I’m currently a co-director at EA Netherlands (with Marieke de Visscher). We’re working to build and strengthen the EA community here.
Before this, I worked as a consultant on urban socioeconomic development projects and programmes funded by the EU. Before that, I studied liberal arts (in the UK) and then philosophy (in the Netherlands).
Hit me up if you wanna find out about the Dutch EA community! :)
Looks fun! I especially like ideas in the media xover events sheet
Oh perfect, I’ve been looking for a podcast that fills exactly this niche :)
I listened to your most recent episode and really enjoyed it, thanks for producing it!
Just jumping in to say I know both Stijn and Remmelt, so can confirm they’ve been heavily involved in EA (Stijn still is, Remmelt I’m not so sure, but he definitely co-founded EA Netherlands — I last spoke with him in 2024).
From what I know, Remmelt was (is?) heavily involved in StopAI.
I don’t want to speak for them but I’m pretty sure they’d both strongly oppose violent acts. Remmelt did so on Twitter here.
I’m not endorsing the rest of your claims in the post, just wanted to corroborate this specific part.
Oh yes for sure. But I assume there will be another edition of the survey. There’s a lot of vibe-based takes in this piece, which is fine, but it’s also nice to look at the best data we have, which is probably this survey (and future editions of this survey). I want to know whether Harry thinks he will update his views as new data comes in.
How much weight do you put on info like the community satisfaction section of the EA survey?
The data from 2024 suggests overall satisfaction has stayed stable:
And self-assessed likelihood of retention doesn’t seem to vary across cohorts:
Obviously, there’s a lot more to your post than worries about community satisfaction.
Thanks Jari, great questions! Also congrats on your first comment!!
1. National orgs haven’t done much experimentation with programme development, so there isn’t much to copy. EA Switzerland recently bolted on an impact cohort to their trad intro programme, which looks promising, and EA UK are going to try and replicate it, so one option would be to just copy that, and I’m considering that. In the past 6 months or so, EA Israel has been trying to build an intro programme from the ground up, but that’s more focused on students and young professionals. CEA’s virtual programme hasn’t been updated significantly in years. CEA’s bootcamp or HIP’s IAP are interesting candidates for copying BUT they’re online and are mostly about helping people switch careers, rather than generally helping people increase their impact/onboarding people to the community.[1]
2. We think we might be able to do more. Last year, we experimented with various GTM strats, and got some decent results (140% increase in completions). But our CAC was a bit high, and this might be an indicator we don’t have PMF. We’re also worried that, as our target audience has changed, the intro course is less likely to produce the outcomes we’re looking for.
It might be the case that we end up doing something similar regardless because, when we interview our target audience, we discover they just really want career programmes. If that’s the case, then we should probably just copy what HIP and CEA are doing (with some adaptations).
My personal messy thoughts on some of the things EA Netherlands (and maybe other community building orgs?) should be doing in the near future (building on our recent post). Sharing to get input. Please tell me what you think I’m getting wrong.
Community building, not talent placement. EAN is primarily a community-building organisation. Our theory of change is to grow community capital — the community’s capacity for future impact, which we take to be roughly (sum of individual career capital) × coordination ability. This is the stock we’re building. Members deploy it over time through three medium-term outcomes: career switches into impact-relevant roles, significant donations to effective organisations, and community organising or advocacy for EA-aligned causes, solutions, and tools. (Organising and advocacy are partly deployments and partly investments back into coordination ability — meta-work, worth flagging as such.)
Over a longer horizon, those flows aggregate into population-level outcomes: shifts in awareness and perception of EA and EA-style thinking, and eventually norm and institutional change.
Career switches and donations in any given year are how we tell whether the stock is being built well; they’re not what we should be optimising for.
This is the difference from talent placement. Placement orgs — BlueDot for AI safety upskilling, AIM for entrepreneurship, MATS for AI safety research — convert specific kinds of career capital into specific roles on roughly annual cycles. That’s a different job, and it’s done well elsewhere. EAN’s contribution is to grow the underlying stock: more members, more career capital per member, more coordination ability across them. What makes this worthwhile is what the approach produces over long horizons: engagement spread across many roles rather than concentrated in one bet, ecosystem infrastructure that no single placement focus would build, values that persist when people switch careers, and the candidate pipeline that placement orgs themselves depend on. Maybe the Dutch ecosystem does need a placement org, but that should probably be an EAN spinout (as was the case with Doneer Effectief).
