The problem with social analytics is not scarcity, it is triage. Ten metrics are displayed, three matter, and the three that matter are rarely the ones on top. Worse, the noisy metrics move more dramatically, which makes them feel more informative than they are.
The metrics that predict growth
| Metric | Why it matters | What it tells you |
|---|---|---|
| Watch time / retention | The dominant distribution input on video platforms | Whether the content holds attention — the only thing you fully control |
| Saves | Signals lasting value, not a reflex | Content worth returning to; strongly weighted on Instagram and Pinterest |
| Shares / sends | Highest-intent action available | Someone put their own credibility behind it |
| Non-follower reach | Isolates discovery | Whether you are reaching anyone new at all |
| Follows per post | Conversion of reach into audience | Whether you are reaching the right new people |
If you tracked only these five, you would lose very little. Retention in particular is worth more attention than it usually gets, because it is both the strongest distribution signal and the one most directly improvable by editing.
The metrics that mislead
- Follower count. A lagging indicator that responds to work done weeks earlier. Useful over quarters, useless over days, and psychologically compelling in inverse proportion to its usefulness.
- Likes. The cheapest possible action, weighted lightly by most platforms. High likes with low saves usually means content that was pleasant and forgettable.
- Impressions. Counts eyeballs without indicating quality of attention. Ten thousand impressions with two-second average watch time is worse than a thousand with fifteen.
- Total views. Same problem — a view often counts after a very short threshold.
- Engagement rate, taken alone. A useful ratio that is easily distorted. Small accounts show inflated rates because a handful of loyal followers is a large percentage; the number drops as you grow even when performance improves.
Reading data without fooling yourself
- Look at medians across at least ten posts, never a single post. Medians rather than averages, because one breakout distorts an average completely.
- Compare like with like. Carousels against carousels, videos against videos. Format differences swamp everything else.
- Watch trends over four to eight weeks, not week to week. Weekly comparisons are mostly noise.
- Segment by content type. The genuinely useful question is which kind of post works, not which post worked.
- Note the outliers separately. A breakout post is worth studying on its own terms, but it should not be included when calculating what is normal.
The question worth asking of your data
Rather than "how did this post do," the more useful questions are structural:
- Which content type has the highest median retention? That is your format.
- Which topics generate saves rather than just likes? That is your subject.
- Which posts brought follows rather than just reach? That is your audience fit.
- Where does retention drop in your videos? That is your editing brief.
- Which posts got comments without you asking for them? That is your genuine interest signal.
Each of those produces an action. "Reach was down 8% this week" produces anxiety and no action, which is a reasonable test of whether a metric is worth your attention.
The retention graph, specifically
On video platforms, the retention curve is the single most actionable chart available and the least examined. The shape tells you specific things: a steep initial drop means the hook is not landing; a mid-video cliff means you lost people at a identifiable moment you can go and watch; a flat curve with a late drop is close to ideal.
Unlike most metrics, this one points at a specific edit. Find the drop, watch that second of the video, and fix what happens there.
A minimal tracking habit
One row per post in a spreadsheet: date, format, topic, retention or watch time, saves, shares, follows. Six columns, thirty seconds per post. After thirty posts you can answer questions about your own audience that no general advice can address, and you will have stopped reacting to single-post noise — which is most of the benefit.