Session replay is a lamp. You point it at a corner of the product and sometimes you see a user hesitate on a permission dialog you had forgotten existed. That is qualitative work. It is closer to sitting behind the glass in a research lab than to App Analytics as we teach it at the desk.
Trouble starts when the lamp is given a number: replays watched, rage clicks counted, “frustration score” averaged. Those figures inherit the biases of who was sampled, which sessions the vendor kept, and which employees had time to watch. They move when the vendor changes a heuristic. They do not move, reliably, when the product does.
We still use replay in Funnel Forensics, with a constraint: you may cite a replay as an illustration after the event counts have been reconstructed on paper. You may not replace the reconstruction with a montage. A montage persuades. It does not measure.
There is also an ethical remainder. Replays are recordings of people who did not sit for an interview. United Kingdom teams should treat them with the same caution they would bring to any other recording: retention limits, access control, and a willingness to turn the lamp off when the question has been answered. Watching for sport is not analysis.
If your north star requires a replay dashboard to “explain movement,” the star is already too vague. Choose an event a computer can count without an audience, then use replay — sparingly — when the count surprises you.