Web Analytics: Which Website Metrics Actually Matter

Why traffic is a vanity metric

Let us say it plainly: raw traffic tells you almost nothing about your business. You can double visits and earn nothing if the wrong people land on the wrong pages. Traffic looks great on a slide, but you cannot build a decision on it.

How traffic misleads

The classic story: an agency reports a 60% jump in visits, the boss is thrilled, and sales notices no difference. Look closer and the growth came from one informational article read in another country. Technically the visit counter is not lying, it is simply measuring something you do not need. Vanity metrics — visits, impressions, reach, followers — almost always go up and commit you to almost nothing: easy to inflate, impossible to check against revenue.

What to measure instead

A simple rule: a metric is useful only if you can act on it. If a number goes up and you have no idea what to do differently, it is just decoration.

• Target actions — form submissions, calls, messenger clicks, orders. Intent, not visits.
• Source quality — which channel brings people who actually reach the request.
• Cost of a result — how much money and time one lead and one customer really cost.
• Behaviour on key pages — where people get stuck and where they leave.

Traffic stays in the report as context: it explains why leads went up or down, while the question of whether the site works belongs to other numbers. And the less web analytics a business has set up, the more obsessed it is with visit counts — it is the only thing visible without configuration.

Business goals first, tools second

The typical order: install a counter, stare at the graphs, then wonder what to do with them. The correct one is the reverse — first define what good means for the business, and only then decide how to measure it.

Three questions before you install anything

1. What do we sell and how does a deal happen? An online payment, a request plus a call, and months of negotiation are three different analytics setups.
2. Which action on the site is closest to money? For a shop it is a completed order, for a studio a filled-in brief, for a service a signup and a first real use.
3. Who will look at these numbers, and when? If nobody is obliged to make decisions from them, the reports will die within a month.

From a business goal to a site metric

The goal increase sales cannot be measured on a site directly, because a site does not issue invoices. You have to break it into a chain: people arrive, reach the right page, understand the offer, leave contact details, and sales closes the deal. Each link has its own metric, and the problem almost always sits in one specific link rather than smeared across the whole chain.

That is why in our work on projects we start not with code or design, but with a conversation about the sales model. And keep goals few: one main goal and two or three supporting ones — when fifteen metrics are important, none of them is.

Sessions, users, sources: what is what

Half the arguments about reports happen because people read the same terms differently.

Users and sessions

A user is a specific visitor, usually identified by an ID stored in the browser. A session (a visit) is a single arrival with a sequence of actions that ends when the person leaves or goes idle. A user is a soft number: the same person on a phone, a laptop and an office computer counts as three, and clearing cookies makes a new one. So unique visitors is an estimate, not a census.

Pageviews and events

A pageview is a page opening. An event is any action: a click, a form submission, a scroll. Modern analytics counts almost everything as events, including the pageview itself.

Sources and channels

A source is the exact place a person came from: a search engine, a partner site, a newsletter. A channel is a group of sources: organic search, paid ads, direct visits, social, referrals. About direct visits: everything the system fails to identify lands here too — clicks from messengers, documents, apps, emails. Bloated direct traffic almost always means your links are not tagged. And tags are the cheapest investment in analytics there is: every link you placed yourself must be tagged, or it vanishes into direct traffic.

Bounce, engagement and page depth

This is the most myth-ridden group of website metrics: the place where a wrong conclusion is easiest to reach.

What is wrong with bounce rate

A classic bounce is a session with one pageview and no further action. The problem is it says nothing about whether the person was satisfied. Someone reads the answer to their question in two minutes and leaves happy — technically a bounce. Someone else sees a mess and leaves in three seconds — also a bounce. One number, two meanings. That is why modern systems moved to engaged sessions: a visit counts as quality if the person spent enough time or fired an event.

How to read engagement correctly

Look at it by page type and source, not as one site-wide figure. High bounce on the contact page is normal — the person found your phone number. High bounce on a paid landing page is an alarm: you are paying for something that does not fit the people arriving. And compare your own segments against each other, not against someone else's norms: your only honest baseline is yourself a month ago.

The technical side

Often bad behaviour is not a copy problem but a loading problem: the page lags, the layout jumps. Before you rewrite headlines, check speed and Core Web Vitals — this explains behaviour dips more often than any wording.

Conversion rate, cost per lead and LTV basics

Here begin the metrics the whole exercise was for: they connect the site to money.

Conversion rate

Conversion rate is the share of visitors who took a target action: target actions divided by sessions (or by users — just do not switch the denominator midway). Three things worth understanding:

• It is meaningless as a site-wide average — a mix of hot buyers and random passers-by. Compute it by channel, page and device.
• You can improve it by killing traffic. That is why conversion is always read alongside the absolute number of leads.
• It is capped by the offer, not the buttons. Button colour buys percentages, a clear offer buys multiples.

