What a Network Is Worth
Forty people are already on your team's tool. You ship one feature — group channels — and not a single new person signs up. The number of ways that network can be used goes up by more than a billion times. Same people, same headcount, same licence bill. Guess the jump first, then switch the capabilities on yourself.
A tool with 40 people on it lets any two of them message each other — that's 780 possible conversations. Now you switch on group channels: any set of people can form a room. Nobody new joins. How many usable combinations do you now have, compared with 780?
Bars are on a log scale: each further twelfth of a bar is another ten times bigger. Metcalfe barely clears a quarter of the track and Reed runs to the end — that's the tell. These aren't three estimates of the same number, they're three different questions about the same people.
Reed counts every group that could form. Almost none of them ever will — people join a handful of rooms and abandon most of those. Drag the slider and watch the live count sit at a rounding error of the ceiling.
This is the honest counterweight to both laws. Briscoe, Odlyzko and Tilly argued in 2006 that real networks track something closer to n·log n, because links are not equally valuable — your hundredth contact is worth far less than your first. Metcalfe's n² was the arithmetic behind a lot of 1999 valuations, and the correction arrived with the rest of the dot-com bill.
So use the laws the way an engineer uses a load rating: the ceiling tells you which design is worth building, and it tells you nothing about what the network earns. A group-forming product has room to grow into that a pairwise one structurally does not — that is the whole claim, and it is enough.
Who is allowed to make a channel
Email is a pairwise network with a distribution list bolted on; Slack and Teams are group-forming. The org that locks channel creation behind an IT ticket has bought a group-forming tool and capped it at Metcalfe — then wonders why adoption stalls after the pilot.
Pairs are easier to copy
Uber matches a rider to a driver: valuable, but pairwise, and a competitor can rebuild it with a price war. Discord, Reddit and GitHub let users build the groups themselves — and what defends the platform is the groups its members made, which no rival can undercut.
An open API is a permission, not a feature
Publishing an API lets combinations form that you never designed or resourced. That is the 2ⁿ term arriving in product form — and it's why platform openness usually beats another quarter of features for anyone trying to become infrastructure.
Governance scales with the same term
Every subgroup you permit is also a room nobody is watching: shadow processes, undisclosed data, harassment, records you can't produce in discovery. Reed's upside and your moderation and compliance load come out of the same exponent — budget for both or permit neither.
Three laws, three different questions about the same room of people: Sarnoff counts ears, Metcalfe counts pairs, Reed counts groups. Which one governs your product isn't decided by how many users you have — it's decided by what you let them do together. So before the next growth push, ask the cheaper question first: are we leaving the group-forming term switched off?