What random coffee is, and the problem it actually solves
Random coffee pairs 2 people who would not otherwise talk and asks them to spend 30 minutes together, with no agenda. That is the whole ritual. It exists because on a distributed team the only people you speak to are the ones your work already requires you to speak to, and that turns out to cost more than it looks.
This page is about the ritual rather than our product. We make a Slack bot that runs it, so we are not a neutral source — the section on when not to bother is there because the honest version is more useful to you than the flattering one.
The problem
What goes missing when everyone is remote
In an office you meet people you do not work with by accident — a kitchen, a lift, a desk you walk past. Remote, nothing is accidental. Every conversation has to be someone’s decision, and the decisions that get made are the ones the work demands.
When Microsoft moved its whole workforce to remote work in 2020, researchers followed 61,182 of its employees for 6 months. The share of their collaboration time that people spent with colleagues outside their own group fell by about a quarter, while each group became more tightly connected inside itself. Communication also moved toward email and chat.
That is one company in one unusual year, so read it as a mechanism rather than a law: teams under remote work did not stop talking, they stopped talking across. Published in Nature Human Behaviour, 2022 — read the paper.
New teammates never get past their own team
Someone who joins in an office meets people by walking around. Joining a distributed team, they meet the 6 people in their standup and nobody else, and they spend months not knowing who to ask.
Nobody hears from the other side of the company
Support knows what customers are angry about. Engineering finds out a quarter later, through a roadmap. The fastest route between those two facts is two people who already talk.
The person who could unblock you is a stranger
Most work gets stuck on something one other person has already solved. Asking them costs nothing once you have met them, and feels like an imposition when you have not.
The failure nobody sees
These programs rarely fail at the pairing
Matching people is the easy part, and any tool does it in a second. The usual way one of these programs dies is quieter: the introduction arrives, both people mean to reply, the week fills up, and the round ends with nothing on either calendar. Then it happens again, and the conclusion drawn months later is that the team was not interested.
Almost nobody finds out, because nothing asks. A program that reports how many pairs it created is reporting the one number that was never in doubt.
What we do about it
After a round, the bot asks each person whether they met — met, not yet, or did not manage it. Whoever runs the program sees how many pairs actually met, round by round, instead of how many were created. And the introduction carries a link with a 30-minute slot already filled in, so the time gets agreed while both people are still reading the message. Whoever clicks it gets the event on their own calendar, and the Google version also invites the other person when we know their address.
None of that makes the number good. It makes it visible, which is what lets you change the rhythm, the group, or the decision to run this at all.
Honestly
When not to bother
Everyone already knows everyone
Below about 15 people this introduces two teammates who sat in the same meeting an hour ago. Run it when the company has grown past the point where one person can hold every name.
It is treated as mandatory
Make it opt-in every round, and say out loud that skipping is fine. A program people join because it would look bad not to produces attendance numbers that mean nothing, and quiet resentment that is hard to see.
It is expected to fix something else
This does not fix a trust problem, a management problem, or a team that is quietly overloaded. It adds conversations between people who were not having them. That is the whole claim.
Nobody checks whether it is still working
Every one of these programs decays. That is an argument for measuring it, not for assuming it either works forever or failed on the day someone stopped mentioning it.
Practical
How to run one without buying anything
Pick who is in
One channel is enough to start. Ask people to join rather than adding everyone, so the first round is made of people who want to be there.
Pick a rhythm and keep it
Every 2 weeks is a reasonable start. Weekly is a lot to ask of anyone with a full calendar, and monthly is easy to forget between rounds.
Shuffle, pair, introduce
A spreadsheet does this. Shuffle the list, pair people off, and send each pair one message with both names in it and something to talk about.
Get the time booked immediately
The introduction is the only moment both people are reading. If a time is not agreed then, it usually is not agreed at all.
Afterwards, ask whether they met
This is the step everyone skips, and it is the only one that tells you whether to keep going. Two rounds of honest answers is better evidence than a year of assuming.
A spreadsheet and a calendar reminder will get you through the first few rounds, and for a small team that may be all you ever need. What it will not do is ask 40 pairs whether they met and add the answers up.
How it looks from the inside, including the spreadsheet-to-bot story: what more than 60 coffee chats taught the person who built CoffeeSlack.
Or let a bot run the rounds
Your team types /coffee to join. Rounds go out on the rhythm you pick, pairs get introduced with a calendar link, and afterwards everyone is asked whether they met. Free for every team, no card, and no cap on how many take part.