Who we target with our outreach. EA-curious people with skillsets identified as being particularly valuable by sources such as the meta coordination forum survey.
Product-Market-Impact Fit (PMIF). Standard PMF asks “do people love and use it?” EAN’s job is harder. PMIF resolves into three fits at once: people engage (PMF); the people engaging could plausibly drive the outcomes above (market-impact); and engagement actually causes the outcomes rather than merely correlating (product-impact). (h/t to people like Peter McIntyre and Jamie Harris, who have been pushing for more of this in EA community building).
Supplement our QuIP research with surveys and interviews of people from the Dutch EA community who have gone on to do cool things. Our recent QuIP research interviewed a selection of programme participants. However, it didn’t select based on impact achieved, just engagement with our work. This would complement it by surveying and interviewing Dutch EAs who have done genuinely impactful things post-engagement. The question we’d be asking: what helped, what hindered? Builds on Open Philanthropy’s 2020 “What helped and hindered our respondents” and CEA/RP’s 2024 “What helps people have an impact and connect”. The distinctive contribution is Dutch-specific signal. Mix surveys for breadth and interviews for depth.
Run generative interviews with our target group. For each priority segment, conduct a series of interviews with people who fit the profile: in the skill bucket, at the right career stage, Dutch-based or Dutch-relevant, currently unknown to EAN, EA-curious, and looking for community and learning. We’re looking to learn about their needs, blockers, jobs-to-be-done, and unmet desires — what they’re looking for, what they’ve tried, what would resonate, what almost made them give up trying.
Combine both, generate solution ideas, test cheaply. The cool-people analysis gives us info on what has tended to help people who succeed in the Dutch context. The generative interviews give us within-segment specificity (what this particular cohort needs). Combine them to generate 3-5 distinct candidate solutions. Then test cheaply before building anything: landing-page tests (a one-pager describing the imagined programme — do they apply?), concierge MVPs (deliver the service entirely by hand to the first five users), interview the solution (“would you use this? what would make you not? what would you give up for it?”). Validate against Cagan’s four risks before scaling: value, usability, feasibility, viability. Pre-commit kill criteria for every programme — including the intro course, which is our current on-ramp because it’s been the on-ramp, not because it’s been proven against alternatives.
Thanks for taking the time to anonymise and share this—good insights here so it’s great to have it in the public domain
Nice piece, thanks for sharing it on here!
Nice! I’ll be there, see you then :)
Great stuff, congrats on launching! Curious how you funded it—EA Australia’s general budget or a specific grant? Asking because we explored something similar at EAN and would love to compare notes. Happy to take it to DM if easier :)
From what I understand, Forum usage declined for a couple of years and has now stabilised.
My guess is that, since the Forum is mostly used by fairly engaged EAs, the post-FTX slowdown in top-of-funnel growth is showing up here on a lag. People don’t usually arrive on the Forum cold; my impression is there’s typically a runway of a year or two between first contact with EA and posting or commenting in earnest. So a recruitment hole in 2023–24 produces a Forum hole now, even if intake has since recovered.
If that’s roughly right, two things should follow: the share of activity from accounts under ~12 months old will have dropped more than overall activity has, and Forum recovery will lag whatever recovery CEA is seeing at the top of the funnel by another year or two.
And then it’s probably compounded by dispersion — as you point out, a lot of the conversation that would once have happened here has migrated to LW, X, or Substack.
Good points! I’ve found the book ‘Change: how to make big things happen’ useful in my work at EA Netherlands. I wrote up a few takeaways in a comment here.
Yup, that’s a risk! I think you can mitigate it by learning how to do marketing well, but that’s easier said than done!
Thanks Vasco!
I think it isn’t used much by EA event organisers because it probably doesn’t really measure precisely what it claims to measure. However, I do think it’s nonetheless a reasonable measure of overall quality, and because it’s widely used outside the EA bubble, you can more easily use it to benchmark (compared to likelihood to recommend scores).
Oh, this is perfect timing. I had just ordered The Goal in an effort to get better at this sort of thing. Great to see how it helped someone like yourself—a decent signal that it’s a useful book to read.
Good to know! If it doesn’t get included in the EA Survey, we might consider doing it ourselves at the national level (M&E budget allowing...).
But obviously, international data that would allow us to compare across regions would be more useful.
I think the magnitude × direction framing is really useful, and I agree the risk is real.
At EA Netherlands, I’ve been thinking about this through the lens of Ben Todd’s notion of “community capital” — roughly, the stock of shared values, trust, human capital, coordination capacity, norm-following, and reputation that a community accumulates over time. The worry you’re describing is essentially that outreach erodes community capital. And it can — if you do it carelessly.
