How comparable AI-safety organisations actually acquire and retain users — and what that means for Lens over the next quarter. Five findings, each tied to a concrete near-term move.
Lens's growth model has to differ from BlueDot's by structure. We can't use selectivity as the marketing, and paid acquisition gets us a CAC the unit economics struggle with. Effort should go to four channels, in priority order.
Measured iteration on DM open rate and conversation depth. Cheapest credible work, data already internal.
Capture pre-qualified rejects from upstream selective programs. Our open-enrolment model is the only place this fits.
Interactive demos that travel beyond the EA network — the channel that reaches new audiences.
Three warm intros, zero cost, hard kill criterion. The only channel where the team has a network advantage.
Paid acquisition stays off the table until the funnel mechanics work.
The two orgs with real budgets — BlueDot ($25.6M from Open Phil) and 80,000 Hours ($2.7M marketing grant, 2025) — still report organic referrals, podcast, newsletter and SEO as dominant. Paid is supplementary even when affordable. AI Safety Quest, AISafety.com, Successif, HIP and CivAI spend essentially zero.
Defer paid social and search. With a marginal cost near $10/learner and open enrolment, paid CAC has to be very low to justify itself — and the evidence says we'll acquire most users elsewhere.
BlueDot, MATS and Iliad use selectivity as the marketing — scarcity signals value. Lens has rejected that lever. What's left: shareable artifacts (CivAI demos reach TIME and policymakers), credentialed output (Algoverse routed 230 students / 60 papers into NeurIPS 2025), and persistent value (80k's newsletter retains users for years).
Pick deliberately. Cheapest fit is persistent value — Lens Coach is that layer if retention works. Most differentiating is shareable artifacts. Probably both: Coach primary, artifacts as a periodic drop.
BlueDot rejects ~1,000 applicants per cohort; 80k advising is application-gated; AI Safety Quest is capped at ~400 calls/year. Each drops motivated, pre-qualified users out the bottom with no next step. Incumbents won't build a consolation handoff — sending rejects elsewhere dilutes their selectivity signal.
Make cross-org referral architecture a Q3 priority. We have starting points (Successif, Effective Thesis, AFFINE, MATS). The work is operationalising — a handoff protocol, tracked volume, one owner — not opening new doors.
80k cites its advising plan-change rate (76%) and newsletter open rate when fundraising — not podcast plays. BlueDot's alumni Slack (7,000+ members, active job/policy traffic) is the retention asset, not the application count.
Lens's equivalents are Coach DM open rate (21%) and conversation depth (anon users drop mid-conversation). A measured pass through Coach logs, an A/B test on the opening DM, and a drop-off analysis is the cheapest credible growth work available — and the data is internal.
Every major comparator is US/UK/EU-headquartered. Across all of Africa, selective AI-safety cohorts (ILINA, AISSA, Cooperative AI Research Fellowship, ML4Good Cape Town, EA JKUAT) total under 500 seats/year. EA Nigeria has multi-city presence (Abuja, Lagos, Ibadan, Kano) and ran EAGxNigeria in 2025. Deep Learning Indaba 2026 (Lagos, 2–7 Aug, ~1,000 African ML researchers) is the year's highest-leverage African event.
Run a small Africa experiment in the next 30 days: three warm introductions, zero cost, hard kill criterion. The cheapest geographic experiment Lens can run — and the only one where the team has a network advantage (board role at AI Safety East Africa; Kampala and Lagos connections).
| Organisation | Paid acquisition | Selectivity | Referral / community | Content / artifacts |
|---|---|---|---|---|
| BlueDot | supplementary | core | core | some |
| 80,000 Hours | funded, secondary | n/a | core | core |
| MATS / Iliad | none | core | some | some |
| CivAI | ~zero | n/a | some | core |
| Algoverse | ~zero | some | some | core |
| AI Safety Quest | ~zero | n/a | core | some |
Qualitative read of stated dominant channels. Lens cannot use the "selectivity" column — which is why the green elsewhere matters.
Selective programs use low acceptance rates as a scarcity signal. BlueDot is markedly less selective (~20–25%) than MATS/Iliad — a point worth stating accurately if it comes up. Values approximate.
Annual seats in selective African AI-safety cohorts. Combined capacity is under 500 — the gap an open-enrolment product can fill without competing with incumbents.
Coach DM open rate, current vs. illustrative target. The single cheapest number to move — an A/B test on the opening DM is a one-week experiment.
Measured pass through Coach conversation logs — where do anonymous users drop?
A/B test the opening Coach DM to lift the 21% open rate.
Conversation-depth analysis — identify the drop-off point and the message preceding it.
All internal data, no spend, fast feedback. Higher-leverage than acquisition at this stage.
Three warm introductions via existing networks (AI Safety East Africa, EA Kampala, EA Nigeria leads).
Frame as "how advanced AI affects African governments, labour, research" — not "existential risk" cold.
One conversion to a Lens-cohort pilot conversation, or park it.
Every comparator is US/UK/EU-based. The African AI-safety ecosystem is under-engaged and not contested by incumbents — and Lens has a real network there that no competitor can match.
Treat as a low-cost experiment, framed for demand that actually exists locally — not a bet on frontier x-risk appetite.
The right channel mix for Lens's specific use case, in priority order:
Iterate on open rate and conversation depth. Internal data, this week.
From upstream selective programs. Operationalise existing partners.
Demos that travel beyond the EA network.
One bounded test where the team has a network advantage.
Paid acquisition stays off the table until the funnel mechanics work.