Session recording is the most tempting feature in any analytics product and the easiest to leave running for a year without learning anything.
What it is good at
One flow that is demonstrably broken. Your funnel says forty per cent leave at step three. Counting has told you where; a recording tells you what they did when they got there.
A form nobody finishes. Field-level abandonment is close to impossible to infer from events and obvious in thirty seconds of replay.
Rage clicks. Repeated clicking on something that is not interactive is the clearest signal in behaviour data, and it is invisible to any report that counts page views.
Something you cannot reproduce. A bug that only happens on one browser, or only for people arriving from one campaign.
What it is bad at
Routine monitoring. Nobody watches recordings every week for long, and a habit of dipping in produces anecdotes rather than findings.
Answering "how many". One session is one person. Use the numbers for scale and the replay for explanation.
Anything you have not framed as a question. Watching without one is the most expensive way to browse your own site.
What it costs
More than the rest, on three counts.
It is the heaviest thing to store, which is why recordings expire soonest on every plan. See how long Skomi keeps your data. It is metered like everything else, so recording a large share of a busy site draws down the same allowance your analytics uses. And it collects more about a person than a page view does, which is a responsibility as much as a cost.
Turning it on deliberately
Three settings make it a considered choice rather than a default.
Sampling. You almost never need every session. A sample gives you the same findings for a fraction of the cost.
Masking. Input contents are not captured, and you can block chosen regions outright. Anything showing personal data should be on that list before you start, not after somebody notices.
Opt-out. A site can honour Global Privacy Control, which excludes visitors whose browser sends it while still counting their visits.
A working rule
Turn recordings on when you have a question, sample as little as answers it, and turn them off when the question is answered. That is a smaller feature than the one being sold to you, and it is the one that pays for itself.
If you do switch it on, two settings decide how much of a commitment it is: masking, which is applied to inputs by default, and the recording sample rate, since a busy site rarely needs every visit filmed.
Related: why your heatmap and your analytics disagree, how long to keep analytics data, and Behavior.