For TikTok analytics and Instagram insights, I’d start with visibility, attention and the action you wanted your content to prompt, compared with your own usual results. You don’t need to track everything: you need a clear question, similar posts and a fair reading point to decide what to try next.
Which TikTok and Instagram metrics matter when you’re small?
I’d organise your analysis around these questions: is the content being seen, does it hold attention, and does it prompt the action you intended? Then I’d add a practical question: can you keep making it without every post taking over your week?
I wouldn’t give every publication the same job. A demonstration, a conversation starter and an introduction to your services deserve different ways of judging whether they helped.
For an explanation, I’d examine attention and the questions people ask afterwards. For a practical reference, I might choose saves as the main signal to investigate. For a service introduction, I’d record relevant enquiries separately rather than treating likes as a substitute.
My starting point is your own typical performance, not a number every small account is supposedly meant to reach. I want to help you answer “Is this improving on what I usually achieve?” before asking whether the total looks impressive.
I’d also separate effort from outcome. You can decide to make the demonstration clearer, deliver the promised answer and finish a manageable test; you can’t decide how many people will choose to watch.
That gives you an effort-first goal alongside your performance question. Even when the result is ambiguous, you can finish with something useful: a clearer script, a repeatable format or a better comparison.
Where do I find TikTok analytics?
TikTok’s creator tools guide describes opening Profile, then the menu, then TikTok Studio; it also describes accessing individual post analytics through More insights beneath a post (checked 2026-10-19).
I’d start with the publication you want to understand, rather than copying the account overview straight into your notes. Before recording anything, I’d write down whether the screen describes a particular post or the account as a whole.
I’d then copy the metric’s exact label and the period being shown. A total without that context is difficult to compare later, even if you remember what you meant when you wrote it.
For this method, I’d use post-level readings to investigate creative choices. I’d keep the account overview for broader questions, such as whether my overall routine feels sustainable and whether the content still reflects what I want to make.
The same guide notes that tool availability varies across app and web experiences, with some features depending on region or access criteria (checked 2026-10-19).
If a field isn’t available to you, I wouldn’t replace it with a guess. I’d label it “not available” and choose a question I can investigate using the information I actually have.
How should I start with Instagram insights?
I’d begin with the insights for the publication I want to assess and read the labels before comparing anything. My rule is simple: don’t carry a definition across from TikTok merely because the wording looks familiar.
In my record, I’d keep reels, carousels and Stories in separate groups initially. I’m not asking you to commit to a single format; I’m keeping each comparison focused on a useful question.
For a reel, that question might concern the opening or the demonstration. For a carousel, I might investigate whether a practical explanation prompted the action I intended. I wouldn’t make either format compete against an unrelated post on a different job.
If you’re still deciding what each format is for, my guide to growing on Instagram with reels and carousels can help you choose an intention. Then bring that intention back to your record before looking for a winner.
I’d also keep “not available” separate from a recorded zero. For example, ‘not available’ means I lack the measurement; a recorded zero means the available measurement recorded no actions of that kind.
That distinction matters more to me than having every box filled in. An honest gap is easier to work with than a tidy-looking sheet built on assumptions.
What’s the difference between views and reach?
In its social content measurement guidance, Meta distinguishes reach as a unique audience measure from impressions or views as the total times content was seen (checked 2026-10-19).
I’d preserve that distinction in your notes and check the definition beside the specific field you’re using. I wouldn’t rename views as “people” or assume that a rise in views means an equivalent rise in unique accounts reached.
Example
Completely hypothetical Instagram example: a reel records 900 views and reaches 600 accounts. I wouldn’t describe that as 900 different people, or assume every reached account watched the same number of times.
Another imaginary reel records 780 views and reaches 650 accounts. The first has more views; the second reaches more accounts, so the preferred result depends on the question you chose.
I wouldn’t treat reach as a grade for your ability. I’d use it as context alongside what you published, whom you meant to help and what happened afterwards.
A distribution question and a usefulness question aren’t interchangeable. If you wanted enquiries, a larger audience might be interesting, but I’d still check whether relevant enquiries arrived.
If you need help turning that intention into a content routine, start with growing on TikTok from zero. For the analysis itself, I’d keep the success criterion fixed rather than changing it after seeing which number looks strongest.
How do I read watch time and retention?
For Instagram reels, Meta defines total watch time as accumulated playback time, including replays, in Meta’s explanation of Reels watch-time metrics (checked 2026-10-19).
