Making "still adopted at a year" the metric
We deleted our engagement dashboard and rebuilt reporting around a number that takes twelve months to know. It changed which features survived.
Every metric we tracked in our first two years was available the same day: sessions, saves, enquiries sent, listings viewed. Every metric that describes whether we did any good takes months. We optimised the fast ones for two years, because they were there.
The number
It is now one line at the top of every internal dashboard: adoptions completed twelve months ago that are still in place today. We call it held placements. In 2026 it sits at 90.9%, against 85.8% for the 2024 cohort.
A metric with a twelve-month lag is useless for steering week to week, which is the standard objection and a fair one. Our answer is a set of leading indicators that we have actually validated against it, rather than assumed:
- Whether the adopter met the animal more than once before taking them home. Strongest single predictor we have found.
- Whether the profile disclosed a behavioural flag. Disclosed flags correlate with fewer returns; hidden ones with more, which is the same finding stated twice.
- Whether the week-one follow-up was answered at all. Non-response predicts a return better than any answer given.
A leading indicator you have not checked against the outcome is a guess with a dashboard.
What it killed
A "similar animals" carousel on the profile page. It lifted saves 19% and enquiries 7%, shipped in 2024, and looked like an obvious win. When Levente ran the cohort analysis a year later, animals adopted via the carousel returned at 16.8% against 9.4% for the rest. It was, in effect, very good at getting people to enquire about an animal that was not the one they came for.
We removed it in March. Saves fell back. Nobody outside the company noticed, which is the usual fate of a feature that was never actually working.
What it protected
The follow-up flow costs us real money — it is largely staffed by people, not automation — and by every fast metric it is a waste. It has no effect on sessions and cannot possibly drive acquisition. By the only number we now care about, it is the single most valuable thing we have built.
That is the whole argument for changing the metric. The dashboard decides which arguments you can win internally, and we kept losing the right ones.