Website Analytics That Answer Business Questions

The Symptom: A Dashboard Nobody Opens

Almost every company we work with has analytics installed and almost none of them trust it. The pattern is consistent: the implementation started with the tool rather than the question, so it now tracks two hundred events, three of which anyone has ever looked at. When a real question arrives — why did enquiries drop last month — the data cannot answer it, because nobody defined what an enquiry was.

The fix is not more tracking. It is deciding what the business actually needs to know and then building the smallest measurement system that answers it reliably.

Start From the Questions, Not the Tool

Sit down with whoever owns the number the business cares about and write down the three or four questions they ask every month. In practice they are nearly always some version of: where do good customers come from, which pages do the work, where do people give up, and is the trend moving in the right direction.

Those four questions define your entire tracking plan. Everything that does not serve one of them is optional, and optional tracking has a real cost — it slows the site, it adds consent surface, and it buries the signal you needed under noise you did not.

Define the Conversion Before You Measure It

Most measurement arguments are definition arguments in disguise. A form submission is not an enquiry if half of them are spam. A trial signup is not a customer. Write the definition down, agree it with sales, and make the tracking match it. If the useful event happens outside the website — a qualified call, a signed contract — say so and plan for how that gets back into the picture.

Why Tracking Everything Costs You

Teams add tracking because it feels free. It is not. Every additional tag is a script the browser has to fetch and run, which shows up in load time and therefore in the experience you are trying to measure. Every additional data point widens your consent obligations and the surface you have to be able to explain if anyone asks what you collect and why.

The subtler cost is attention. When a report has forty numbers on it, nobody knows which three matter, so the meeting becomes a tour of the dashboard instead of a decision. Restraint is not a limitation here; it is the feature that makes the data usable.

Events That Survive Staff Turnover

Naming is not cosmetic. A tracking plan with form_submit, Form Submitted and contactFormSuccess in it is a plan that will be rebuilt in two years because nobody can tell which is authoritative. Pick one convention — object then action, lower case, underscores — and apply it without exception.

Document each event in one shared place: what fires it, what parameters it carries, and which question it answers. That last column is the one that keeps the plan honest, because an event that cannot name its question should not exist. This is the same discipline that keeps a publishing operation from drifting, described in our guide to content operations.

Somebody Has to Own It

Analytics implementations rot because ownership is ambiguous. Marketing assumes the developers maintain it, the developers assume marketing checks it, and a release quietly renames a button six months later so an event stops firing. Nobody notices until a quarterly review shows an inexplicable cliff in the data.

Name one owner. Give them the tracking plan, the right to say no to new events, and a short checklist to run after every significant site change: do the core events still fire, do the values still look sane, and has anything been renamed. Ten minutes of that after each release prevents most of the data disasters we get called in to reconstruct.

Measuring With Consent, Not Around It

Consent requirements are now part of the design, not an obstacle to it. That means accepting that a portion of your traffic will not be measured directly, building a consent banner that does not dark-pattern people into agreement, and understanding which of your numbers are modelled estimates rather than counts.

The practical consequence is that you should treat analytics as directionally accurate rather than exact. Comparing this month with last month within the same imperfect system is usually valid. Comparing your analytics total with a platform’s reported conversions is usually not — those two systems are counting different things, a problem we cover in attribution after cookies.

It also helps to be explicit about what you will never be able to measure. Word of mouth, a recommendation in a private group, a conversation at a conference — these produce real revenue and leave almost no trace in any tool. Pretending otherwise leads teams to over-credit the channels that happen to be measurable and quietly defund the ones that are working invisibly.

From Metric to Decision

A metric earns its place when a change in it changes what somebody does. Bounce rate rarely passes that test. Enquiries by source, cost per qualified enquiry, and the conversion rate of the three pages that carry most of your traffic usually do, because each has an obvious action attached.

Build the report around those actions. One page, the same shape every month, with the number, the direction, and the decision it triggers. Anything that needs a paragraph of explanation before it can be read is not a report, it is a research project — and it belongs somewhere else. The same principle guides conversion work: measure the thing you are willing to act on.

Be equally deliberate about who reads the report. A number that only the marketing team sees will only ever change marketing behaviour. The same number in front of the person who sets the budget, once a month, in the same format, is what turns measurement into influence — and it is usually the difference between an analytics setup that survives a reorganisation and one that quietly stops being maintained.

Search Data Is Business Data

Do not leave search performance in a separate world. The queries people use before they arrive tell you what they think you sell, which is often not what you think you sell. Reading that alongside on-site behaviour turns two partial pictures into one useful one, and it is why the reports in Search Console belong in the monthly review rather than in an SEO silo.

A Cadence Instead of a Panic

Data reviewed only when something goes wrong is data used badly. A short monthly review, with the same four questions and the same page, does two things: it catches drift early, and it builds the shared vocabulary that stops the next disagreement from becoming a debate about whose dashboard is right.

Once a quarter, review the tracking plan itself. Remove events nobody used. Add the ones the last three months of questions revealed you were missing. An analytics setup is a living thing, and pruning it is as important as extending it. If you want that built around your commercial questions rather than a template, our performance marketing team starts with the questions and works backwards — and keeps the measurement tied to the metrics that matter, as set out in our guide to ROAS, CAC and LTV.