There is something slightly strange about modern marketing reporting.
The numbers have never looked more precise.
£73,482.17 in revenue.
4.82 ROAS.
£31.47 CPA.
312 conversions.
Everything neatly arranged inside a dashboard, calculated to two decimal places and presented in a way that suggests there is very little room for interpretation.
And yet the reality underneath those numbers has arguably never been less certain.
That distinction matters.
Precision is not the same thing as certainty.
And I think one of the biggest changes happening in marketing reporting is that our job is no longer simply to explain performance.
Increasingly, we also have to explain how confident we are in what we're seeing.
Reporting was designed to give us answers
For years, the basic promise of marketing reporting has been fairly straightforward.
Spend £X.
Generate Y conversions.
Produce £Z in revenue.
Calculate ROAS.
Compare it to last month.
Make a decision.
Obviously, anyone working deeply in measurement knows reality was never quite that simple.
Customers have always interacted with multiple channels before converting. Attribution models have always involved assumptions. Offline behaviour has always been difficult to connect perfectly to online activity.
But our reporting interfaces made everything feel remarkably definitive.
Put enough numbers into a clean dashboard and uncertainty becomes surprisingly easy to hide.
The problem is that those uncertainties are becoming harder to ignore.
Even within Google's own ecosystem, attribution is based on choices about how credit should be distributed across the customer journey. Google Analytics, for example, offers data-driven attribution alongside last-click approaches, and its data-driven model uses machine learning to estimate the contribution of different interactions.
Change the model and you can change the story.
Change the conversion window and you can change it again.
Neither result necessarily means the data is wrong.
It means the number has context.
And increasingly, that context matters as much as the number itself.
Two platforms can disagree without either being broken
This is one of the hardest things to communicate in reporting.
A client opens Google Ads.
Then they open GA4.
Then perhaps Shopify, HubSpot or their CRM.
The revenue doesn't quite match.
The conversion count is different.
The channel attribution is different.
And the natural response is:
Which one is right?
Sometimes there is genuinely a tracking problem.
But sometimes the uncomfortable answer is:
They can all be right within the rules of what they are measuring.
Google itself acknowledges that discrepancies between Google Ads and Analytics can happen even when the implementation is correct.
Differences can come from attribution settings, lookback windows, conversion timing, counting methodology, invalid traffic treatment and even whether a conversion is recorded against the date of the ad interaction or the date the conversion actually occurred.
That creates an important distinction.
A discrepancy is not automatically an error.
Sometimes it is a difference in methodology.
But dashboards are generally very good at displaying the result and fairly bad at displaying the methodology that produced it.
So we end up comparing two numbers that look equivalent when they are actually answering slightly different questions.
Some of the numbers are now estimates
This gets even more interesting when we move beyond attribution.
Some marketing conversions are no longer directly observable at all.
Privacy restrictions, browser limitations, consent requirements and cross-device behaviour can prevent platforms from connecting an advertising interaction to a later conversion.
Platforms increasingly use modelling to fill those gaps.
Google describes its modelled conversions as estimates for conversions where the link between the advertising interaction and the conversion cannot be observed directly. The models use observable data to estimate what happened within the unobservable portion.
That's not inherently a bad thing.
Quite the opposite.
A well-designed model can give us a more useful representation of reality than simply pretending the unobservable customers don't exist.
But it changes the nature of the number.
A conversion total might contain:
- directly observed conversions;
- modelled conversions;
- attributed conversions;
- cross-device assumptions;
- different lookback windows;
- and different rules about which interaction receives credit.
The dashboard still gives us one number.
But there is a lot happening underneath it.
More data doesn't necessarily create more certainty
There was a period where the answer to marketing measurement problems seemed obvious:
Get more data.
Connect more platforms.
Build a bigger dashboard.
Create a customer data platform.
Bring CRM, advertising, analytics and revenue information together.
And much of that is absolutely useful.
But there's a strange point where more data can actually make the reporting conversation harder.
Imagine an agency reporting for an ecommerce brand.
Google Ads says one thing.
Meta says another.
GA4 provides another version of the customer journey.
Shopify tells you what was actually purchased.
The CRM adds information about the customer afterwards.
Perhaps there is also email, organic search, affiliate marketing and increasingly AI-assisted discovery somewhere in that journey.
You haven't necessarily created one version of the truth.
You may have created five different perspectives on the truth.
That isn't a technology failure.
It is partly the unavoidable consequence of measuring a customer journey from multiple vantage points.
The mistake is pretending those perspectives are interchangeable.
AI makes this even more important
And then AI arrives.
This is where I think things get particularly interesting.
One of AI's greatest strengths is its ability to take complicated information and explain it quickly.
