What is web analytics?

There are three collection methods in general use, and most real setups combine them.

• Client-side JavaScript. A script on the page reports pageviews and events. It sees everything the browser sees, including engagement time and viewport, and it can be blocked.
• Server-side. Your backend sends events directly. Nothing can block it, and it cannot observe anything that happens in the browser.
• Server logs. Every request your server handled, including bots and assets. Complete but coarse, with no view of client-side behaviour at all.

Client-side is the default because it answers the widest range of questions. Its weaknesses, blocking and lost requests, are what a first-party proxy and server-side events for critical conversions are for.

Whatever the vendor, the core set is the same, and it is worth knowing which layer each metric belongs to.

• Volume: pageviews, sessions, unique visitors. Three counts of the same traffic at three levels of grouping.
• Acquisition: referrers, search, campaigns, direct. Where the traffic came from.
• Behaviour: entry and exit pages, engagement time, bounce and exit rates.
• Outcomes: goals, conversion rate, revenue. The layer that connects the rest to money.

The outcomes layer is the one teams most often skip, and the one that makes everything above it worth reading. Traffic without conversions is a number nobody can act on.

Being clear about the boundary saves a lot of wasted effort.

It cannot tell you why. It records that 60% of people abandoned a form, never what they were confused by. Session replay narrows the gap; user interviews close it.

It cannot see what it was never sent. Dark traffic, cross-device journeys, word of mouth and anything a blocker removed are all absent, and absent data does not announce itself.

It cannot count people, only browsers. One person on a phone and a laptop is two visitors in every tool ever built, unless they log in.

And in a cookieless setup it cannot follow anyone across days, which is a deliberate design property rather than a defect. Long-window questions need either an optional cookie mode with consent or your own user id.

mrkr is web analytics that runs cookieless by default: it stores nothing persistent on the visitor's device, so it does not require a consent banner for its own measurement, and it therefore counts all of your traffic rather than the consenting fraction.

It covers the full set above, including goals and revenue, plus session replay and funnels. Bots and referrer spam are classified and dropped at ingest rather than filtered in the dashboard, so historical aggregates are clean too.

Where you need cross-day identity, an optional per-site cookie mode provides it and requires consent on that site. The default is the other way around because most teams want the banner gone more than they want the longer window.