How we do it

How we build analytics around the decisions your business needs to make

Start with who reads the reports. Understand the decisions they need to make. Then work out what to measure.

Written by Sam Bailey
16 November 2026

 

Most website reports I see are put together the same way. Someone logs into Google Analytics at the start of the month, copies the figures into a slide or a Word document, and compares them with last month, looking at where visitors came from, how many there were, which pages they looked at, and how many forms were submitted.

The numbers go up or down, and the report says so, but it often can't explain why or answer the questions the people running the business are actually asking. They want to know which parts of the business the website is working for, and whether any of that activity is turning into work.

That's reporting rather than analysis. Reporting tells people what happened; analysis starts with a decision someone needs to make and works back to the data that would help them make it.

In this article, I've outlined how we put that into practice, from setting up the tracking and bringing different data sources together to building dashboards and keeping the reporting accurate and useful over time.

 

What are leading and lagging measures, and why do they matter?

For most professional services firms, the result that matters often happens months after someone first visits the website. Someone reads an article in March, comes back twice in April, gets in touch in May, and the work is agreed over a couple of meetings in June. The website has played its part, but by the time the result arrives, it's recorded in the CRM or finance system rather than Google Analytics.

March
April
May
June
What happened
Reads an article
Comes back twice
Gets in touch
Work agreed over two meetings
Where it's recorded
Google Analytics
Google Analytics
Google AnalyticsThe CRM
The CRM or finance system
Nothing in Google Analytics
Kind of measure
Early signal
Early signal
Headline measure
Lagging measure
The website's record stops in May. The result arrives in June, somewhere else.

That's a lagging measure. It tells you what the outcome was, but usually arrives too late to help you understand what's happening now or where you should focus your attention.

Leading measures give you earlier indications of what might happen. The right audiences finding the right service pages, more people searching for you by name, or repeat visits from a target organisation could all be signs of growing interest. None guarantees an enquiry or a piece of work, but they give you something to monitor while you wait for the results.

You need both, and you need to understand how they're used. The way we organise our reporting owes a lot to Avinash Kaushik, whose writing on analytics we've followed for years, particularly his approach to giving different people the measures they need to make decisions.

We group those measures into three levels.
 

Level What it tells you Who it's for Examples
Headline measures Whether the website is contributing to the business Leadership team or board Enquiries and the value of work won
Supporting measures Where changes in the headline figures are coming from Marketing and business development leads Enquiries by service, sector and source
Early signals Where interest is developing and what might need attention Marketing and the teams responsible for acting Visits to priority service pages, searches for your name, repeat visits from target organisations

 

Each group makes different decisions, at a different pace, and a report that shows everyone everything tends to be useful to no one.

There's one more distinction worth making before any of this gets built, because it determines how much confidence you can put in the numbers.

Some figures are counted directly, such as recorded form submissions or work won in the CRM. Others depend on attribution rules, such as crediting an enquiry to the first article someone read. And some are estimates, because the exact figure isn't available, such as the expected value of an enquiry or a third-party measure of visibility in AI answers.

All three have their uses, but they shouldn't be treated as though they're equally precise. We use directly recorded figures wherever we can, make the attribution rules clear, and label estimates so nobody mistakes them for actual results. A lot of reporting goes wrong when those distinctions are lost.

 

HOW WE DO IT

1 Start with the decisions, not the data

Every project starts with a conversation about who's going to read the reports and what they need to decide. It takes an hour or two of the client's time in the first week, but it's time well spent because it determines everything we need to build afterwards.

For each group of readers, we work through three questions.

  • What decisions do they make? Someone deciding where to invest next year needs something very different from a marketing executive deciding which article to promote this week.
  • How often do they make them? Some decisions are made weekly and some once a year. A board member doesn't need a daily dashboard, and if they've been given one, that's usually a sign that they're being asked to look at the wrong measures.
  • What would change their mind? If no number could change the decision, it probably doesn't need to be on their report.

Choosing what goes into each report

Those answers tell us which measures belong in each of the three levels we set out earlier. Leadership gets the headline measures that show whether the website is contributing to the business, marketing and business development get the supporting detail to understand what's changing, and the people responsible for acting get the early signals that tell them where to look.

For the headline measures, we also agree what good looks like with the people who'll be judged on them; a number on its own tells you very little, but the same number against a target tells you whether there's something worth investigating.