Cost per lead and per customer

Cost per lead is a channel's budget divided by its number of requests. On its own it deceives: a channel can deliver cheap leads of terrible quality. So next to it always sits cost per customer — budget divided by actual closed deals. The gap between those two numbers is the quality of the channel.

LTV in plain terms

LTV is how much a customer brings over the entire relationship, not just the first purchase. No sophisticated model at the start: take the average order value, multiply by the average number of purchases per year and by the customer's lifespan. Crude, but enough to see the main thing: a channel with expensive leads can be more profitable than one with cheap leads if it brings customers who stay. Without LTV you will cut your most profitable channels because they look expensive in last month's report.

Goals and events: what to track on a site

A goal is an event that has value to you. Without goals, analytics is just a counter.

The mandatory minimum

• Form submission — fired on a successful server response, not on a button click. Otherwise you count failed attempts.
• Phone click — a click on a tel: link. On mobile this is often the main lead channel.
• Messenger or email click — a separate event per channel.
• Add to cart and checkout — for a shop these are the foundation of the funnel.
• Scroll to 50% and 90% — cheap and informative for long pages.

Useful beyond the minimum

Form start paired with submission gives you the abandonment rate: if a hundred people start and twenty submit, the problem is the form, not the traffic.

Naming discipline

Set a rule before the first event: one case, a predictable structure like object_action (form_submit, phone_click, cart_add). A year from now the account will hold fifty events, and untangling a zoo of Lead!! and Form (bottom) is impossible. And decide which events are goals and which are observation: a goal must mean money. Declare everything a goal and 40% of visitors convert because they scrolled to the footer.

Sales funnels and attribution without jargon

Individual metrics show a state. A funnel shows exactly where you lose people.

How to build a funnel

A sales funnel is a sequence of steps toward a goal, where at each step you see how many moved on. For a service: visit → service page → form opened → form started → submitted → qualified lead → deal. For a shop: product page → cart → checkout → payment. The meaning lives in the transitions: find the step with the biggest drop-off — that is where your money is.

The funnel does not end at the site

The most valuable part lives after the request: how many leads turned out relevant, how many became deals. Without feedback from sales you optimise the number of requests, not revenue. We have seen projects where, once CRM data was connected, the best channel turned out to be the worst: a flood of requests, zero deals. To start you need no integration — a monthly export matched against sources by hand is enough.

Attribution in plain language

Attribution answers one question: who gets credit for a request when a person touched you five times?

• Last touch — credit goes to the last channel. Most systems count this way by default, but it overrates closing channels like branded search.
• First touch — credit goes to the channel that first brought the person in. Good for judging what creates demand.
• Distributed models — credit is split across all touches. Closer to reality, but harder.

Advice: do not hunt for the correct model, there is none. Look at the same campaign in two models — if the picture agrees, the conclusion is solid.

Choosing a tool: first-party, privacy, cookies

The tool is the last question, not the first. But the choice decides how far your numbers can be trusted.

Classic cloud systems

The big free platforms give you rich reports and ad integration, but their scripts live on third-party domains that ad blockers know by heart, so part of the data is lost.

What a first-party counter is

A counter is first-party when its script and data collection run from your own domain — for example, a subdomain like stats.yoursite.com. The browser sees a request to your site rather than to an external tracker. What that gives you:

• Blockers do not cut it — filter lists are built on the domains of known trackers, and yours is not on them.
• The data is fuller — the share of people with blockers is significant and, in technical or younger niches, well above average; in a classic counter those people do not exist.
• Cookies live longer — browsers cut first-party cookies far more gently than third-party ones.
• You own the data and do not depend on someone else's retention policy.

We run a first-party counter on Ostohlo's own site, and the gap against a classic counter is plain to the eye: two counters on one site show different numbers, and the first-party one is closer to the server logs.

Privacy and consent

Analytics is personal data, even without names. Decency looks like: an honest consent banner if you work with the EU or the UK; never sending phone numbers, emails or form contents into analytics; IP anonymisation; not storing data longer than needed. And the choice is simple: a small site needs only a counter with goals; a shop with ads needs analytics plus CRM; a technical project should look toward first-party.

How often to look, and how not to fool yourself

Data misleads not because it lies, but because we want to see confirmation in it.

Rhythm instead of panic

You need to look daily only in two cases: an active ad campaign is running, or you have just shipped a change. In normal mode a healthy rhythm is: weekly — a short glance at leads and sources; monthly — a review with conclusions and hypotheses; quarterly — reconciliation with money and strategy. Daily staring at graphs makes you react to random noise.

