Analytics is not the thing you add at the end.It is what makes everything else checkable.
Analytics and reporting set up so the numbers can be trusted, and read so your team knows what they mean.
Without it there is no before and after, nothing to base an assumption on, and no way to confirm what worked. That is why it is in every engagement we take, and why it is usually the first step.
Nice to have, or the first thing you do
How this is usually sold
As a service you might add later. Dashboards, a tag manager container, a monthly report with a lot of charts in it.
Nice to have. Bought last, cut first.
Why it is actually first
Without it there is no before and after, nothing to build an assumption on, and no way to confirm what worked and what did not.
Every recommendation we make rests on facts and figures. This is where those come from.
Why this page sits underneath all the others
Everything else on this site depends on this page being true
Read the rest of our services and you will find the same promises repeatedly: work done in measured batches, a control group, a number stated before we start, a falsifier written down in advance, and an honest account of what did not work.
None of that is possible without measurement you can trust. A control group you cannot separate in the data is not a control group. A before and after across a replatform means nothing if the tracking changed at the same time. “It worked” is an opinion until something counted it.
That is why analytics and reporting is included in every engagement we take, and why it is almost always the first thing we do. Not because we are an analytics company. Because the alternative is advising you on instinct and calling it strategy.
Gathering the data and reading it are different jobs
Setting it up is the easy half
Analytics, tag management, Search Console, Bing Webmaster Tools, dashboards that show the few things that matter. Necessary, and increasingly something a competent developer can do.
Anyone can install a tag.
Reading it is the half that pays
Knowing which number is real, which is an artefact, which movement is noise at your volume, and which one means stop.
We teach your team to do this, because a report nobody can interpret is a cost, not an asset.
This is the same distinction we make about an audit, one layer down. The audit gathers evidence and the assessment interprets it. Here, the setup gathers and the reading interprets. In both cases the second half is where the exercise is won or lost, and in both cases it is the half nobody sells.
What reading the data actually finds
Five things we found by reading the data, not by installing anything
- A dashboard tile showing the same figure for two unrelated companies. Not a coincidence, a field mapping defect. The correct numbers were in the account all along, several thousand apart, and the tile had been reported to both clients for months.
- Two stores on the same stack, one telling the truth. One reconciled its transactions to its purchases exactly. Its sibling reported 149 against 364 for the same period. Every revenue and conversion figure downstream of that was unusable and nobody had noticed.
- The trusted tracker was sending nothing. Two analytics plugins were installed and a double count was feared. In fact one was authenticated to the wrong account entirely and had never sent a single event. The feared problem was not real. The actual one was worse.
- Organic reported two ways, differing by roughly a third. Sessions overstate organic reach compared with users, and the gap widens exactly when referral patterns change. So we report organic on clicks and users, and say which we used.
- A perfectly continuous chart that was measuring two different things. After a replatform on the same domain, the analytics property continues, so the before and after looks directly comparable. If the tracking was rebuilt even slightly differently at cutover, it is not.
Every one of these was found by someone reading numbers rather than displaying them. Four of the five were invisible in the dashboard that was already in place.
What we work on
Analytics setup and repair
GA4 configured to answer a question rather than to be installed. Most of our work here is repair rather than setup, because something is usually already there and nobody is sure whether to believe it.
Tag management
Google Tag Manager as one deliberate system instead of six plugins that each think they own tracking. Knowing what is firing, and what is firing twice.
Search Console and Bing Webmaster
Both, verified properly, because they answer questions analytics cannot: what you were shown for, what got clicked, what is indexed and what is not.
Conversion and event tracking
What counts as a conversion, whether it is counted once, and whether the figure in the ad platform agrees with the figure in analytics. It frequently does not.
Campaign tracking
UTM conventions your team can actually follow, including QR codes, print and offline campaigns. Inconsistent tagging is the most common reason a channel looks like it is doing nothing.
Dashboards
Few numbers, chosen because a decision depends on them. A dashboard is a place to look when you have a question, not a monthly performance of activity.
Automated reporting
Pulling advertising, analytics and search data into one place on a schedule, so nobody spends a day a month copying numbers between tabs and introducing errors while they do it.
Teaching your team to read it
Which numbers are readable at your volume and which are not, what a real change looks like against normal variation, and when to do nothing. The part that outlasts us.
Knowing which numbers to ignore
The most useful thing a report can tell you is that a number is not readable
Most business reporting is built to look decisive. A percentage change, an arrow, a colour. At low volume, which is most B2B, the arrow is usually noise dressed as a finding, and acting on it costs real money.
So we say which of your numbers can carry a decision and which cannot. A conversion rate off eleven conversions is not a conversion rate. A ratio without its denominator is decoration. Indexation and markup are observed states you can check today; rankings are directional in two weeks and decision-grade in four; conversion needs a bigger sample than most catalogs produce in a month.