We also try to get those headline measures as close to the money as we can. An enquiry is a reasonable place to start, but it's still a long way from a piece of work being won. The best set-ups we've worked on are the ones where the client has connected their CRM to the website data, so a deal that closes months later can be traced back through the recorded digital activity, with the value of that work sitting alongside the visits and enquiries.

Not every organisation can do that on day one, so we start with the closest thing to the money the data will support, and plan the route to the rest.

Separating branded and non-branded search

One decision we make early, because it can make a big difference to how the headline numbers are interpreted, is to separate the people who searched for the organisation by name from those who didn't. Branded searches are largely driven by existing awareness, whether that's come from reputation, referrals or previous marketing activity, whereas non-branded searches give us a clearer view of people discovering the organisation while looking for help with a problem.

Both matter, but they tell us different things. Reporting them separately gives us a much better understanding of where the website is helping the organisation reach new audiences, rather than treating all search traffic as though it represents the same opportunity.

Keeping the scope under control

Ask a room of senior people what they'd like to see and you'll get forty measures, most of them interesting, but few of them capable of changing a decision.

We hold the headline level to six measures at most, and everything else has to earn its place lower down. The scoping is also what sets the cost of the build, because a data layer that has to break everything down by twelve things is a much bigger job than one that breaks it down by three or four. So the conversation about what people actually need to see pays for itself twice.

Making sense of small numbers

Another key consideration is volume, and it's one that most reporting ignores. Professional services websites don't typically produce thousands of enquiries a month. If a service gets around twelve enquiries a month, one month with fifteen isn't necessarily a trend; it could just be the normal variation in a relatively small number.

At those volumes, even a meaningful improvement can take several months to become clear, and smaller changes take longer. Reacting to every monthly movement risks making decisions based on noise rather than a genuine change in performance.

So before we set a target, we check whether the data could actually show the improvement we're looking for. Where it can't, we either adjust how we're measuring progress, report over a longer period or lean more heavily on the early signals in between. It's much better to know that at the start than to spend a year reacting to noise.

With the decisions and measures agreed, the next step is making sure the website records what we need.

 

HOW WE DO IT

2 Add a data layer so you can report by service, sector and content type

Out of the box, Google Analytics knows very little about what your pages mean to the business. It records their addresses, titles and other basic information, but it doesn't know that one page is about employment law, another is an article by a particular director, and a third is written specifically for manufacturers.

A data layer helps fill that gap. It's a way of making additional information about each page available to the tracking, such as which service it belongs to, which sector, what type of content it is, who wrote it and when it was published.

Much of that information already exists in the CMS. It's the tagging your editors apply to organise the website and decide where content appears, so rather than asking them to maintain another set of tags for reporting, we make that information available through the data layer and configure Google Analytics to record it.
 

What GA4 knows about a page by default
Basic page information, without the business context
Page address Page title
Data layer
→
CMS metadata passed through and configured in GA4
What GA4 can report on with a data layer
Existing page information, available as reporting dimensions
Page address Page title Service Sector Content type Author Campaign Date published
Making better use of the information already in your CMS, so you can compare content performance by service, sector and content type.

Once that's in place, we can start answering questions that would otherwise mean working through pages and spreadsheets. Which service areas are attracting visits and enquiries? Are the guides the employment team spent the summer writing actually being found? Which types of content are people visiting before they get in touch?

Content type earns its place on that list for a reason that isn't immediately obvious. A guide will often answer questions from earlier in the audience journey, while a service page may become more relevant when someone is comparing providers. That isn't always the case, and page type alone doesn't tell you exactly what someone is trying to achieve, but it's a useful way of grouping content and understanding how people move through the site.

It's also where this work connects with the audience mapping approach I covered in my previous article. The map identifies the questions people ask at each stage; the tagging helps us understand how the content written to answer those questions is being used.

For one large law firm, every service page, article and profile carries its service group in the data layer, so reporting across more than a hundred service lines comes down to a filter rather than a spreadsheet exercise, and that breakdown is now what the group's content strategy is built on. For a building society, tagging pages by product and content type helps the team understand which content people use before applying, and track demand product by product.