I wouldn’t read that total as the time a typical person spent watching. I’d keep the exact label so I don’t accidentally compare accumulated time with an average.
The same explanation describes average watch time as the average time spent playing a reel, calculated from watch time and plays (checked 2026-10-19).
Compare videos of similar length
I’d begin with videos of broadly similar length and purpose. That makes it easier to investigate whether the explanation holds attention without choosing a winner solely because it accumulated more seconds.
Example
Hypothetical example: a 20-second video has an average watch time of 12 seconds. A 50-second video averages 18 seconds.
The longer video records more average seconds, but that alone doesn’t establish that it holds attention better throughout its full explanation. I’d keep both the duration and the watch-time reading in the comparison.
Use a retention curve to form a question
If a retention curve is available, I’d read it while watching the video. I’d note what you’re saying or showing around a drop, without treating the timing as proof of its cause.
My hypothesis might be “The introduction delays the useful demonstration.” I’d test a more direct opening while keeping the topic, structure and length as similar as practical.
The guide to planning TikTok videos and reels from idea to script can help turn that question into a draft. I’d take a specific observation into the script, not the vague instruction to make everything more entertaining.
What do saves, shares and comments tell me?
I’d treat interactions as clues, not complete explanations of what someone thought. I wouldn’t assume a save proves buying intent or that a comment proves satisfaction.
Instead, I’d assign the signal a role before publishing. For a reference guide, I might investigate saves; for a discussion, I’d examine the questions raised; for something designed to be passed along, I’d watch shares.
Those are editorial choices for the test, not universal rankings of valuable interactions. I’d still read the content and context before deciding what to repeat.
Read the questions, not just the comment total
I’d reserve space for recurring questions, rewritten briefly in your own words. Separate confusion about the explanation from requests for another example or questions about working with you.
A useful question can suggest the next publication even when the overall comment count isn’t remarkable. I’d ask whether the conversation helps you understand the people you want to reach.
I’d also avoid turning every response into a new content direction. Keep the questions that fit your purpose, and let unrelated comments remain unrelated.
Keep the counts beside any ratio
If you calculate actions divided by views or reach, I’d write down the calculation. Keep the denominator consistent and explain which actions you included rather than giving different sums the same “engagement” label.
Example
Hypothetical example: a post receives 8 saves from 400 accounts reached, while another receives 12 saves from 1,200 accounts reached. Saves divided by reach are 0.02 and 0.01 respectively.
The second post has more saves in total; the first has more saves relative to its recorded reach. I’d retain both counts and avoid turning this isolated comparison into a rule for the whole account.
I’d describe this as saves per account reached, not the chance that an individual viewer will save. That wording keeps my interpretation close to the calculation rather than claiming more than it tells me.
Should I use the average or the median?
Example
Illustrative example of defining your usual result: I’d start with the median by sorting the values and taking the middle one. If the centre falls between a pair of values, I’d use their midpoint.
I’d keep the average available, but I wouldn’t let it define “normal” on its own. I want to see whether an exceptional publication is pulling the summary away from most of the group.
Example
Entirely hypothetical results from five comparable organic videos, all recorded at the same age: 300, 360, 420, 480 and 3,000 views. The median is 420; the arithmetic average is 912.
Using the average as the expectation would leave most of these imaginary videos below “usual”. I’d investigate the exceptional result without making it the obligation for every future publication.
The median only describes the group you selected. It can’t repair a comparison that mixes unrelated formats, different reading ages or paid promotion with organic distribution.
I’d choose a recent, relevant comparison group and record unusual circumstances. I wouldn’t quietly remove an inconvenient result or include an unrelated success to make the baseline look better.
How I'd measure it
The routine below is an illustrative working method, not a platform requirement.
I’d use this as a learning routine, not a way to predict the next publication. Before checking results, I’d write down the question, the comparison group and the point at which I’ll review them.
Choose a question and a main metric
I’d begin with a question that could lead to a decision. “Should I show the finished result before explaining the process?” is more manageable than “How do I make every metric go up?”
Then I’d choose a main metric and a supporting observation. For an opening test, I might examine the available attention measure while checking whether the video still delivers its promised answer.
I wouldn’t reward an opening that attracts attention by promising something the rest of the video doesn’t provide. My test needs to improve the content, not merely improve a detached number.
Build your own comparison group
I’d compare similar organic publications within the same platform. I’d note collaborations, paid promotion and other distinct circumstances separately rather than blending them into the main baseline.