But my ambition is to try to monitor this over time. If we surveyed relevant aspects of the community periodically — tracking its values, the degree to which people are following community norms, the actions people are taking, the degree of interconnectedness between members, and the quality of human capital coming in — we might be able to detect whether outreach efforts are degrading the direction term, and course-correct if they are.[1]
If that’s feasible, it turns a binary question (“should we grow or not?”) into an empirical one (“is this particular form of growth maintaining alignment?”). Some forms of outreach might pass the test, and others might not. But you’d want to check rather than assume.
I think the implicit model in your comment is one where we have to choose — stay small and aligned, or grow and dilute. But perhaps there’s a third option: invest seriously in both outreach and monitoring community capital, and course-correct over time.
Out of interest, has this been considered, @David_Moss?
Thanks Joris, and an excellent question. The honest answer is that I can’t cleanly attribute the application increase to marketing. If I had to put a number on it, I’d estimate marketing contributed roughly 30–35% of the application increase — but this is genuinely uncertain, and I’d put the range at somewhere between 20% and 50%. Here’s the reasoning and the data behind that.
What we did differently this year
The standard promotional activities — Forum posts, reaching out to group organisers across the region and asking them to share the event — were roughly the same as previous years. We also ran paid ads in both years, but the 2025 campaign was substantially different in quality and targeting.
In 2024, the ad campaign was set up by a very capable student who threw herself into it but had no prior experience with paid advertising. She ran campaigns on both LinkedIn and Meta. In 2025, we hired a professional agencies to run the Meta campaign on Instagram and support us with organic content on LinkedIn and Instagram.
Here’s how the paid campaigns compare. In 2024, we spent ~€2,640 and generated 454,904 impressions and 4,162 clicks (CPM of €5.80, CPC of €0.63). In 2025, we spent ~€3,200 and generated 257,698 impressions and 6,442 clicks (CPM of €12.41, CPC of €0.50). CPM is cost per 1,000 impressions; CPC is cost per click.
So the 2025 campaign was more expensive per impression but cheaper per click — it reached fewer people but got more of them to engage. And the critical difference was what happened after the click: in 2024, there was no lead capture mechanism, just link clicks. In 2025, 709 leads were captured via instant forms and entered an automated email follow-up sequence.
I asked Claude to compare the 2025 campaign stats against industry benchmarks, and the results suggest the agency did a good job. Our CPC of €0.50 is well below typical Instagram rates (industry averages run $1.83–$3.35). Our CTR of 2.62% is roughly 3–10x the Instagram norm of 0.22–0.88%. And our CPL of €4.07 is a fraction of the average Facebook lead campaign CPL of ~$28. The one caveat is that instant forms with pre-filled data make conversion very easy, which flatters these numbers — the real question is always what those leads are worth further down the funnel.
On top of the paid campaign, we invested significantly more time in organic LinkedIn content with support from the agency. My personal profile generated 103k impressions over the campaign period (March–December), reaching 25,129 unique members, up 814% on the prior period. The EAN company page added another 26k impressions, with a clear ramp from ~500/month early on to 7,100 in November, dropping back to ~700/month after the event. We also distributed posters across university campuses in the Netherlands.
A caveat on the LinkedIn figures: the organic content also included campaigns for our intro course (which saw a 140% increase in completions YoY), so those numbers overstate the EAGx-specific reach. I can’t easily disaggregate the two.
The mechanics of the Meta campaign
We ran Instagram ads using Meta’s instant form format. Someone scrolling their feed sees an ad, taps it, and a form pops up within the app pre-filled with their name and email — they can submit with essentially one tap. Those emails then synced to a Mailchimp automation that sent two follow-up emails linking to the application page.
The funnel looked like this: ~50,000 people reached → 6,000 clicks → 709 instant form leads → 612 entered the email automation → ~30–40 click-throughs to the application page. We generated those 709 leads at €4.07 each.
We originally wanted to run a website conversion campaign, which would have let us track the full journey from ad to application. The EA Global website (where applications happen) doesn’t support a tracking pixel, so we built our own landing page — but then we messed up the pixel there as well. So we fell back to the instant forms, which gave us reliable lead data but broke the chain between ad exposure and actual applications.
What the feedback survey says
The post-event survey asked attendees how they heard about EAGxAmsterdam. Only ~2% of respondents credited Instagram ads and ~3% credited LinkedIn. On the surface, this looks like marketing had no effect.