Setting up conversions

Alongside the tagging, we set up the conversions; the actions on the site that count. We keep to one main conversion, usually a completed enquiry, and a handful of supporting ones, such as a brochure download or newsletter sign-up. Track twenty things as conversions and you risk treating actions of very different importance as though they all represent the same level of interest.

The data layer matters here too, because a conversion on its own is just a count. With the page information and the visitor's route recorded alongside it, we can see which services and content types someone engaged with before getting in touch, rather than just knowing that another form was submitted.

Where the CRM can tell us what happened next, we bring that back in; whether the enquiry was genuine, whether it became work, and what that work was worth. Where it can't yet, we can use an estimated enquiry value based on historical conversion rates and the typical value of work, where the data supports it.

We label that clearly as an estimate, rather than presenting it as revenue. It's the third kind of number we discussed earlier; useful for understanding direction, but not something to read to the pound.

Getting the tagging right

The first hurdle is the tagging itself. The data layer is only as good as the information in the CMS, and in most organisations that information has been added by different people over several years. Articles with no service attached, pages tagged to a service that's since been renamed, articles with the author missing; all of them show up as gaps in the reporting.

We fix the existing tags before trusting the breakdowns, and agree a consistent set with whoever publishes the content so the problem doesn't keep coming back.

A key hurdle is that this doesn't work backwards. The additional information is recorded when a page is viewed or an action takes place, and you can't go back and add missing data-layer values to events that have already been collected.

You may be able to reconstruct some of the historical picture from page addresses, but that isn't the same as having recorded the information properly at the time. It's a good reason to start early, even with a handful of useful tags, rather than waiting until everything is perfect.

One other thing worth remembering is that Google Analytics doesn't give you a complete count of everyone visiting the website. Consent and tracking configurations affect what gets recorded, and some implementations include modelled data, so everyone using the reports needs to understand what the figures do and don't represent.
 

Beyond tracking: using data to inform original research

If your website has a tool or calculator, it's worth treating it as a source of data in its own right. For the building society above, the dashboards already show how people who use its calculators behave compared with everyone else, and the calculators are used thousands of times a month, considerably more often than the forms on the site.

There's potentially another opportunity in what people enter into those tools. Depending on the questions being asked, a calculator or estimator might reveal patterns in the kinds of problems people are trying to solve, the budgets they're working with or the outcomes they're considering.

Where that information can be collected appropriately and analysed in aggregate, without exposing individual users, it could provide the basis for original research into what your audiences are planning, worrying about or budgeting for.

That's something your organisation may be uniquely placed to publish, and one of the reasons the audience mapping approach can lead to digital tools and calculators rather than just more pages.

With the website recording what we need, the next step is bringing in everything that happens before and after the visit.
 

HOW WE DO IT

3 Bring in what Google Analytics can't see

Google Analytics records what people do on your website when they're being tracked, but much of what helps explain how they got there and what happened afterwards sits in other systems.

We bring those sources together so they can be read alongside the website data, giving the people making decisions a more complete picture of how the website is performing.
 

What Google Analytics can't see

The website visit is only part of the picture.

Discovery and visibility

How people find you and where your content appears

  • + Search Console: search queries, impressions and clicks
  • + Google AI visibility: impressions of links in AI Overviews and AI Mode
  • + Tracked campaigns: visits from identifiable campaign links, including offline advertising and QR codes
Website activity

What visitors do on the site, where tracking is available

  • ✓ Google Analytics: pages viewed, engagement and enquiries
  • ✓ CMS data layer: service, sector, content type and other page information
  • + Company identification: some of the organisations visiting, where identification is possible
Commercial outcomes

What happened to enquiries and what work they became

  • + The CRM: qualified enquiries and sales opportunities
  • + Work won: which opportunities became clients
  • + CRM or finance system: the value of the work won
Separate evidence source: AI server logs

Requests from verified and reasonably trusted AI crawlers and assistants show which pages automated systems are accessing and whether those requests were served. They don't prove that a page appeared in an AI-generated answer or that a person visited the website.

Google Analytics gives you part of the picture. The other sources help explain how people find you, what happens to their enquiries and where AI systems are accessing your content.

Bringing the different sources together

Here's what each source contributes.

