Replay tools are good at the thing they are good at: showing a gesture that a funnel step cannot narrate. A thumb that hunts for a control, a paywall that covers the one button that would have counted as value, a permission dialog that interrupts the first session. We use clips in Event Grammar Intensive as exhibits, not as evidence of magnitude.
The sampling problem you already know
Most teams cannot store every session. Sampling arrives, then a well-meaning filter for “angry taps”, then a highlight reel that confirms the hypothesis that entered the room first. None of this is wicked. It is simply not a measurement system. If your activation rate moved, replay will not tell you whether it moved by four hundred people or by four, and it will not tell you whether Android moved while iOS stayed still.
Privacy is not a footnote
Mobile replay sits closer to the body than a web recording. Notifications, message previews, photos — the surface of a phone is not a marketing site. Store review and UK data protection both have opinions. In Funnel Signal Studio we treat masked fields as part of instrumentation, not as a legal afterthought. If you cannot describe what is redacted, you do not have a telemetry design. You have a hope.
What to keep
Keep replay for diagnosis after a quantitative surprise. Keep a short library of clips attached to a ticket, with the event names that should have fired underneath. Then go back to the dictionary. The strategy is the named action: first saved item, first completed transfer, second open inside a window you can defend. Watching someone fail is moving. Counting the failure twice, on two platforms, is the work.
If your current stack offers replay and almost no trustworthy events, start with the events. We will not take a Studio Seat whose warehouse is empty and try to fill it by watching videos. That limitation is in the syllabus on purpose.