Blog · Explainers

Reel analytics: the four numbers worth reading

Read retention at three seconds, average watch time as a percentage of length, shares, and follows from the video. Each one maps to a different fault: retention to the hook, watch time to the pacing, shares to whether the video was worth passing on, and follows to whether the account is legible. Views are the output of the other four and diagnose nothing.

Most people check views, feel something about the number, and change nothing specific. The useful version is narrower: four numbers, each of which points at one part of the video, so that the next edit is a decision rather than a mood.

Why views are the wrong number

Views are downstream of everything else. A reel gets shown to a small sample, and if that sample watches, it is shown to more people. So the view count is the system's summary of the other four numbers — by the time you read it, it has already happened, and it tells you nothing about which part went wrong.

NumberWhat it measuresWhat to fix if it is low
Retention at 3sWhether the opening earned the watchThe first frame, first cut, first line
Average watch timeWhether the middle heldPacing, length, cut density
SharesWhether it was worth passing onThe idea, not the edit
Follows from this videoWhether the account is legibleConsistency across your last nine posts
ViewsAll of the above, combinedNothing directly

Retention at three seconds

The single most diagnostic number available. It is the percentage of people still watching after the opening, and it is almost entirely a function of decisions you made in the first second.

  • Below about 50%: a hook problem, and nothing further down the video is being seen. Fix the first frame, the first cut and the first line before changing anything else.
  • A cliff in the first second: the opening frame did not read at all. Usually a wide shot, a fade, or text nobody could parse at speed.
  • A gentle slope from the start: the hook worked and the promise was vague. People stayed to find out and then found out.
  • Above 70%: the opening is doing its job. Move your attention to the middle.

Everything that moves this number is in reel hooks: what the first three seconds have to do, and the cheapest single fix is a cut inside the first second — a video that has not cut yet reads as an upload rather than an edit.

Average watch time, as a percentage

Read it relative to length, not in seconds. Twelve seconds on a fifteen-second reel is excellent; twelve seconds on a sixty-second reel is a video that lost three quarters of its audience.

LengthWeakGoodStrong
7–15sunder 50%60–80%over 90%
15–30sunder 40%50–65%over 75%
30–60sunder 30%40–55%over 60%

The most common cause of a bad percentage is not a bad video — it is a video that was longer than its idea. Cutting a thirty-second reel to eighteen usually raises both the percentage and the absolute watch time, which is the rare change that improves two numbers at once. How long should a reel be covers where the length actually comes from, which on a beat-matched edit is the track.

Look at the shape of the graph, not the average. A drop at a specific second is a specific moment: a shot that held too long, a caption nobody could read in time, a section where the music dropped out. Watch that second and you will usually see it immediately.

Shares, and the ratio worth watching

Shares are the strongest signal a short video can generate, because sharing costs the viewer something — they are putting their own name behind it. A reel with high shares relative to views is being passed between people, which is the only distribution that compounds.

  • Shares per thousand views is the number to track over time. The absolute count is meaningless without the denominator.
  • Shares come from the idea, not the edit. A beautifully cut video of an obvious point does not get shared. A plainly cut video of a useful one does.
  • Saves are the sibling metric and mean something different: saves say "I will need this", shares say "you should see this". Educational content wins saves; opinions and surprises win shares.
  • Low shares with high retention is the signature of a well-made video about nothing much. That is a content problem and no amount of editing fixes it.

Follows from the video

The number closest to whether any of this is building anything. A reel with a hundred thousand views and four follows reached people who enjoyed one video and saw no reason to expect another.

That is usually not a fault in the reel. It is a fault in what a visitor finds when they arrive: nine posts that look like nine different accounts. The fix is consistency — one grade, one text system, one pace — which is the argument running through why reels look amateur and the practical reason for reusable text overlays.

Reading them together

PatternDiagnosisNext change
Low 3s retention, everything else unknownHookFirst frame and first cut
Good retention, low watch time %Too longCut 30% of the length
High watch time, low sharesWell made, not worth passing onChange the idea
High shares, low followsOne-hit videoMake the account legible
Everything fine, few viewsNot enough volumePost more often
All numbers down at onceNothing — this is varianceWait five more posts

How to actually use this

  1. Review weekly, not daily. Daily numbers are noise, and reels keep accumulating views for days.
  2. Judge on five videos, never one. Single-post variance on short video is enormous, and reacting to one post is how people rebuild a format that was working.
  3. Change one thing at a time. New hook, same everything else. Two changes teach you nothing about either.
  4. Write the number down before you change anything, so you have something to compare against in a fortnight.
  5. Keep the winners as templates. A structure that beat the others twice is not luck, it is a format — and it should become the recipe you generate from rather than something you rediscover.

The reason this discipline pays is that most accounts change everything every week and therefore learn nothing in a year. Holding the look, the pace and the length constant while varying one input is slower for a fortnight and the only thing that produces a format you can rely on — which is what makes batch production worth setting up in the first place.

Read next