  • Search Console: This shows what people were searching for, how often your pages appeared in Google and how many people clicked through. It's also one of the most useful measures of brand awareness we've found that doesn't need a survey; an increase in searches for your organisation by name is a useful indication that more people are looking for you specifically. One thing to watch is that Google occasionally changes how it counts and reports search activity, so year-on-year comparisons need checking before anyone reads too much into them.
  • The CRM: This tells us which enquiries were genuine, which became opportunities, whether they turned into work and what that work was worth. It's where the lagging measures live, and it's why connecting the CRM to the website data is worth the effort. For a consultancy we work with, the website, email channels and CRM are joined up, so a piece of work signed in June can be traced back through the recorded marketing activity to a campaign six months earlier. Part of making that work was establishing a clear distinction between marketing enquiries and sales opportunities, so the sales team only sees qualified opportunities rather than every form submission.
  • Offline campaigns: Print adverts, local advertising and QR codes on posters can all be given trackable routes into the website, allowing visits and subsequent activity to be attributed to the relevant campaign. For the building society, that's how the marketing team follows a brand campaign across print, paid social and local advertising in one view, rather than trying to piece together the results from separate reports. It doesn't tell you how many people saw an advert without responding, but it gives you a much clearer picture of what happened when they did.
  • Company identification: Tools like Leadinfo or Lead Forensics can identify some of the organisations visiting a website, even when nobody fills in a form. For professional services firms selling to other businesses, seeing that a target organisation has been visiting relevant service pages can be a useful early signal. There are limitations, though. Not every organisation can be identified, and identifying a company isn't the same as knowing which person visited or what they were looking for. The tracking also needs to be set up in line with the relevant privacy requirements.
  • AI assistants: Google Analytics now has a dedicated channel for visits referred by AI assistants, but it doesn't capture all of them because some arrive without identifiable referral information and may be recorded as direct traffic. We read that alongside two other sources. Search Console's generative AI reporting shows how often links to your pages appear in Google's AI Overviews and AI Mode, while our server logs show requests from AI systems we've verified or classified as reasonably trustworthy. These are three different measures, and none gives you the complete picture on its own. The logs also provide two useful signals for the technical team; the proportion of AI requests the server successfully handles, and whether a bot that normally accesses the site regularly has suddenly stopped appearing. I've covered how we collect, verify and interpret that log data in How we read AI crawler data.

Making sense of data that doesn't agree

The first hurdle is that the sources don't agree. They count different things, use different time zones and sometimes identify the same page in different ways, so bringing them together takes more care than it sounds.

A visit in Google Analytics and a click in Search Console aren't the same thing, and a dashboard that simply adds them together is wrong. We need to understand what each source measures, where the data can genuinely be joined and where it should be compared side by side.

The second hurdle is the gap between the website and the commercial outcome. Even with the CRM connected, not every piece of work can be traced back to a visit. Some clients come through referrals or conversations at events, while others use several devices or visit long before anyone records an enquiry.

We show what can be traced, say plainly what can't, and where we've had to estimate, we show a range rather than a single number.

With the different sources brought together, the next step is turning the data into something people will actually use.

 

HOW WE DO IT

4 Build dashboards that answer "so what?"

We build dashboards in an interactive reporting tool, typically Google's Data Studio, which updates automatically as new data comes in. That saves someone having to log into Google Analytics every month and copy figures into a slide or Word document, giving the team back hours of work.

It's usually the first benefit people notice, but it isn't the main point. What matters is that every page of the dashboard is designed to answer one of the questions we agreed in step 1.

Different reports for different readers

The leadership view carries the handful of headline measures, each sitting alongside its target and the same period last year, so there's some context for deciding whether anything needs attention. We highlight the changes worth investigating and leave the rest a click away; a page listing every service line every month can easily bury the two that actually changed.

Each page also carries a short explanation of what's happened, what we think might have caused it, and what needs doing next. That's the "so what?" that a page of charts on its own can't provide.

Lower down, every early signal has someone responsible for it. If the proportion of AI requests the server successfully handles has dropped, that's for the technical team to investigate; if the sector pages are appearing in search but attracting few visits, that's something for the content team to look at. It's a small discipline, but it helps turn a dashboard from something people look at into something they act on.

Showing the numbers honestly

Where the data is uneven, we avoid relying on averages alone. A handful of large pieces of work can distort the average value of a project, so we show the median and the spread alongside the total. We take the same approach to the time between someone's first recorded visit and their enquiry, where there's enough data to make a useful comparison.