Seasonality and small samples

Never compare one week to another without accounting for the season. The right comparison is against the same period last year plus the previous period: holidays, vacations and industry cycles move numbers more than your edits. And the golden rule of small numbers: at 10 leads a month, the difference between 8 and 12 is not a trend, it is chance. The fewer the events, the longer the window you need.

A/B tests, honestly

An A/B test is when two versions of a page are shown to randomly chosen visitors at the same time. Three rules:

1. Simultaneously, not one after the other — otherwise you compare the weather, the season and the ads too.
2. One hypothesis at a time — change five things and you only learn it got better, never why.
3. Decide the duration in advance — stopping when your variant is ahead is the most popular way to fool yourself.

If traffic is low, an honest A/B test is off the table — session recordings, click maps and surveys work: five sessions often teach more than a month of watching a graph.

Common web analytics mistakes

We meet these on nearly every project that arrives with someone else's setup.

Tracking everything

It feels logical: collect it all, sort it out later. In reality, a year on you have two hundred events and no one uses the data. Analytics is not an archive, it is an instrument: collect what you will actually look at.

A counter with no goals

The most widespread situation: the counter is installed, visits roll in, there are no goals. That means no conversion, no funnel, no channel evaluation — only traffic, and we already know what that is worth. Setting up basic goals is a few hours of work that gets postponed for years.

Trusting a single number

Any metric alone lies. Conversion rose — but leads fell. Bounce dropped — but you broke the counter and it now counts every scroll. The rule: confirm any change with a second independent number, and better still with a fact from reality.

Double counters and dirty data

The code is pasted into the template and fired through a tag manager too — traffic doubles. Your own visits, bots and test orders are not filtered out. All of this is fixed once but distorts data constantly.

Comparing against someone else's benchmarks

A normal conversion is 2%. Whose? In which niche? For some, 0.5% is excellent; for others, 8% is a failure. The only correct reference is your own trend.

Analytics with no human

The reports are set up, the dashboards are pretty, but nobody owns the conclusions. Data becomes a ritual. If a metric has no owner, the metric does not exist.

Turning data into decisions

This is the step that decides whether analytics pays off or stays an expensive toy.

The loop that works

1. Spot an anomaly or a bottleneck — not traffic dropped, but conversion on the service page halved on mobile.
2. Form a hypothesis — the form is too long on a phone, people quit at the third field.
3. Verify with other data — session recordings, the abandonment rate, a chat with a manager.
4. Make one change — a short form, two fields.
5. Measure over a sufficient period and record the result in writing, failures included.

The key here is the last step. Keep a change log: date, what you did, what you expected, what happened. A year on, it is more valuable than any dashboard — the only thing that answers what actually works for you.

Where decisions usually hide

The cheapest wins almost always sit in three places. First — the request form and the path to it. Second — the hero of the landing page and the clarity of the offer; the approach in our piece on selling landing page structure helps here. Third — the mobile version of key pages. A good example at scale is our project, the 24freelance marketplace: analytics is built into the product rather than bolted on the side, and decisions are made from real behaviour data.

Analytics setup checklist

Let us gather it all into a practical list. Most sites stumble within the first third.

Foundation

• One main business goal for the site is defined and written down in words.
• The counter is installed on every page exactly once — no duplicates.
• Your own visits, office IPs and test orders are excluded from stats.
• You have decided first-party versus classic, and why.
• A cookie consent banner is set up if you work with the EU or the UK.
• The privacy policy describes what you collect in plain language.

Goals and events

• Each form submission — an event on success, not on click.
• Clicks on phone, email and every messenger — as separate events.
• Scroll to 50% and 90% on long pages.
• For a shop: product view, add to cart, checkout start, payment.
• Form start — so you can see the abandonment rate.
• An event naming scheme is agreed and written down.
• Only events that mean money are declared goals; the rest are observation.

Sources and money

• All your links are tagged: newsletters, posts, banners, QR codes.
• Ad accounts are linked to analytics, and direct traffic is checked for an abnormal size.
• Cost per lead is calculated for each channel.
• There is at least a manual monthly match of requests to deals, and LTV is estimated.

Process

• Every key metric has an owner — a specific person.
• A rhythm is set: week for a glance, month for a review, quarter for strategy.
• A change log is kept, with expectations and results.
• An alert is set for leads dropping to zero — tracking breaks silently.
• Every quarter you check that all goals still fire: they fall off after redesigns.

What to do tomorrow

Do not try to close the whole list at once. Do three things: set up tracking for form submissions and phone clicks, tag your own links, and appoint someone who looks at the report once a week. And if you want analytics done right the first time and tied to real sales — get in touch, and we will map your funnel and show where the money leaks out.

FAQ