For years, being good at digital marketing meant being good at optimisation.
You adjusted bids.
Refined keywords.
Built audiences carefully.
Changed placements.
Moved budgets.
Tested landing pages.
Rewrote ads.
A/B Test and many more.
A good performance marketer could spend hours inside an advertising platform finding small improvements that, collectively, made a meaningful difference.
But for the past 4 years or so that job is changing.
Google's AI Max for Search can now expand beyond existing keywords, match ads to relevant searches using broader signals, customise ad copy and even select the landing page it believes is most relevant to a user's intent. Google describes it as a continuous optimisation layer for Search campaigns.
And that is only one example of a much broader shift.
More of the tactical optimisation that once happened manually is moving inside the platform.
Which creates an interesting question:
When the platforms increasingly optimise themselves, what exactly is left for the marketer to optimise?
Quite a lot actually (and way more that it might seem from first glance).
But the control surface is moving.
And that shift is one of the reasons AI marketing analytics is becoming much more important.
The marketer's advantage is increasingly moving upstream: towards the quality of the data, the goals being supplied, the context surrounding them, the creative ideas available and the measurement used to judge whether the system actually worked.
AI can optimise very quickly.
The harder question is whether we have given it the right thing to optimise for.
But first things first let's establish some terminology.
What is AI marketing analytics?
AI marketing analytics is the use of artificial intelligence and machine learning to analyse marketing data, identify patterns, explain performance, forecast outcomes and support marketing decisions.
In practical terms, it can help marketers understand questions such as:
-Why did conversions fall last week?
-Which channel is contributing most to revenue growth?
-Are there unusual changes in campaign performance?
-Which customers are most valuable?
-Where should additional budget be allocated?
-What is likely to happen if current trends continue?
Increasingly, however, AI isn't simply analysing marketing after the activity has happened.
It is participating in the activity itself.
Advertising platforms can now use machine learning to determine bids, audiences, search-query matching, placements and creative combinations while campaigns are running.
That collapses a traditional sequence.
Marketing analytics used to look roughly like this:
Campaign → Data → Report → Analysis → Decision → Change
Increasingly, it looks more like:
Data → AI decision → Campaign adjustment → New data → AI decision
The feedback loop becomes continuous.
And once that happens, the quality of what enters that loop becomes extremely important.
Automation doesn't remove optimisation. It moves it upstream.
One of the easiest mistakes to make with AI in marketing is assuming that automation means fewer things need to be optimised.
The opposite may be true.
There may simply be fewer manual campaign settings to optimise.
Google's AI Max is a useful example.
Its search-term matching can use broad-match and keywordless technology to identify searches beyond the advertiser's existing keyword list. Its text customisation can use existing ads, assets and landing-page copy to generate more relevant messaging. Final URL expansion can select another relevant page on the advertiser's website when the system believes it is more appropriate for the query.
That reduces the amount of manual decision-making required inside the campaign.
But it increases the importance of everything feeding the campaign.
If your landing pages are poor, allowing AI to choose between them doesn't solve the problem.
If your conversion tracking is wrong, smarter bidding doesn't solve the problem.
If your creative is indistinguishable from everyone else's, generating more combinations doesn't solve the problem.
If your definition of a valuable customer is wrong, the system may become extremely efficient at finding the wrong customers.
This is where I think the marketer's optimisation stack is changing.
The new marketing optimisation stack
For a long time, optimisation happened mainly at the execution layer:
-Bid.
-Keyword.
-Audience.
-Placement.
-Campaign.
-Ad.
Increasingly though, the more valuable optimisation is happening one level above that.
I would think about the new stack in five layers:
Data → Goals → Context → Creative → Measurement
The advertising platform can increasingly handle execution.
The marketer has to make the other five layers better.
1. Data: what are we actually teaching the system?
This is the least glamorous part of AI marketing.
It may also be the most important.
Across marketing reporting and analytics work, one problem appears again and again: companies often have plenty of data without necessarily having usable data.
-Google Ads says one thing.
-GA4 says another.
-The CRM has different campaign names.
-Revenue exists in an ecommerce platform.
-Lead quality exists somewhere else.
-Offline sales may never make their way back into the advertising platform.
-Some conversions are duplicated.
-Others are missing.
And then AI gets placed on top of it all.
The intelligence layer may be sophisticated.
The data underneath it can still be messy.
That matters because AI doesn't magically know that a field has been incorrectly configured or that two systems define a customer differently.
It simply has more processing power with which to analyse the information it receives.
Consider a basic example.
An agency has a client generating leads through paid search.
Campaign A generates 200 leads.
Campaign B generates 120.
If the advertising platform only sees lead submissions, Campaign A looks much better.
But suppose the CRM tells a different story.
Campaign A produces ten qualified sales opportunities.
Campaign B produces thirty-five.
The advertising platform hasn't necessarily failed.