We also mark events that might have affected performance directly on the charts, whether that's a change to how Google counts, the launch of a new AI model or a site migration. Those events don't necessarily explain a change, but they're useful context when someone is trying to understand what happened, and help avoid taking the credit or blame for something outside the organisation's control.

Search and AI in one view

Someone finding you through an AI assistant and someone finding you through Google may be trying to answer the same question, so it makes sense to look at both in the same report.

We bring together search visibility, clicks, AI referrals and AI crawler activity so we can see how each is changing over time. They're different measures, and we don't add them together, but looking at them alongside one another helps us understand how visibility and traffic are changing across the different ways people discover your content.

Testing it with the people who'll use it

Before connecting everything to live data, we share a draft so the client can challenge the structure while it's still relatively easy to change.

And before a dashboard page goes live, we try it on someone who'll actually use it. We show them the page for five seconds, take it away, and ask what stood out and where they'd look next. If they can't identify what matters, the page gets reworked. It's a simple test, but it can reveal problems that are easily missed in a review meeting.

Starting small and building on what works

The first hurdle is the temptation to build the ultimate dashboard. Everyone wants to see everything, and the result can be something impressive that nobody opens after the first month.

We'd much rather start small and add to it as people begin asking for more. The building society's dashboards have roughly doubled in size since they launched, and every addition came from someone asking a question the existing pages couldn't answer. That's the right way round.

The second hurdle is that a dashboard doesn't analyse itself. It can show that enquiries for one service have dropped for three months running, but it can't tell you whether that's because of the website, the market or a senior person who's left. That still needs someone to investigate, understand what might have changed and recommend what to do about it.

Once the dashboards are live, the work changes from building them to keeping them accurate and useful, which is the last step.

 

HOW WE DO IT

5 Keep the data clean, and keep asking what it means

The numbers need looking after. Automated traffic can cause sudden spikes in visits, spam can appear as fake referrals, and a change to the website can break the tracking on a form without anyone noticing. Left alone, any one of those can distort the reporting and lead people to make decisions on the wrong figures.

We check the data every month, investigate anything that looks out of place and fix problems with the tracking as they arise. Some unwanted activity can be excluded from reports, but it isn't always possible to correct historical data, so prevention and early detection matter.

Then there's the analysis; looking at what's moved, working out why, and recommending what to do about it. Over time, three habits make that process more useful.

Keeping a record of what we expected

Every target gets three things written next to it; the figure we're aiming for, what we think will move it, and what actually happened.

So if the plan is that six new sector guides will increase enquiries from that sector by a fifth over the year, we record that expectation at the start and put the actual result alongside it at the end.

Over time, that record becomes one of the most useful parts of the reporting. It helps us understand which activities have been associated with improvements, which haven't delivered what we expected, and where we need to look more closely before drawing conclusions.

It also makes next year's targets easier to set, because they're informed by what happened previously rather than starting from a guess.

Retiring measures

Some measures stop earning their place, either because the decision they informed has been made or because the business has moved on. We review what's being reported regularly and expect some measures to change each year.

A dashboard that never changes is often a sign that nobody is questioning whether it still answers the questions people are asking.

Watching where search is going

As AI assistants answer more questions directly, some of the visits that previously came through search may disappear, but the effect won't be the same for every type of content.

Once a quarter, we look at searches that don't include the organisation's name and group the pages by content type, using the CMS tagging from step 2 alongside the Search Console data. We then compare each group with the same period the previous year.

Three figures help us understand what's changing; how often the pages appeared in search, their average position, and how often people clicked through.

If impressions and rankings are holding steady but clicks are falling, people may be getting the information they need without visiting the site. AI answers could be part of that, although changes to the search results and other competing content can have the same effect.

If impressions are falling, we need to investigate whether demand has changed or whether the pages have become less visible. And if rankings are dropping, we look at what's changed in the search results, the competing content and the pages themselves.

None of those figures tells the whole story, but together they help us distinguish between changes in demand, visibility and the proportion of searches that result in visits.

In the sites we monitor, guides and explainers are often more exposed to these changes than service pages, which is why we look at content types separately. It helps us understand where the risks are developing and where it might be worth investing more effort.

