Most social media reporting is a slideshow of numbers that nobody acts on. Impressions up, followers up slightly, engagement rate down a tenth of a percent. Everyone nods. Nothing changes, because nothing in there suggested a change.
The fix is not more metrics. It is fewer, chosen by a single test: would a change in this number change what we do next? If a metric goes up and your plan stays the same, and it goes down and your plan still stays the same, that metric is scenery.
This guide covers which numbers pass that test, how to read the awkward ones properly, and the one piece of analysis that is worth more than any chart.
Pick your goal before your metrics
The same number can be essential or pointless depending on what the channel is for. So start from the goal, of which most small businesses have one or two, not five.
| Goal | What to measure | What people measure instead |
|---|---|---|
| Get found by new people | Reach, and the share of it from people who do not follow you | Follower count |
| Build a relationship with an audience | Comments and replies, saves and shares | Likes |
| Drive visits to a website | Sessions from social at your own site, by campaign | Impressions |
| Drive bookings or orders | Completed bookings tagged to the campaign | Clicks |
| Answer customers quickly | Median first reply time and the oldest open item | Number of messages received |
| Be visible locally | Calls, direction requests and website taps on your listing | Post impressions, which Google no longer reports anyway |
The last row is worth pausing on. Per-post metrics for a Google Business Profile do not exist for anyone: the measurements were retired in 2023 with no replacement. What is still reported is how people found the listing and what they did next, which is the more useful half. Our guide to what each network reports back covers the other gaps like it.
The five numbers to track per channel
Followers. Not because the number matters, but because its direction does, and because it is the denominator for several other things. Watch the rate rather than the total: our follower growth rate entry has the arithmetic.
Impressions or views. How many times your posts were shown. Useful as a sanity check and as the denominator of an engagement rate. Not useful as a goal, because it rises with volume.
Engagements. The interactions: likes, comments, shares, saves, clicks, depending on the network. Count them, but care more about the ones that cost something. A save or a share is a stronger signal than a like, and a comment is stronger than both.
Engagement rate. The most useful single number on this list, and the easiest to report misleadingly. See below.
Clicks, measured at your own site. Not the network’s click count, which is defined differently everywhere and sometimes not reported at all. Put tracking parameters on your links and count arrivals. Our free UTM builder builds them.
That is the list. If you want a sixth, make it replies answered within target, because customer service is the part of social media most likely to produce revenue and least likely to appear on a dashboard. Our guide to social media response time covers how to measure it honestly.
Engagement rate: state the denominator
Engagement rate is engagements divided by something, and the something changes the meaning entirely.
Engagements per impression answers: of the people who saw this, how many did something? This is the right measure for judging a post. It is independent of your audience size, so it lets you compare this month’s post with last year’s.
Engagements per follower answers: how active is my audience? This is the right measure for judging an account over time, and it falls naturally as an account grows, because large audiences are less uniformly interested.
Those are different questions and the two numbers are not comparable. A post can have an excellent rate per impression and a poor one per follower at the same time, and both facts are true.
Three rules keep it honest:
- Say which denominator you used, every time, in the report itself.
- Use the same one all year. Switching mid-year makes your own history useless.
- Never compare your rate with someone else’s unless you know their denominator, which you usually do not.
Our engagement rate entry has the formulas, and the free engagement rate calculator does the arithmetic both ways.
Compare like with like, or compare nothing
This is where most analysis goes wrong, and it goes wrong invisibly because the chart still looks fine.
If you compare a vertical video posted on a Thursday evening with a link post from a Tuesday morning, you have learned nothing about either the format or the time, because two things changed. If you compare this month’s engagement totals with last month’s and you posted twice as often, you have measured your volume.
The discipline:
- Same network. Always. Numbers from two networks are different quantities with the same names.
- Same format. Carousels with carousels, video with video.
- Similar kind of post. Teaching with teaching, promotion with promotion.
- Same metric, same denominator.
