Anyone can pull up a dashboard. The skill that actually improves a social programme is interpretation: knowing which number answers the question you asked, which number only answers a similar-looking question, and which number is quietly absent because the platform never reported it. Most bad decisions in social media are not caused by missing data. They are caused by a number being read as proof of something it never measured.

This guide is the interpretation layer. It assumes you already know which metrics deserve a place in the report — if you do not, start with the metrics that actually matter — and focuses on what happens after the numbers land on your screen.

Start with the question, not the dashboard

Every number becomes useful only when it is attached to a decision. Write the question first, then pick the metric that answers it:

• "Is the content reaching people beyond our followers?" Non-follower reach, if the platform reports reach at all. Not impressions, not likes.
• "Is the content worth stopping for?" Engagement divided by reach, compared against your own recent posts rather than a published benchmark.
• "Did anyone intend to do something?" Saves, shares and link clicks, in that order of seriousness for most accounts.
• "Did the schedule hold?" Posts published versus posts planned — a publishing metric, not an audience metric.
• "Is the account growing into the audience we want?" Follower trend over weeks, read together with which posts brought the followers.

A number answering none of your questions is decoration. It can stay on the screen — dashboards are allowed to be pleasant — but it must not enter the argument.

What each number can and cannot prove

The eight metrics a social dashboard typically carries each have a narrow honest range.

• Impressions counts deliveries, including repeats. It proves the platform showed your post. It does not prove a human registered it, and comparing impressions across platforms is close to meaningless because each network counts differently.
• Reach counts distinct accounts that saw the post, where the platform reports it. This is the fairest denominator for engagement rate and the first number to check when distribution changes.
• Likes prove a low-cost positive reaction. They correlate weakly with anything downstream, and they are the easiest number to inflate by accident with a change in posting time or audience mix.
• Comments prove someone had enough energy to type. Read the comments themselves; the count only tells you whether to read them.
• Shares are the strongest social proof available: someone put your post in front of their own audience. Treat a share as worth several likes and never average them together.
• Saves mean the content was judged useful enough to return to. On the platforms that report them, saves are the best proxy for durable value.
• Clicks are the only metric in the set that points off-platform. They are also the one most often missing — several networks do not expose them to third-party tools at all.
• Video views are platform-defined: a play counts after a threshold each network chooses differently. Compare a video to your own previous videos, never to another platform's.

The pattern worth internalising: the further down this list, the closer the number sits to intent, and the scarcer it is.

The honesty layer: absent, zero and stale

Three states look identical on a screen and mean entirely different things.

Absent is not zero. A metric the platform cannot provide comes back as a null — a placeholder, not a measurement. Telegram and Google Business Profile have no post-level insights endpoint at all, so every metric for those networks is absent by design. X reports impressions only on accounts with the paid metrics tier. LinkedIn engagement arrives, but impressions generally require Marketing Developer Platform access. YouTube reports views and little else. Tumblr's note total mixes likes, reblogs and replies into one figure. The full support matrix is in the analytics documentation, which also shows which dashboard tiles are still placeholders until the sync fills them.

Zero is a real measurement — a post with no likes has zero likes. The difference matters because averaging a null into a total as if it were zero silently drags every average down, which is exactly how a healthy month starts looking bad.

Stale is neither. Metrics are swept periodically rather than streamed, and only posts published within a recent window are actively refreshed. Before you read a number as this week's result, check when it was last captured: a figure that has not moved in two days is a stored value, not a live one.

Three habits cover most of this. State the date range on every chart you quote. Never present a total without saying how many posts in the set actually reported. And when a platform is missing a metric entirely, write "not reported" rather than leaving a blank that a reader will fill in with optimism.

A fifteen-minute weekly reading pass

Reading analytics weekly is a discipline, not an investigation. Three passes, in order.

Pass one — check the pipeline, not the performance. Are the accounts still connected? Did anything fail? How fresh is the last capture? Two minutes. If the data layer is wrong, everything after it is fiction.

Pass two — compare like with like. Take this week against the same kind of week before: same platforms, similar posting volume, same post jobs. Compare medians rather than averages, because one unusual post moves an average and barely dents a median. If you published nine posts instead of five, do not celebrate a total-engagement rise without checking engagement per post.

Pass three — extract three sentences and one decision. Something rose. Something fell. Something is missing data. Then one change for next week: a format to try more of, a topic to drop, a posting window to move. If you cannot name the change, you have read the numbers but not interpreted them.

The decision is the entire point. A reading pass that ends in "interesting" will be abandoned by week four, because it costs time and returns nothing.

Vanity metrics: keep them on a leash

Likes, follower totals and raw impressions are not banned — they are benchwarmers. They earn a place in the argument in exactly three situations:

• Trend context. Follower growth rate over a quarter says something that a single week cannot.
• Someone senior asks. Answer the question, then annotate what it does not prove. Refusing a number creates suspicion; supplying it silently creates a bad decision.
• Diagnosing a collapse. When every number falls at once, the vanity metrics are the smoke detectors — useful for noticing the fire, useless for cooking.

Outside those, a vanity metric in a report is a number someone will quote back at you later as evidence for a decision you never made.

Comparing fairly, or not at all

Cross-platform comparison is where most reading goes wrong. Three rules keep it honest.

Use the same denominator everywhere. Engagement divided by reach is the fairest general-purpose ratio because reach already accounts for how many people had the chance to react. Engagement divided by followers flatters big, quiet accounts and punishes small, active ones.

Compare each platform to itself. A 3 percent engagement rate means something on one network and nothing on another. The useful comparison is this post against your own last twenty posts on that network — your own history is the only benchmark that controls for your audience.

Give a number time to be a trend. One week is weather. Four weeks is a direction. When you cannot wait four weeks, say in the report that the finding is provisional rather than presenting a single week as a conclusion.

Where PostHub fits, and where it does not

PostHub carries the numbers your own publishing produced — impressions, reach, likes, comments, shares, saves, clicks and video views, captured per post and per network, org-scoped to your workspace, with the date filters and the ranked top-posts list you need to run the three passes above. It is also explicit about what it does not have: a metric nobody reported arrives as null rather than a flattering zero, and networks without an insights API are marked unsupported instead of quietly omitted.

What it will not do is interpret for you. No tool can tell you that this week's dip follows a two-post week rather than an algorithm change, that your median is fine because one tutorial post flopped, or that the campaign label you forgot to apply is why the report cannot separate paid from organic. The reading pass — check the pipeline, compare like with like, extract one decision — stays yours.

Key takeaways

• Write the question before reading the number; a metric that answers none of your questions is decoration.
• Impressions prove delivery, reach proves opportunity, shares and saves prove intent — each number has a narrow honest range.
• Absent, zero and stale are three different states. Never average a null as a zero, and always state the capture date.
• Run a fifteen-minute weekly pass: pipeline, like-for-like comparison, three sentences and one decision.
• Compare engagement by reach, compare every platform only to its own history, and treat one week as a snapshot rather than a trend.

FAQ

How often should I actually look at the numbers?

Weekly for decisions, monthly for the stakeholder summary, quarterly for strategy. Daily checking produces reactions to noise — a single post having an odd morning says nothing that a week's median will not confirm.

What do I do when a platform does not report a metric I need?

Say so in the report and go to the platform's own analytics for that one figure. It is a slower path, but a blank labelled honestly is more useful than a substitute number that invites the wrong conclusion.

Does a dip in engagement mean the content stopped working?

Not on its own. Check posting volume first, audience mix second, and outliers third — then decide whether reach fell or engagement-by-reach fell. Those two branches lead to completely different fixes, and the metrics guide walks the full diagnosis in order.