It has been given an incomplete definition of success.
This is exactly why platforms are investing heavily in first-party data infrastructure.
Google Ads Data Manager, for example, is designed to bring customer information from external sources such as CRM systems and data warehouses into Google Ads so it can be used for measurement, audiences and optimisation.
Google's enhanced conversions similarly use first-party data to improve conversion measurement and bidding.
The broader lesson is simple:
Better AI does not reduce the need for better marketing data. It increases it.
2. Goals: what does “good” actually mean?
Once the data is available, the next question becomes surprisingly difficult.
What should the system optimise towards?
Marketers often default to the easiest measurable outcome.
-Form submissions.
-Purchase.
-CPA.
-ROAS.
But those are proxies for business value.
Sometimes very good proxies.
Sometimes terrible ones.
Imagine an ecommerce company selling two products.
Product A generates £200 of revenue with a £150 gross margin.
Product B generates £300 of revenue but only £80 of gross margin.
A system optimising purely for revenue may prefer Product B.
The finance team might prefer Product A.
Neither interpretation is mathematically wrong.
They are optimising towards different definitions of success.
Lead generation creates the same problem.
If one channel generates cheap leads that rarely become customers while another produces expensive leads that close regularly, cost per lead can actively push optimisation in the wrong direction.
Google's own guidance around offline and enhanced conversions recommends feeding later-stage outcomes such as qualified leads and converted leads back into Google Ads.
That's an important direction of travel.
We are slowly moving from:
“Find me more conversions.”
towards:
“Find me more of the conversions that actually matter.”
And that requires marketing, sales, finance and data teams to agree on something algorithms can't decide for them:
What does a valuable outcome actually look like for this business?
3. Context: the algorithm doesn't work inside your company
There are things every marketer knows about their business that don't naturally appear inside Google Ads or Meta Ads.
-A particular service may have no capacity left this month.
-One product might generate far more repeat purchases than another.
-A certain customer segment may produce a much higher lifetime value.
-One region might be strategically important despite lower short-term ROAS.
-A campaign may deliberately sacrifice immediate conversions to create future demand.
--A sales team may know that a certain type of enquiry almost never closes.
The algorithm cannot automatically infer all of that from a click.
This is where business context matters.
More data helps.
Connecting CRM and revenue systems helps.
Better conversion events help.
But there is still a layer of human judgement between what the system can optimise and what the organisation actually wants.
This becomes more important, not less, as platforms automate execution.
If humans are no longer making every tactical decision, their responsibility increasingly becomes defining the environment within which those decisions are made.
The marketer moves from constantly turning the knobs to deciding what the machine should be trying to accomplish in the first place.
4. Creative: producing assets is becoming cheap. Ideas aren't.
Generative AI has dramatically reduced the cost of producing marketing assets.
A marketer can generate dozens of:
-headlines,
-advertisement variations,
-images,
-email subject lines,
-landing-page sections,
-video scripts,
-and audience-specific versions
in minutes.
This is useful.
It also creates an uncomfortable economic reality.
If everyone gains access to roughly the same production capability, production itself becomes less of a differentiating factor.
The scarce resource moves elsewhere.
-The idea.
-The positioning.
-The customer insight.
-The offer.
-The perspective.
-The emotional understanding behind the creative.
AI can give you fifty variations of an advertisement that says:
Save time with our innovative solution.
The fact that it produced fifty versions does not make the underlying proposition interesting.
This is why I don't think creative automation makes marketers less important.
It may make creative judgement more important.
The machine can generate permutations.
Somebody still has to give it something worth permuting.
5. Measurement: did the algorithm improve the business, or just the metric?
This is the layer I think becomes most important over time.
Advertising platforms are very good at optimising metrics inside their own environments.
Businesses operate outside those environments.
A platform might report:
-more conversions,
l-ower CPA,
-higher ROAS,
-or greater conversion value.
All useful information.
But the business may need to ask different questions.
-Did total revenue increase?
-Did profit increase?
-Did we acquire more new customers?
-Did lead quality improve?
-Did customer lifetime value change?
-Did we create incremental demand?
-Would some of those customers have bought anyway?
-Did one channel merely take credit from another?
Those are harder questions.
They require cross-channel marketing analytics, CRM information, revenue data and sometimes experimentation rather than relying entirely on platform attribution.
This is where AI marketing analytics has the potential to become genuinely valuable.
Not because it can write a nicer paragraph underneath a dashboard.
Because it can help marketers investigate increasingly complex relationships across large datasets faster.
But measurement still needs independence.
There is an obvious conflict in allowing an advertising platform to both optimise your advertising and act as the sole judge of whether that advertising worked.
Platform reporting is valuable.
It shouldn't necessarily be the entire measurement system.
Where automated marketing reporting fits
AI marketing analytics and automated marketing reporting are closely related, but they are not the same thing.
Reporting automation solves the movement of information.
