Twitter analytics means two things, and only one scales
Twitter analytics is either the tab on your own account or measuring a conversation you do not own. They answer different questions and need different data.

Twitter analytics means two different things, and the confusion costs people a lot of time. One is the tab on your own account. The other is measuring a conversation you do not own.
One is the tab on your own account. The other is measuring a conversation you do not own. They answer different questions and only one of them scales.
What the built-in analytics tab covers
Your own posts, their impressions, their engagement and some audience information. It is free, it is accurate and it stops at the edge of your account. That is enough to answer whether your posting is working. It cannot answer anything about a hashtag, a campaign involving other people, a competitor or an event.
If you need the second kind, that is what we do, within the Twitter compliance rules.
The metrics that exist, and which ones matter
A full report on X data breaks into a handful of groups: volume, ratios, concentration, rankings of posts and accounts, composition and movement over time. Volume is the easiest and the least useful on its own. The ratios are where a report starts saying something, because they put two raw numbers against each other.
Concentration is the one most reports skip and the one that changes decisions. It tells you whether a handful of posts carried everything or the load was spread.
Why every figure should be countable
Anything derived from counting, summing, dividing or sorting the underlying posts can be checked. Anything produced by a model cannot. Impressions are counted by the platform. Engagement is counted. Ratios are arithmetic on counted things.
Estimated reach, automatic topics and sentiment percentages are model output. They can be useful, but they belong in a different category and should be labelled as such in any report that will be questioned.
The measurement people get wrong most
Dividing engagement by followers instead of impressions. It produces a bigger, friendlier number that answers a question nobody asked: how your theoretical audience behaved. Dividing by impressions answers how the people who actually saw the post behaved, which is the one that matters now that timelines are algorithmic.
What changes when you measure someone else's conversation
Scale, mostly. Your own account produces a few hundred posts a year. A hashtag can produce that in an hour. That changes the job from reading posts to analysing a dataset, and it changes what you need: every matching post with its own figures, not a sample and not an estimate.
It also brings the terms of the API into play, which govern what can be collected and stored. Everything we deliver is gathered Twitter compliant, which is what keeps it usable for client reporting and for research.
Where to start
Pick one question. Whether the campaign worked, who is really amplifying you, how you compare with a competitor. Then take the three or four metrics that answer it and ignore the rest. A report built around one question beats a dashboard showing forty.
Choose a plan and you get the report and the dataset behind it, so every number in it can be rebuilt.