But I’d be cautious about taking that at face value. Only 98 out of 159 respondents answered the channel question. It’s plausible that attendees who came via ads — likely newer to EA and less engaged — were less likely to fill in a post-event survey, meaning they’d be systematically underrepresented.
There’s also a basic attribution problem. Someone sees an Instagram ad, doesn’t click, then hears about the event from a friend a week later and applies. They’d truthfully answer “through a friend” on the survey, even though the ad planted the seed. Self-reported attribution tends to undercredit awareness channels and overcredit the last touchpoint.
How I get to the estimate
Taking all of the above together: we received 906 applications in 2025 versus 570 in 2024, an increase of 336 (59%). Attendance grew from 381 to 517 (36%).
One way to get at the question is to compare against the broader trend. CEA’s recent post reports that overall EAGx attendance grew 20% in 2025 (2,866 → 3,238). Amsterdam’s 36% attendance growth significantly outperformed this. And there are reasons to think the baseline is even lower than 20%: that figure partly reflects expanding from fewer to 10 events globally (including new ones in Nigeria and São Paulo), so the per-event growth rate may be close to flat. There were also factors working against us: EAGxBerlin was close to us on the calendar (competing for the same regional audience), and EA Connect — CEA’s largest ever virtual event — ran around the same time, potentially satisfying people’s appetite for EA events without needing to apply to an in-person conference. In 2024, EAGxUtrecht was in the summer, further from other events.
So Amsterdam significantly outperformed a baseline that was probably close to flat, despite calendar headwinds. The main local factors that could explain the outperformance are: Amsterdam being more well-known than Utrecht, and our marketing investment.
Of those 336 additional applications, how many did marketing drive? 709 people submitted an instant form after seeing an ad. But the gap between tapping a pre-filled form on Instagram and actually paying ~€120 and filling out a substantive application is wide. For a paid event with a non-trivial application process, a lead-to-application conversion rate of 5–15% seems realistic — that gives roughly 35–105 applications from Meta ads. LinkedIn and the poster campaign are harder to quantify: LinkedIn works as an awareness channel that primes people to respond when they encounter the event elsewhere, and posters are essentially untrackable. Together they probably contributed a few dozen applications at most that would show up in the data as “EA Global website” or “through a friend.”
Setting those against the 336 additional applications, and accounting for the confounders above, I’d estimate marketing explains somewhere in the range of 20–50% of the growth, with 30–35% as my central estimate. The range is wide mostly because the lead-to-application conversion rate is doing all the heavy lifting and I have no data to pin it down. I hold this loosely.
What would actually help us figure this out
The biggest gap is that we can’t connect our leads to actual applications. Fixing the tracking pixel or matching our 709 lead emails against the applicant list would give us a hard floor on ad-driven applications. It’d be great if the EA Global team could find a solution for this.
The other thing that would help is knowing whether Amsterdam outperformed or tracked with the broader trend. @Arthur Malone🔸 / @Kiryl Shantyka / @Niki Kesseler — would you be able to share application trends across EAGx conferences over the past couple of years? The overall EAGx attendance figure includes new events, so a per-event comparison would give a cleaner baseline. If Amsterdam’s growth was in line with returning events, that points to ecosystem-level factors. If it significantly outperformed, that helps isolate the local effect.
I think most of what you’re after already lives under the EA community building topic.
I’m co-director of EA Netherlands, and the main reason I think this work might be a good use of my time is precisely that it helps develop motivation for unusually effective altruistic action. So any time I’m discussing community building best practices, I’m in effect doing the thing you say should be the priority - ‘debating the development of motivational strategies for altruistic action’. There’s a fair amount of it about; it’s just filed under a heading you may not have looked at yet. And working out what to be motivated about is part of the same project, which is why those object-level debates about animal welfare, AI and so on aren’t a distraction from it—they’re the substance it’s meant to motivate.
I’d also gently push back on the opening premise. ‘The greatest good for the greatest number’ is the standard shorthand for utilitarianism, and EA isn’t the same thing as utilitarianism—so it’s not quite right to treat that slogan as the benchmark EA fails to meet. The only commitment effective altruism really requires is that helping others matters. Plenty of effective altruists aren’t utilitarians and care intrinsically about things like rights, freedom, or fairness; most people end up giving some weight to several ethical theories at once. There’s a good summary of this in the FAQs on effectivealtruism.org under ‘Is effective altruism the same as utilitarianism?’.