Product Analytics
2 post(s)
-
Inference cost vs UX: why cheaper AI models can quietly reduce engagement (Amplitude)
Amplitude ran a real production test swapping an agent’s model to cut cost. Conversion held, but latency doubled and people asked fewer questions. The point: your success metric needs to include ‘time-to-answer’ and ‘messages per session’, not just dollars and a top-line conversion proxy.
-
The 7% retention rule: why Day-7 retention is a brutal early signal (and how to move it)
Amplitude argues that ~7% Day-7 retention is a useful benchmark because first-week activation correlates strongly with longer-term retention. This summary focuses on the practical takeaway: instrument a clear activation event and shorten time-to-‘aha’.