The interesting thing about Airbnb, Facebook, and Slack is that none of their metrics directly measured whether people loved their products. Instead, they tried to identify a behavior that suggested users were actually getting value. But this is also where Goodhart’s Law comes in: once a metric becomes a target, teams can start optimizing the number rather than the experience it was supposed to represent.
Airbnb : Nights booked
Airbnb focused on nights booked because it represented an actual transaction between hosts and guests. Unlike app downloads or website visits, a booking suggested that someone trusted the platform enough to spend money on a place to stay. The problem is that increasing bookings doesn’t necessarily mean improving the experience. A team could push aggressive discounts, encourage bookings through misleading listings, or make cancellations difficult. Nights booked might increase, but guests could end up disappointed and hosts might lose trust in the platform.
Facebook : 7 friends in 10 days
Facebook’s early growth team used this metric because they believed people who connected with seven friends within their first ten days were more likely to stick around. It was a proxy for reaching the point where Facebook became socially useful. However, a team could game this by aggressively suggesting random friends or pressuring users to send requests. Suddenly, users have more connections, but those connections might not mean anything. The product becomes focused on collecting friends rather than building meaningful relationships.
Slack: Teams sending 2,000 messages
Slack’s metric was based on the idea that teams sending around 2,000 messages had started incorporating Slack into their everyday workflows. The number was a proxy for adoption and collaboration. But a team could optimize for message volume by encouraging unnecessary notifications, splitting conversations into multiple messages, or making communication more fragmented. This could make Slack look more successful while actually making people less productive.
What I would take away
I think the biggest issue with these metrics is that correlation doesn’t necessarily mean causation. Facebook users who made seven friends might have stayed because they already had strong social networks, not because reaching seven magically made the product better. Similarly, Slack teams sending 2,000 messages might already have been highly collaborative.
As a PM, I would still use these metrics, but I wouldn’t evaluate success through them alone. I’d want to look at retention, user satisfaction, and whether people are actually accomplishing what they came to the product to do. Otherwise, we risk designing products that are really good at producing numbers but not necessarily good for the people using them.