Then I’d calculate the median for the chosen metric. I’d keep the individual results visible so I can see whether a group looks consistent or depends on an exceptional piece.
If I discover that the selection was unfair, I’d explain the change and recalculate using the revised rule throughout. I wouldn’t adjust the group only for the variant I wanted to win.
Read posts at the same age
I’d agree a review point in advance instead of comparing a newly published video with an older one. This is a working convention for the experiment, not a claim about when either platform finishes distributing content.
Example
Hypothetical review routine: record each TikTok video or Instagram reel at 72 hours, then take another reading at 7 days. Compare the early readings with each other and the later readings with each other.
The 72-hour reading is an organisational choice in this imaginary test, not an official deadline or a declaration that the content has stopped reaching viewers. If the later reading changes the conclusion, I’d record that too.
Change one variable at a time
If I’m testing the first frame, I’d aim to keep the topic, promise, structure and length similar. If I’m testing length, I’d keep the core explanation intact and note what I removed.
I don’t expect laboratory conditions. I want to reduce avoidable differences and record the rest so I don’t attribute every change to my favourite explanation.
Example
Hypothetical experiment: make 4 videos opening with a question and 4 opening with the finished result. Alternate the variants and compare their median attention readings at the same age, without deliberately changing the music or editing style.
That batch size is a practical choice for this imaginary experiment, not a platform requirement or proof of statistical reliability. I’d still inspect the individual results and production notes.
Decide whether to repeat, adjust or wait
I’d repeat a variant when the improvement appears across comparable pieces and the effort is manageable. I’d adjust when a specific observation suggests another test, while keeping the explanation labelled as a hypothesis.
I’d say “not clear yet” when the groups overlap, important information is missing or the apparent advantage depends on an exceptional post. That is a complete analytical answer, not an unfinished task.
I also wouldn’t keep extending an unclear test indefinitely. I’d simplify the question, improve the comparison or choose a different variable that is easier to investigate.
How do I track analytics without constantly checking them?
I’d start with a simple note or spreadsheet you can update from your phone. Before building a dashboard, I’d check whether the record helps you make an actual decision.
- Identify the post, platform, format and publication date.
- Write its intended job and the variable being tested.
- Copy the main metric’s exact label, value and reading age.
- Keep a supporting observation and note unusual circumstances.
- Record production effort and finish with a decision.
For effort, I’d use plain categories such as light, manageable or difficult to sustain. You don’t need to time every task if that creates more work than it resolves.
I’d finish each review with a short sentence: “Keep this structure”, “Test a more direct opening” or “Need more comparable posts”. If the record never leads to a decision, I’d remove fields until the useful question becomes visible.
I’d also note what I’m deliberately leaving unchanged. Keeping the middle of the explanation stable while testing the opening prevents every review from becoming a complete creative rebuild.
Which analytics mistakes would I avoid?
I’d avoid choosing the goal after the results arrive. If the post was intended to generate relevant enquiries, I wouldn’t declare that job done merely because another metric looked encouraging.
I also wouldn’t attribute account-wide follower growth to a particular video without information supporting that connection. I’d write “This happened during the same period” rather than “This video caused it”.
After a disappointing result, I’d resist changing the opening, length, music and topic together. I’d choose a smaller question so the next reading has a better chance of teaching me something.
Finally, I wouldn’t turn a low number into an automatic explanation about the algorithm or your audience. I’d first check the comparison, the available information and whether the publication delivered what it promised.
Frequently asked questions
How many views are good for a small account?
I wouldn’t set a universal target. I’d compare with your median for similar posts and check whether the content prompted its intended action at an effort level you can repeat.
Should I check my analytics every day?
I’d choose review points rather than reopen the numbers whenever I feel uncertain. For this method, comparable reading ages and a useful decision matter more than frequent checking.
What should I do when a metric is missing?
For example, I’d record an unavailable metric as ‘not available’, rather than entering 0. Then I’d investigate a question supported by the information you have rather than filling the gap with an estimate.
Can I compare TikTok and Instagram directly?
I’d compare each platform with itself first. Afterwards, I’d compare creative lessons and production effort without assuming similar labels describe identical measurements.
What does “not clear yet” mean?
It means I wouldn’t confidently choose a variant from this comparison. I’d keep what I learned and improve the next test rather than turn an ambiguous difference into a rule.
What should I do next with my analytics?
Before your next publication, create a note and complete this sentence: “I’m testing ___, comparing it with ___, and deciding by looking at ___.” If that still feels too broad, make the question smaller before you publish.