self-reported success metrics.

Founder departure risk. Joey Savoie was the intellectual architect -- the philosopher who could articulate why CE exists and what its epistemic commitments are. Samantha Kagel brings professional management but may struggle to maintain the distinctive intellectual culture. The risk is that CE becomes a well-run incubator rather than a thought-leading institution.

Scalability ceiling for research/advocacy charities. CE's biggest successes (Fortify Health, New Incentives) are implementation charities that can scale by delivering physical goods/services. Many CE charities focus on research/advocacy, which is harder to scale and harder to evaluate. The "1B animals affected" number relies heavily on speculative estimates about shrimp sentience and MOU compliance.

Revenue source opacity. Only ~50% of CE's $6.3M FY2024 revenue comes from identified sources (OP, EA Funds). The other ~$3M is from unknown donors, making it hard to assess financial sustainability and incentive alignment.

Cross-References

Catalyze (AI safety incubator founded by someone who applied to CE) is the most direct cross-reference -- it is CE's model applied to the domain CE won't enter. Comparing Catalyze's outcomes to CE's over the next 3-5 years would be highly informative about whether the incubation model works for high-uncertainty cause areas.

GiveWell is CE's primary external validator. Multiple CE charities have been evaluated to GiveWell top-charity standard, which is the strongest possible endorsement of the incubation model's output quality.

Open Philanthropy is CE's largest identified funder and the primary funder of several CE-incubated charities. The relationship is symbiotic: OP funds promising new charities to test, CE supplies them.

80,000 Hours positions charity entrepreneurship as a high-impact career path, with CE as the primary vehicle. They have featured multiple CE founders on the podcast.

The broader EA ecosystem treats CE as a success story for the neartermist/GHD wing -- evidence that evidence-based, measurable impact work can produce impressive results without requiring speculative longtermist reasoning.

What Would Change This Assessment

• CE incubates an AI safety charity: Would signal epistemological expansion and make CE directly relevant to AI risk reduction. Currently zero probability.
• Independent meta-evaluation finds CE's cost-effectiveness is mediocre: Would undermine the core case. If CE charities perform only marginally better than counterfactual giving, the incubation overhead isn't justified.
• Post-Joey leadership crisis or mission drift: If the AIM expansion results in diluted focus and declining charity quality, the track record argument weakens.
• Catalyze succeeds with CE's model applied to AI safety: Would prove the model is transferable to high-uncertainty domains, potentially making CE's refusal to enter AI safety look like a missed opportunity.
• Multiple CE charities revealed to have overstated impact: Would undermine the track record argument. The shrimp sentience challenge is the most live version of this risk.

Self-Critique

Weakest claim: My assessment that CE's governance is "poor by any external standard" may be too harsh given that small UK charities commonly have founder-dominated boards, and the EA ecosystem's interlocking directorates are structural rather than corrupt. The cross-organizational ties may reflect a tightly-knit community rather than governance failure.

Potential bias: I may be underweighting CE's track record because of the AI safety lens. From a global health and animal welfare perspective, CE is arguably one of the most impactful organizations in the EA ecosystem. My framing -- emphasizing what CE doesn't do (AI safety) over what it does (measurably reduce suffering) -- reflects an AI-safety-researcher lens more than an objective assessment of CE's importance.

Missing sources: I was unable to access the full text of several EA Forum critical posts (due to bot detection), including "Most research/advocacy charities are not scalable," "Why isn't there a CE program for AI Safety," and "AIM launches for-profit accelerator instead of AI safety." These might contain stronger arguments than what I reconstructed from search snippets.

What a thoughtful disagreer would say: "You're applying AI safety criteria to an organization that never claimed to work on AI safety. CE should be evaluated on its own terms -- as a meta-charity for measurable, evidence-based cause areas -- and by that standard it's one of the most successful organizations in EA history. The fact that it doesn't address your preferred cause area is not a weakness of CE; it's a limitation of your analytical framework."

What would most change my view: An independent, GiveWell-quality meta-evaluation of CE itself -- assessing dollars-of-impact-generated per dollar-of-incubation-cost. If this showed CE's meta-charity cost-effectiveness is genuinely competitive with direct giving to GiveWell top charities, my assessment of CE would shift substantially upward. If it showed CE's overhead makes it only marginally better than direct giving, the incubation model's case would weaken.