What a dimension describes
A dimension is the descriptive attribute a number is grouped by. Country, device category, landing page, session source, campaign name, browser and page title are all dimensions. When a report says “sessions by device”, sessions is the metric and device is the dimension doing the describing.
Dimensions arrive from three places. Some are collected automatically by the analytics tool from the request itself, such as device and country. Some come from the link that brought the visitor, which is what UTM parameters exist to supply. And some you define yourself and send with an event: membership tier, form type, logged-in status, product category.
Dimensions also have scope. Some describe a single event, some the whole session, some the user across sessions. Mixing scopes in one report produces numbers that look wrong because they are answering a slightly different question from the one you thought you asked.
Why dimensions matter
Totals hide almost everything worth knowing. A conversion rate for the whole site is a single figure describing nobody; the same rate split by device usually reveals a healthy desktop experience and a mobile one quietly losing people. In Nepal, where most traffic arrives on a phone, that split is often the entire story of the account.
They are also how you find waste. Cost by campaign is routine reporting. Cost by search term, by device, by hour and by location is where money that is not working becomes visible, and where the first sensible cuts come from.
Common mistakes with dimensions
Untagged links are the biggest. Any visit arriving without campaign parameters lands in a general bucket, and once it is there no report can recover where it truly came from.
The second is creating a custom dimension with far too many distinct values — a unique identifier per visitor, for instance. That pushes a report into high cardinality, where the tool begins grouping rare values into a catch-all row and the detail you built the dimension for disappears.
The third is naming values inconsistently. “Facebook”, “facebook” and “FB” become three separate rows that nobody thinks to add together. Agree the spelling once, in writing, before anyone builds a campaign.
How to use them
Choose the dimensions you need from your measurement plan rather than adding them opportunistically, and give each one a written definition and an owner. Keep the number of distinct values low and stable over time; a handful of clean categories beats a long list nobody can read.
Then use them in pairs. One dimension shows that a difference exists; two show where the difference lives. Device and landing page together, or channel and country, will explain most surprises without any further tooling. Where the same value has to appear in more than one tool, set it once in your Google Tag Manager setup so it stays consistent everywhere it is read.