- Set aside the outliers. A post shared by a large account, or one published on a public holiday, will distort any group it lands in. Note it and remove it.
And respect sample size. Three posts in a category is a hint. Ten is an argument. The most useful habit here is writing the count next to the claim: “carousels outperformed single images, across nine carousels and fourteen images” is a sentence you can judge. “Carousels perform better” is not.
The analysis worth more than any chart
Here is the question most dashboards do not answer, and it is the one that changes behavior: what do my best posts have in common?
Not when they went out, which is a separate and well-trodden question. What they are. You can answer it with a spreadsheet and an afternoon:
Take your last few months of posts on one network. For each one, record the format, roughly how long the caption was, how many hashtags it had, whether it contained a link, and whether it asked a question. Then split the posts by each of those dimensions and compare the engagement rate of each group against your own average.
What comes out is usually surprising and immediately actionable. Posts with a question in them do better, or worse. Short captions win on one network and lose on another. Links cost you reach, or do not. Three hashtags beat ten.
Two rules stop this turning into superstition:
- Normalize against your own average, per channel, so one large channel does not decide what the brand should do.
- Ignore any group with too few posts in it. A dimension needs at least two groups with a reasonable number of posts each before it can say anything at all.
Picifer does this analysis from your own published posts and shows the sample size next to every finding, for exactly that reason: a conclusion drawn from four posts should look like one drawn from four posts. The free social media audit tool walks through a simpler version by hand.
What to ignore, politely
None of these are worthless. They are just not decisions.
Follower count as a goal. It rises with time and with giveaways, and neither correlates well with revenue. Watch the rate, not the total.
Impressions as a goal. Post more and impressions rise. That is arithmetic, not achievement.
Likes in isolation. The cheapest interaction there is. Fine as context, weak as evidence.
Industry benchmarks. Collected from audiences that are not yours, often with an undisclosed denominator, usually by someone selling something. Your own trend is better data.
Best-time charts from other people’s audiences. Useful for the first week, replaceable by your own data after a month. Our guide to finding the best time to post covers the method.
Anything you cannot act on. The cleanest test there is. If a metric went up 20 percent and nothing about next month would change, take it off the report.
Three kinds of evidence that are not metrics
Some of the most useful information about your social media never reaches a dashboard, and it is worth collecting deliberately rather than noticing by accident.
What customers say when you ask them. “How did you hear about us”, at the counter, on a booking form, in the order confirmation. It is unfashionable, the sample is messy, and it is the only measurement that sees the person who watched a video, told nobody, and walked in three weeks later. Every API ever written is blind to that person. Our dark social entry covers why.
The questions your inbox repeats. If forty people a month ask whether you deliver, that is not a customer service statistic, it is a note that your posts and your profile are not answering an obvious question. The inbox is a list of gaps in your own communication, free of charge, continuously updated.
What your audience does with a post rather than to it. Someone who screenshots your price list, sends it to a friend and books the next day leaves no trace in any engagement count. The nearest visible signals are saves and shares, which is why they are worth more attention than likes, and why the posts worth repeating are often not the ones with the best numbers.
None of these three belongs in a report with a decimal place. All three belong in the monthly conversation, because they routinely change what to post next in a way that a 4 percent movement in impressions never does.
The monthly review
Twenty minutes, once a month, with the same four questions. More often and you react to noise.
Did we keep the plan?
Count the posts that went out against the slots you have. If you filled half your slots, that is the finding, and the schedule is the thing to change.
What were the five best posts, and what do they share?
By engagement rate, on comparable posts. Look for the shared trait, not the winner.
What happened to clicks and bookings?
Measured at your own site, by campaign. This is the number that connects to money.
What is one change for next month?
One, or at most two. If you change five things and results improve, you have learned nothing about which one did it.
Write the decision down, with the date. Over a year those notes become the only genuinely reliable social media guide for your business, because every line in it was tested on your own audience.



