A default analytics dashboard shows something like thirty numbers. Perhaps five of them will ever change a decision you make. The rest are either restatements of each other or facts with no available action attached.
Being able to tell the difference is most of what separates useful measurement from a weekly ritual of looking at charts.
The test
Before treating a metric as important, answer: if this number moved twenty per cent, what would I do differently?
If the answer is "nothing", or "I would look at another number", it is not a metric you need on a dashboard. It might still be worth having available for diagnosis — that is a different job from being watched.
The ones that survive
Conversions, by source
The only metric that connects to the reason the site exists. Everything else is an intermediate.
The important part is by source. Total conversions tell you how the month went; conversions by acquisition channel tell you where to spend the next one. And conversion rate by source is more useful still, because a channel sending a tenth of the traffic at four times the rate is the one to grow.
The trap: conversions attributed to the click rather than the visit. Someone arriving from search, leaving, returning directly and converting is usually credited to "direct", which is a category meaning "we do not know". Attribution deserves care.
Entry pages, with what happened next
Not "top pages" — entry pages. Where people arrive is where first impressions are formed and where most of your acquisition surface actually lives.
Paired with what those visits went on to do, this is the most actionable table in analytics. A page bringing significant traffic that converts at nothing is either attracting the wrong audience or failing them.
Engagement time, if it is measured honestly
Genuinely useful when it means visible-and-active time, and worthless when inferred from the gap between requests, which counts an abandoned background tab as attention. Check which yours is before trusting it.
Funnel step drop-off
For any sequence you care about, the step that loses the most people is the highest-value thing analytics can tell you, because it is specific and fixable.
Only if the funnel respects order — a visit counted at step three must genuinely have passed one and two, in sequence, in that visit. A funnel that counts them independently is showing you four unrelated numbers in a shape that implies a relationship.
Where you cannot see
Meta, and important. If a fifth of your traffic is invisible to your measurement, every rate above is computed on a skewed sample. Knowing your own blind spot is a precondition for the rest of the list meaning anything.
The ones that are mostly theatre
Bounce rate
The most-quoted and least-useful number in analytics. The classic definition — a session with one pageview — makes a page that fully answered someone's question look like a failure. Someone who read your article, got what they needed and left is not a problem.
It is only meaningful when it accounts for engagement: one pageview and under some threshold of active time. Even then, it is a diagnostic rather than a goal. Nobody should have a bounce rate target.
Total pageviews
A vanity number. It goes up when you publish more and when a bot gets through. It cannot go down in a way that tells you anything. Use it for capacity planning, not for decisions.
Average session duration
An average over a distribution with a long tail and a large spike at zero. The mean describes almost nobody. If you must, use a median, and expect it still not to change any decision.
Pages per session
Rises when navigation is confusing and when content is compelling. Two opposite meanings, one number.
Real-time visitors
Genuinely enjoyable and almost never actionable. Useful for exactly two things: confirming a deploy worked, and watching a launch. Neither is a reason to have it open all day.
A dashboard worth keeping
Five things:
- Conversions and conversion rate, by source, this period against last
- Entry pages, with conversion rate beside each
- Funnel drop-off for your one most important sequence
- Traffic by channel, watching the trend rather than the level
- Your measurement gap, so you know what the four above are computed on
Everything else belongs one click away, where it can be consulted when a question arises rather than glanced at when none has.
Common questions
Is bounce rate a useful metric?
Rarely, as usually defined. A session with a single pageview counts as a bounce even when the visitor read the whole article and got exactly what they came for. It becomes somewhat useful when it also requires low engagement time — one pageview and under ten seconds active — but even then it is a diagnostic to investigate rather than a target to move.
What is the single most useful analytics metric?
Conversion rate by acquisition source. It is the only common metric that connects what you do to why the site exists, and breaking it down by source turns it from a scorecard into a decision about where to spend effort. Everything else on a typical dashboard is either an input to this or a restatement of another number.
Should I track average session duration?
It is a weak metric because it averages a distribution with a huge spike at zero and a long tail, so the mean describes almost no real visitor. If you want to know whether people engage, measure active visible time per page directly and look at the median. Even then, be honest about whether the number would ever change a decision.