Your analytics numbers are lower than reality
Start here, because everything downstream depends on it and almost nobody has been told.
UK sites use a consent banner. Configured correctly, analytics storage is denied by default, and a visitor only becomes a session once they accept. Everyone who declines, ignores the banner, or leaves before deciding is invisible to GA4. They still visited. They still bought or did not buy. They are simply not in the denominator.
Here is that gap measured. On an e-commerce site I run, over one 28-day window: Google Merchant Centre reported 337 Shopping clicks, while GA4 reported 54 Organic Shopping sessions and 288 sessions across all channels combined. Merchant Centre counts clicks at Google's end, before any banner exists. GA4 counts sessions after consent. Both are correct measurements of different things, and the difference is enormous.
The fix is not to remove the banner. It is to keep a second, cookieless count of page views server-side — no personal data, no cookie, no consent required — and use that as the denominator for anything you are going to make a decision on. GA4 then does what it is good at: behaviour and journeys among the people who did consent.
Deciding what counts as a conversion
Before any code, one conversation: what does a good outcome look like for this business? The wrong answers here cost more than any technical error.
- For a service business: an enquiry, by any route. Form, phone tap, WhatsApp, chat, email link. All of them, or the number is fiction.
- For e-commerce: a purchase, with its value. Add-to-basket and checkout steps are useful diagnostics but are not the outcome.
- For a business selling something considered and expensive: often a partial step — a brochure download, a quote request, a booked call.
The most common error is counting a page view as a conversion. A thank-you page view is a proxy for a submission, not the submission — it is triggered by refreshes, by bookmarks, and by anyone who lands on it directly. It is also the easiest thing to set up, which is exactly why it is everywhere.
The second most common error, in trade and service businesses specifically, is counting only form fills. If most of your customers ring, an account optimised on form submissions is optimising toward a minority of your business — and Google Ads will faithfully spend your budget chasing the wrong people.
How it is actually set up on a new site
Six steps, in this order, and the order matters because each one depends on the last being right.
- Consent first. The banner, configured so that analytics and advertising storage default to denied and are granted only on acceptance. Doing this after the tags are live means every number from the first month is a different measurement from every number after it.
- GA4 with real events, not just page views. One event per enquiry route: form submitted, phone number tapped, chat opened, email link clicked. Named consistently, so that six months later a report can be built without archaeology.
- A cookieless page-view count alongside it. Server-side, no personal data, no cookie. This is the honest denominator and the reason any conversion rate quoted afterwards means something.
- Search Console connected. Free, unaffected by consent, and the only source of truth for what people searched before they arrived. Anything that matters gets stored in a database the client can query rather than left to expire at sixteen months.
- Google Ads and Meta conversions fed from the same events. Not separately defined in each platform — the same events, so that the channels are at least measuring the same thing when they disagree about who deserves the credit.
- Reconcile against reality. Count last month's enquiries in the inbox and the phone log. Compare with what the tracking says. If they differ, the tracking is wrong, and it is wrong now rather than in three months when a decision is made on it.
That last step is the one that separates working tracking from tracking that merely exists. Almost every setup I inherit has never been reconciled with an actual count of enquiries, and a surprising number of them are out by more than half.
Is server-side tracking worth it?
Sometimes, and it is oversold. Server-side tracking sends events from your own server rather than from the visitor's browser, which makes them more reliable — browser extensions, ad blockers and tracking prevention all remove browser-side events.
When it is genuinely worth the complexity:
- You spend enough on ads that a 10–20% improvement in signal quality is worth real money.
- You sell to a technical audience, where ad-blocker use is high enough to distort the picture materially.
- You need a durable server-side record of conversions independent of any platform, for reconciliation.
When it is not: a local business spending £500 a month on ads. The setup and ongoing maintenance will cost more than the improvement is worth, and the same effort spent on counting phone calls properly would return several times as much. A simple cookieless page-view endpoint gets you most of the honest-denominator benefit for a fraction of the work.
Reading the numbers without fooling yourself
- Always name the source and the window. "288 sessions" is meaningless; "288 GA4 sessions, all channels, 28 days to 4 September, consented only" is a fact.
- Never blend a consented number with an unconsented one in the same ratio. Merchant Centre clicks over GA4 sessions is not a conversion rate, it is a category error.
- Compare full calendar months against full calendar months. Partial periods and rolling windows produce changes that are entirely artefacts of the dates.
- Exclude yourself. Your own visits, your developer's, and your office. It sounds trivial until you notice a small site where internal traffic is a fifth of everything.
- Expect the channels to disagree. Google Ads, Meta and GA4 each attribute by different rules and will each claim the same sale. The reconciliation is a judgement, and anyone presenting a single blended number without saying so is hiding the judgement rather than making it.
Every case study on this website carries the tool it was read from and the date, for exactly these reasons. It is not decoration; it is what makes the figures checkable — and a figure nobody can check is a claim, not a measurement.