How we're using AI

Google Analytics now offers ways to ask questions about your data in plain English, and there are tools that connect AI assistants directly to analytics platforms. They're useful, and they're improving quickly, but they're only as good as the information available to them.

Ask one how the employment team's articles are performing and, if the pages don't carry a service tag, it has no reliable way of identifying which articles belong to that team. It may give you an incomplete answer or simply tell you it can't answer the question.

The data layer from step 2 is what makes that information available, whether it's a person or an AI assistant asking.

On our side, AI has done much the same for analytics as it did for the log pipeline in our article on AI crawler data. It's helped us build things more quickly, from the tagging rules to the first version of each dashboard, and it drafts the monthly commentary for someone to check and correct.

What it doesn't do is decide what success looks like. For a professional services firm where the commercial outcome might arrive months after someone first visits the website, analytics data alone can't tell an AI assistant which outcomes matter most to the business or what targets it should be working towards.

Those decisions still need to come from the people running the business, which takes us back to where we started in step 1.

If you're going to try this yourself

There are three things worth keeping in mind, because they're likely to trip you up if you're not expecting them.

The wish list

Every stakeholder will have measures they'd like to see, and most of them will sound reasonable. Start with the decisions people need to make, keep the headline level to a handful of measures, and be prepared to say no to the rest, or at least move them into the supporting detail.

And make sure branded and non-branded search are reported separately before anyone gets attached to the combined figures. They tell you different things, and treating them as one can give a misleading picture of how the website is performing.

The tagging

A data layer is only as good as the information in your CMS, and the additional data only starts being recorded once the tracking is in place.

Before building anything on top of it, check that your pages are consistently tagged by service, sector and content type, and agree with whoever publishes the content how those tags will be maintained.

Don't wait for the perfect version, though; three or four reliable groupings in place now are better than twelve that arrive next year.

The numbers you can't see

Google Analytics doesn't record every visitor, it can't tell you what happened to an enquiry after it left the website without connecting other systems, and at the volumes most professional services firms work with, a lot of month-to-month movement is simply normal variation.

Know which figures are counted directly, which depend on attribution rules and which are estimated. Show a range where you're estimating, check whether there's enough data to detect a meaningful change before setting targets, and make sure everyone reading the dashboard understands what it doesn't capture.

We run this as a short project, usually over about four weeks, and then keep the data accurate and the analysis going for as long as it's useful.

None of the principles depends on us, though. An organisation that starts with the decisions its leadership needs to make, gets its CMS tagging right and builds a focused dashboard for each group of readers will already have a much better basis for making decisions than one copying figures into a slide every month.

 

If you'd rather get some help Start measuring what matters

In a free consultation, we'll look at what you're reporting on now, who it's for, and what it would take to make it answer the questions your business is actually asking.
 

Book a free consultation

FAQs

Questions we get asked about analytics and reporting

A lagging measure records an outcome after it has happened, such as new work won or the value of that work. It's accurate but late. A leading measure moves first and tends to come before an outcome, such as target organisations visiting service pages, more searches for an organisation by name, or the same company returning several times. Useful reporting uses both, and is clear about which is which.

A data layer is a small piece of code on every page of a website that passes extra information to Google Analytics each time the page is viewed, such as which service or sector the page belongs to, what type of content it is, who wrote it and which campaign it's part of. The information comes from the tags editors already apply in the CMS. It lets every figure in the analytics be broken down by the groupings an organisation actually reports on, and it records from the day it's installed; it can't be applied to past data.

Someone who searched for an organisation by name was already going to visit, so the credit belongs to its reputation and referrals rather than to the website. For most organisations that sell expertise, those searches are the majority of search visits. Taking them out of the headline figures leaves the people who were looking for help with a problem and found the organisation, which is the part the website can influence.

By joining the website data with the CRM, so that enquiries can be followed through to the work they became and what it was worth. Where that isn't possible, an enquiry can be given an estimated value based on the share of enquiries that become work and the typical value of that work. Either way, figures that are estimated should be labelled as estimates and shown as ranges.

Partly. Google Analytics now has a channel for visits from AI assistants, but many of those visits arrive without identifying where they came from and are recorded as direct, so the figure undercounts. It's best treated as a baseline and read alongside Search Console's report on Google's AI answers and the website's server logs, which show which AI systems are reading which pages.