Analytics platforms report several hundred metrics. Most businesses look at three, act on none, and would be better served by five specific numbers and a habit of checking them.
Key takeaways
- A metric is only worth tracking if a plausible reading of it would change what you do next week.
- Sessions and pageviews are volume, not performance. They rise and fall for reasons you did not cause.
- Segment by acquisition channel before drawing any conclusion, site-wide averages hide the entire story.
- Bounce rate and time on page are the two most confidently misread numbers in the industry.
On this page
The test a metric has to pass
Before tracking anything, write down what you would do if the number doubled and what you would do if it halved. If both answers are “nothing”, it is not a metric, it is decoration.
This one filter removes most of a typical dashboard. It also exposes the uncomfortable case where a number would change your decision but you have no way to influence it: interesting, but not actionable.
The numbers that pass this test are your candidates for KPIs. To turn them into goals with a target and a deadline, see our marketing KPI examples by goal.
The five that earn their place
Almost every business can run on conversion rate, cost per acquisition, channel mix, landing page performance and returning visitor share. Everything else is diagnostic detail you reach for once one of these moves.
| Metric | Tells you | Decision it drives |
|---|---|---|
| Conversion rate by channel | Which traffic is worth having | Where to spend more or stop |
| Cost per acquisition | What a customer costs to win | Whether the channel is viable |
| Channel mix over time | How dependent you are | Where the concentration risk is |
| Landing page performance | Which entry points work | What to fix or replicate |
| Returning visitor share | Whether anyone comes back | Retention and content investment |
Click to download this table as an image
Note that four of the five are ratios or comparisons. Absolute counts are almost never useful on their own, because they conflate how much traffic arrived with how well the site handled it. If landing page performance shows where people drop off but not why, a UX and conversion review is the next step, and our paid ads guide covers how to judge cost per acquisition against customer value.
The metrics that mislead
Bounce rate and time on page are not quality measures, and treating them as such produces confidently wrong decisions. Both are artefacts of how measurement works rather than descriptions of user satisfaction.
Bounce rate
In GA4, bounce rate is the share of sessions that were not engaged: under 10 seconds, one page and no key event. So someone who reads your opening hours for fifteen seconds and leaves no longer counts as a bounce. A quick bounce on a page that answers the question completely can still be a success, and the same number on a category page is a failure. The metric cannot tell the two apart; only you can, by knowing what the page is for.
Time on page
GA4 measures engagement time while the tab is in the foreground and sends it when the visitor leaves, so the old Universal Analytics problem of six-minute reads recorded as zero seconds has mostly gone. The interpretation problem has not: a long time can mean real interest, or a visitor who was confused or left the tab open. It is a weak signal in both directions.
Site-wide averages
An average conversion rate across all channels is close to meaningless. Branded search converts several times better than cold display traffic, so the average mostly tracks how your traffic mix shifted, not how your site performed. Always segment before concluding.
Getting a setup you can trust
Most analytics problems are data collection problems, and they are invisible unless you look for them. A dashboard built on broken tracking is worse than no dashboard, because it produces confident decisions from noise. Consent refusals add another gap on top; an analytics and dashboard setup checks both before you rely on the numbers.
- Define your conversions before installing anything. GA4 has called them key events since 2024. Two or three real ones beat fifteen vanity events.
- Exclude internal traffic. Your own team browsing the site can distort a small business’s numbers substantially.
- GA4 excludes known bots automatically; still check for spam referrals quarterly.
- Check that campaign tags survive every redirect. One redirect that strips parameters can silently reattribute an entire channel to direct traffic.
- Verify events fire once, not twice. Duplicated conversion events are common and quietly halve your apparent cost per acquisition.
- Write down what each event means. In six months nobody will remember what
form_submit_2was.
How often to look
Weekly for spend, monthly for performance, quarterly for strategy. Checking daily produces reactions to noise, and small sites are almost entirely noise at a daily resolution.
A site with forty conversions a month sees swings of thirty per cent from ordinary randomness. Reacting to those is not analysis, it is superstition with a dashboard. Give any change enough time to separate itself from the variance before drawing a conclusion.
The purpose of a report is to shorten the argument about what to do next. If it does not do that, it is a screensaver.
Serhii Yelbaiev, Well Web Marketing
Frequently asked questions
Usually not at first. One well-configured analytics platform plus Search Console covers most small businesses, and an SEO and AI-search audit shows what Search Console is telling you. Add a session recorder only when you have a specific question about behaviour that numbers cannot answer.
They always will. Ad platforms count conversions against the click that led to them and use their own attribution window; analytics assigns credit by its own model. Compare each source against itself over time rather than against the other. Our guide to attribution models explains why they never reconcile.
The only useful benchmark is your own previous figure for the same channel. Published industry averages combine businesses with nothing in common and mostly measure differences in traffic quality.
Enough conversions that ordinary variance cannot explain the change: usually a few hundred, which for many small sites means several months rather than several weeks. The same maths decides whether an A/B test is worth running.
Sources and method
- Google Analytics Help, Engagement rate and bounce rate
- Google Analytics Help, Known bot-traffic exclusion
- Google Analytics Help, Conversions vs. key events in Google Analytics
- Reflects analytics implementation and reporting work for small and medium clients. Last reviewed October 2026.