Retention is the only vote that counts

There is a number on our internal dashboard that nobody outside the company sees, and it is the only number I check every single day: the share of users who started eight weeks ago and ran Y0 this week, unprompted. No campaign, no re-engagement email, no push notification — just a person, on a Tuesday, choosing us again.

Everything else a startup measures can be bought or flattered. Signups respond to launch posts. Daily actives respond to notifications. Revenue responds to discounts. Week-eight unprompted return responds to exactly one thing: whether the product is genuinely, repeatedly useful. It is the only metric I know of that cannot be bribed.

Acquisition is what you say about the product. Retention is what the product says about itself.

Why eight weeks

Novelty has a half-life of about three weeks for a productivity tool — we measured it. Cohorts one through three flatter you; the curve is still riding the launch dopamine. By week eight, the user has had a vacation, a crunch, a slow Friday, and at least one moment where your product failed them. If they are still here, the habit is real. Our cohort curves go flat at week six or seven; the height of that plateau is, as far as I am concerned, the value of the company.

What this does to roadmaps

Running on retention changes what you build, not just how you score it. Feature ideas at Mynd Labs are argued in a specific format: which week of the lifecycle does this touch, and which failure does it remove? A flashy capability that improves day-one wow but adds a permission prompt to week-five usage is a net negative, and we have killed several. The features that move the plateau are almost embarrassingly boring — the runtime remembering an invoice template, a trace loading in 200ms instead of two seconds, a context fetch that no longer asks twice.

• ↳We ship to the plateau, not to the launch tweet.
• ↳Any feature that taxes a retained user to impress a new one is rejected by default.
• ↳Re-engagement spend is capped near zero — if the product cannot pull people back, paid pushes only disguise the disease.

The vote that compounds

There is a strategic layer too. Our entire ecosystem thesis — eight lanes, entered in sequence — assumes each lane is funded by the trust and margin of the previous one. That chain has retention as its first link. A retained user gives you the two currencies an ecosystem is actually built from: permission to ask for more context, and patience when the next product is version one. Neither is for sale at any CAC.

So when people ask how the grand ambition is going, I give them the small answer. The plateau moved up again this quarter. That is the whole report.