Coded Session Recording Review With Denominators

A recording review is a qualitative study with a sample, not a highlight reel. The usual version — someone watches thirty sessions and writes up the four they remember — produces stories nobody can argue with because nobody can check them. This one produces coded observations with a denominator. Address entry abandoned mid-field in 11 of 84 mobile checkout sessions is a finding. Users seem confused by the address form is an anecdote wearing a finding's clothes. Sampling: Sessions are drawn deliberately, never from whatever the tool lists first. Strata are device class, entry channel, outcome — abandoned at a named step, completed, returned inside the window — and session length, because both the 12-second bounce and the 9-minute struggle carry information the median session does not. Watching continues past saturation in each stratum, which in practice means roughly fifteen consecutive sessions producing no new code. Whether that lands at 60 sessions or 200 depends on how varied your traffic is, and it is not a number I will quote before I have started. Coding: Everything that matters goes into a shared taxonomy with a timestamp and a link to the recording, so anyone on your team can open the moment and disagree with my reading of it. Codes stay behavioural: reversed a selection, scrolled past then returned, opened a second tab, abandoned on a validation error, tapped a non-interactive element three times. Interpretation happens once, at the end, on top of counts, and it is labelled as interpretation so it can be rejected without rejecting the data. What tends to surface: Not the things teams expect. Validation firing before a field is finished. A filter state surviving back navigation and silently hiding inventory. A payment sheet dismissed and never reopened. An app deep link landing the user on a store listing rather than the item they tapped. On furniture catalogues the recurring one is the delivery estimate rendering after the price, which sends people out to a search tab and frequently to a competitor. These cost little to fix and almost never appear in a backlog, because aggregate analytics cannot see them at all. What comes back: A coded observation table with frequencies and denominators per stratum, clips assembled by theme, and a short list of what I would test next set against what I would simply fix. That split is the point. A defect hitting one in eight mobile sessions does not need an experiment, it needs a ticket, and putting it in the test queue delays the repair by a quarter and burns a slot you needed for a real question. Not included: No analytics engineering, no test setup, no moderated research, no survey work. I do not attend standups. If the recording tool is not installed, or is masking the fields that matter, that is a prerequisite rather than something I can work around. Not for you if: You want a rolling weekly digest of interesting sessions. That habit manufactures activity and decides nothing, and I will not sell it at any price. You need identifiable behaviour tied to named accounts; I review masked recordings only, and before starting I will ask you to confirm your tool suppresses input contents, which on default configurations it often does not. Your tool samples under 20 percent of sessions and the rate cannot be raised for the study window, in which case the strata will not fill and the counts you get back will not carry the weight you want to put on them.