A team that knows which numbers to ignore makes better decisions than a team with a better dashboard.
Who this is for, and who it is not
It fits businesses whose figures have never quite reconciled, teams inheriting a setup nobody documented, companies about to commission expensive work who want a baseline first, anyone whose reporting takes a day a month to assemble, and B2B operations where the conversion counts are too small to steer by and nobody has said so out loud.
We will say no when:
- You want a bigger dashboard. More tiles is almost never the problem, and we would be taking money to make the real issue harder to see.
- You want the numbers to support a decision already made. They might not, and we would rather tell you before you spend the budget than after.
- Nobody will own it afterwards. Measurement decays. Without someone responsible, you are buying a snapshot and calling it a system.
- Your setup is genuinely fine. Sometimes it is. We will say so and not invent work.
Where this sits beside the rest
This is the foundation under everything else we do. On-page work is measured in batches against a control group, paid search is scaled only once it is proven, and a website migration lives or dies on whether the numbers after launch are comparable to the ones before it.
If you want the whole picture assessed rather than the instrumentation fixed, that is an SEO audit. For AI visibility specifically, measurement works differently and we cover it under answer engine optimization and generative engine optimization. The broader practice is on the services page.
Questions we get asked
Q. Do we have to buy this separately, or is it included?
Both, and the distinction matters. It is included in every engagement we take and it is usually the first thing we do, because we cannot make an evidence-based recommendation without it. It can also be the only engagement: plenty of companies need the instrumentation and the training and nothing else, including companies whose delivery work is being done by somebody else entirely. What it is not is an upsell attached to something else you bought.
Q. We already have Google Analytics. Is that enough?
Usually not, and usually for a boring reason rather than a dramatic one. Something was installed at launch, the business changed, and nobody revisited whether it still counts the right things. The common findings are conversions that count twice, a conversion that is really a page view, an ad platform and an analytics property that disagree by a third, and a tracker everyone trusts that stopped sending data months ago. The question is not whether you have analytics. It is whether anyone has checked what it is actually counting.
Q. What is a GA4 audit, and do we need one?
It is a check of what your analytics is really measuring against what you believe it is measuring, before anyone makes decisions with it. We look at what counts as a conversion, whether events fire once, whether the figures reconcile with your ad platforms and your own order records, and what is being lost to consent, filtering or a misconfigured property. You need one if any number you report has ever been questioned and the answer took longer than a minute.
Q. Can you set up dashboards for us?
Yes, and we will argue about what goes on them. A useful dashboard holds few numbers, chosen because a decision depends on them, and it is a place you go when you have a question. A dashboard with forty tiles is a performance of activity, and the more complete it looks the less anyone consults it. We would rather build the small version and teach your team to interrogate the underlying data when something looks odd.
Q. Can you automate our reporting?
Yes. Pulling advertising, analytics and search data into one scheduled place removes the day a month somebody spends copying numbers between tabs, and it removes the transcription errors that come with it. The part worth saying out loud: automation makes reporting cheaper and faster, and it does not make the numbers more correct. If the tracking underneath is wrong, you now have wrong numbers arriving reliably.
Q. Who actually reads the reports?
That is the right question and it changes what we build. A report nobody can interpret is a cost, not an asset, so part of the work is teaching your team what they are looking at: which numbers are readable at your volume, what normal variation looks like, and when the correct response is to do nothing. That last one saves more money than any optimization we have ever recommended.
Q. How do you measure AI search visibility?
Differently from organic traffic, and with the instrument named every time. Some sources report a live single query, which is a sample of one because AI answers are not deterministic. Others aggregate across a window and are far more stable. We label which is which, because a single live query that fails to mention you is not evidence that you are invisible, and it is the most common way people are misled on this right now.
Q. Our numbers have never reconciled. Is that normal?
Common, and it is not something to live with. Analytics, ad platforms and your own order records count different things at different moments with different definitions, so small gaps are expected. Large ones mean something is genuinely broken, and until it is fixed every decision made on those figures is a guess with a chart attached. We find the cause rather than averaging the discrepancy away.
Q. Do you use Search Console as well as analytics?
Always, and Bing Webmaster Tools too. They answer questions analytics cannot: what you were shown for, what was clicked, what is indexed, what was crawled and when. Analytics tells you about people who arrived. Search Console tells you about the far larger number who did not, which is usually where the opportunity is.
Q. What does this cost, and how long does it take?
It depends on how much is already in place and how much of it can be trusted, and we agree a fixed figure before anything starts. A straightforward audit and clean-up is short. Rebuilding measurement across several properties with a training component is not. We will tell you at the first call which of the two you are looking at, and if the honest answer is that your setup is fine we will say that instead.
Start with a conversation
Thirty minutes on what you currently report, who reads it, and whether anyone has checked what it is counting. Bring the number you trust least.