40,000 Signups, No Change in Retention

40,000 signups landed over one weekend. A tweet went semi-viral on Saturday, and by Sunday night the dashboard looked like a hockey stick that had found religion. Monday morning the team was already debating which feature had done it, and how to do it again.

Six weeks later, weekly active users sat within four percent of where they had been before the spike.

The 40,000 behaved the way a crowd summoned by a tweet behaves. They signed up, looked around, skipped onboarding, and never came back. The product had not changed. The audience had not changed. What changed was the team's confidence in a story about what worked, and that story was mostly wrong.

That is the expensive part. A spike rewrites the team's intuition about which lever moved which number, and the rewrite usually points the wrong way. The next quarter of roadmap gets spent chasing the moment. The slow, unglamorous work of fixing retention gets deprioritized, because everyone is busy reproducing an accident.

Product-market fit does not show up at the top of the spike. It shows up six weeks later, in the shape of the line where the spike used to be. If retention fell back to baseline, the product did not move; the audience did, for reasons no one at the company controls. If retention stepped up even slightly, something happened, and the real question is what made the step and whether it repeats.

The team eventually dug into that four percent. Three quarters of it came from a single reply thread on the tweet, a thread describing a use case no one had marketed. That use case became the next quarter's positioning. It was the only real signal in the whole spike. Everything else was weather. The same trap shows up when one account distorts the read on demand, the way it did for the biggest customer buying the wrong product.

Do not optimize for the spike. Optimize for what survives after the spike forgets you.