The five things that worked
1. Letting people opt out of matches without explaining. When we required a reason, people said yes to lunches they didn't want. We removed the reason field and the no-show rate dropped by half.
2. Sending the conversation prompts the night before, not the morning of. The morning-of version felt like homework. The night-before version felt like a heads-up.
3. Picking the restaurant for them. Two researchers picking a place together is a 45-minute Slack thread. Us picking is 30 seconds and they don't argue.
4. Capping each lunch at 90 minutes. Beyond 90, energy drops. The follow-up rate for lunches that ran 75–90 minutes was double those that went 105+.
5. Asking for one open question, not a bio. Bios produce small talk. Open questions produce conversations.
The three things that didn't
1. Theme nights. Methods Tuesday, Senior Researcher Friday — they all produced lower follow-up rates than mixed lunches. The theme constrained who showed up; the constraint was the problem.
2. Optional video introductions. Almost no one made one. The few who did made the lunches awkward — the other person had already "met" them and the in-person felt redundant.
3. Asking for follow-up commitments at the table. "Send each other a paper by Friday" sounded good in theory and produced exactly zero papers. The follow-ups that happened were the ones nobody asked for.
The one we keep getting wrong
Matching new members with each other in a city. We default to it because it feels safer — both people are figuring out the format together — but the data says it's the lowest-yielding pairing we run. New members do better with someone who's done two or three lunches before. We know this. We still under-do it because the operations are slightly harder. That's our problem to fix, not yours to absorb.
What's next
More cities. More follow-up data. We're starting to track six-month follow-ups instead of one-week — early signal is that the lunches that lead to year-long collaborations look different from the lunches that lead to a single follow-up email. We'll write that up when we have enough of it.
If you're a researcher who wants in on the next thousand, the next round of lunches is open for sign-ups. Apply at /join.