Instead of a marketer spending two hours looking through campaigns, an AI system can potentially identify anomalies, summarise performance and generate an explanation within seconds.
That is incredibly useful.
But there is a new risk.
AI is very good at making ambiguous information sound definitive.
Give a system an incomplete dataset and it can still produce a beautifully structured explanation.
Revenue dropped because of this.
ROAS increased because of that.
Campaign X drove the improvement.
Audience Y caused the decline.
The prose can be convincing long before the evidence is.
That changes the responsibility of the reporting layer.
We shouldn't just be asking:
Can AI explain the data?
We should also be asking:
Does the data actually support the explanation?
The wider measurement industry is already wrestling with this. IAB's 2026 State of Data report describes privacy regulation, signal loss, platform optimisation and fragmented data environments as challenges to connecting advertising exposure with outcomes, while also warning that AI without sufficient transparency, governance and data quality can reinforce black-box decision-making.
In other words, faster analysis doesn't remove uncertainty.
Sometimes it simply allows us to reach a confident answer faster.
AI search is giving us a preview of what comes next
Search is a particularly interesting example.
Traditional SEO reporting has decades of familiar metrics behind it.
Rankings.
Clicks.
Impressions.
Sessions.
Conversions.
Now brands also want to know how visible they are inside ChatGPT, Gemini, AI Overviews and other AI experiences.
But the measurement infrastructure isn't equally mature.
A brand might be repeatedly mentioned or recommended inside AI-generated answers without receiving a click that can be neatly attributed in its analytics platform.
Semrush's current guidance around AI visibility explicitly distinguishes this environment from traditional search measurement and recommends using directional signals where direct attribution isn't yet possible.
I think we'll see more of this across marketing.
Not fewer measurements.
Different classes of measurement.
And reporting needs to get much better at explaining the difference.
Perhaps every metric needs a confidence layer
This is where I think reporting needs to evolve.
Instead of treating every number as equally definitive, perhaps we need a language for describing what kind of number we're looking at.
Something like:
1. Observed
We directly recorded the event.
A transaction occurred.
A form was submitted.
Revenue entered the CRM.
There may still be implementation questions, but the underlying event itself has been observed.
2. Modelled
We cannot observe the complete picture, so a statistical model estimates the missing portion.
This can still be extremely useful.
But the user should understand that modelling exists.
3. Attributed
The event happened, but we are deciding which marketing activity receives credit.
The conversion may be certain.
Its ownership is not.
4. Directional
The metric tells us something useful about movement or relative performance, but we shouldn't pretend it establishes exact causation.
AI visibility is a good current example.
So are many brand metrics and upper-funnel signals.
None of these categories means good data or bad data.
They answer different questions.
The problem starts when we present all four with exactly the same level of confidence.
Maybe the dashboard needs to show its working
This could change how we design reporting.
Imagine clicking a revenue metric and seeing:
Source: Shopify Measurement: Observed transaction revenue Attribution: GA4 data-driven attribution Lookback window: 90 days Known limitation: Some cross-device journeys modelled Confidence: High for revenue, moderate for channel attribution
That is far more useful than simply:
Revenue: £184,392
It doesn't make the reporting less trustworthy.
I think it makes it considerably more trustworthy.
Because uncertainty isn't the same thing as unreliability.
Unacknowledged uncertainty is.
This changes the job of the reporting team
For a long time, marketing reporting was largely about collecting information and presenting performance clearly.
Then the challenge became connecting more platforms.
Then blending the data.
Then automating the dashboards.
Now we're moving into something slightly different.
The reporting team increasingly has to become an interpreter.
Not just:
What happened?
But:
What do we know happened?
What do we think happened?
What is being modelled?
What depends on the attribution methodology?
What is only directional?
And what decision can we reasonably make from all of that?
That last question is probably the most important one.
Because executives don't really need perfect measurement.
Perfect measurement rarely exists.
They need enough reliable information to make a better decision.
Good reporting shouldn't remove uncertainty
For years, we've designed dashboards to make uncertainty disappear.
Everything has a number.
Everything has a decimal point.
Everything has a chart.
And everything looks wonderfully precise.
But marketing doesn't become certain just because we've visualised it nicely.
As customer journeys become more fragmented, privacy changes what can be observed, platforms rely increasingly on modelling, and AI starts interpreting the numbers for us, I think our definition of good reporting has to change.
The best reporting won't be the reporting that pretends to have every answer.
It will be the reporting that clearly separates what we know, what we estimate, what we attribute, and what we can only use as a directional signal.
Because a dashboard can make uncertainty invisible.
Good reporting should make it understandable.