Instead of manually downloading Google Ads data, exporting Meta Ads, checking GA4, copying CRM numbers and rebuilding spreadsheets every week, marketers can connect those sources automatically.
That creates a consistent data layer.
AI analytics can then operate on top of it.
The distinction matters.
Reporting automation answers:
Can we get the data into one place reliably?
AI marketing analytics answers:
What can we understand from it?
There is a third question that remains harder:
What should we actually do?
AI can increasingly help answer that too.
But this is where context and judgement become critical.
Suppose an automated report identifies that conversion rate has fallen by 18%.
An AI system may trace most of the decline to paid social.
Useful.
But the real explanation might be:
-a tracking change,
-a stock shortage,
-a website issue,
-a seasonal shift,
-a deliberate campaign change,
-a pricing increase,
-or poorer lead quality being filtered further down the funnel.
The analysis has shortened the investigation.
It hasn't necessarily completed it.
That is an important distinction when thinking about AI-powered marketing reporting.
The real value isn't removing humans from the process.
It is removing the repetitive work that prevents humans from getting to the interesting questions quickly enough.
Cross-channel marketing analytics becomes more important as individual platforms automate
There is another consequence of increasingly automated advertising.
We may understand less about exactly why an individual platform made every decision.
That makes the wider view more important.
A business might advertise through:
-Google Ads,
-Meta Ads,
-LinkedIn Ads,
-TikTok,
-email,
-organic search,
-affiliate marketing,
-and increasingly AI-driven discovery.
Each platform has its own reporting logic.
Each may claim some contribution to the same customer.
Looking at them independently gives us useful operational information.
Looking across them gives us a better business picture.
This is why cross-channel marketing analytics matters.
The interesting questions increasingly sit between systems.
-Which channels create demand and which capture it?
-Do branded searches increase after paid social campaigns?
-Which acquisition sources produce the best customers six months later?
-What happens to total revenue when one channel's spend changes?
-Are multiple platforms taking credit for the same conversion?
-Where does marketing performance look strong inside an ad platform but weak once CRM outcomes are included?
These aren't dashboard questions in the traditional sense.
They are investigation questions.
And they are exactly the kind of questions AI becomes useful for once the underlying data is organised properly.
The biggest AI marketing problem may still be a data problem
There is a tendency with every major technology shift to focus on the exciting layer.
With AI, that layer is intelligence.
-Ask your data questions.
-Generate commentary.
-Predict results.
-Automatically optimise campaigns.
-Build agents.
But after working through enough marketing datasets, there is a much less exciting truth underneath all of this:
You can't intelligently analyse data you don't understand.
-The same metric can mean different things across systems.
-The same campaign can be named differently across platforms.
-Revenue can be attributed differently.
-Conversions can use different windows.
-Customer identifiers may not match.
-Historical data may contain tracking changes nobody documented.
Suddenly the question:
“Why did revenue fall last month?”
isn't a simple AI prompt.
It is a data architecture problem!!!
The future of marketing analytics therefore isn't just about better AI.
It is about creating an environment where AI has something trustworthy to reason from.
And I suspect this is where many businesses will discover the real work is.
How marketers should prepare for AI-driven campaign optimisation
The temptation is to respond to increasing automation by learning every new AI feature as quickly as possible.
There is value in that.
But I would prioritise something more fundamental.
Build the data foundation first.
Make sure important marketing, analytics, CRM and revenue information can be connected and compared.
Define the commercial outcomes that matter.
Give advertising platforms better conversion signals instead of relying purely on easy-to-measure top-of-funnel actions.
Understand where business context exists outside the marketing platforms.
Invest in differentiated creative ideas rather than simply generating more assets.
And maintain a measurement layer capable of challenging what the individual platforms report.
Because ultimately, automation changes execution much faster than it changes strategy.
The marketer isn't disappearing. The control surface is moving.
There will probably be fewer reasons for marketers to spend their days manually changing bids by 10%, creating endless audience variations or adjusting hundreds of campaign settings individually.
Machines are simply better suited to make many of those decisions.
They can process more signals, more frequently and at far greater scale.
But that doesn't make marketing automatic.
It moves the valuable decisions from:
Which bid should I change?
to:
Which business outcome should we optimise towards?
From:
Which audience should I select?
to:
What do we actually know about our most valuable customers?
From:
Which campaign gets credit for this conversion?
to:
Did this activity create incremental business value?
From:
How do I optimise the platform?
to:
How do I give the platform better information to optimise from?
That is the shift I think matters most.
The marketer's job is increasingly becoming the job of teaching the system what “good” looks like.
And in that environment, the business with the most sophisticated AI won't automatically win.
The advantage may belong to the business with the best combination of:
data, commercial goals, context, creative judgement and measurement.
The AI sits on top.
The foundation still determines what it can do.
Skip this step and results can be more than unpleasant!
Until next time.