AI Will Not Fix a Weak Marketing System. It Will Expose It.
Part 3 of 5 — AI in Marketing: From Experimentation to Commercial Advantage
Many conversations about AI in marketing eventually reaches the same honest admission: the results are not what we expected.
The tools are impressive. The demos are compelling. The early experiments seemed to work. But something between the proof of concept and the commercial outcome is not adding up.
Understanding what is genuinely blocking AI value in marketing is the most important thing a CEO can do before spending another pound on it.
Why Is AI in Marketing Underdelivering for Most Businesses?
The most consistent finding across every credible research source on AI in marketing is this:
the quality of AI output is entirely dependent on the quality of the input.
AI does not fix bad data. It operationalises it at speed.
Salesforce's State of Marketing research, drawing on nearly 5,000 marketers globally, frames this directly. Marketing teams want better personalisation and smarter customer engagement — but those ambitions depend on data that is unified, trusted and accessible.
In most businesses, that is not what exists.
Customer data sits across CRM systems that are not fully maintained. Campaign data lives in platforms that do not talk to each other. Sales conversations are not captured in a structured way. Web behaviour, product usage and customer service interactions are all separate.
When AI is applied to this environment, it is faster production of outputs based on incomplete or unreliable information.
Does More AI Content Actually Help Marketing Performance?
No — and this is one of the most important distinctions to make clearly.
One of the most visible side effects of AI adoption in marketing is a sharp increase in content volume. Teams that previously produced ten pieces of content a month can now produce fifty.
That sounds like productivity. In many cases, it is a problem.
Content that is fast to produce but generic in its positioning, weak in its insight and disconnected from a specific customer's actual need does not drive commercial outcomes.
It adds to the noise.
In a market where AI is making content production cheap for everyone, the brands that stand out will be the ones that invest in genuine distinctiveness.
CIM's 2026 guidance for marketers raises this directly: the risk is that AI produces entirely average content at a scale that drowns out sharper, more differentiated work.
A proposition that is not clearly defined will not become clearer because AI is producing more variations of it. Weak positioning scaled through AI is still weak positioning.
What Is the Real Governance Risk of AI in Marketing?
HubSpot's AI Trends data is striking:
- 67% of marketers believe AI will significantly change how their function works.
- But 30% say their organisation actively discourages AI use;
- And 20% say there is no clear policy at all.
The majority of marketing teams are making individual decisions about what to use, how to use it, and what to do with the output, without any agreed framework.
The DMA UK's 2026 AI position statement is explicit: customers should be informed when AI or automated decision-making is being used, should be able to request an explanation, and should be given the means to object.
In practice, most marketing teams are not applying this consistently because no one has defined where it applies, who is responsible for it, and what the process looks like.
For businesses in regulated sectors, like financial services, healthcare and insurance, this is a brand risk as well as a regulatory one. The FCA and ICO both have clear expectations around how customer data is used in AI-assisted processes. AI increases the risk if the underlying controls are not already in place.
Governance is not a bureaucratic overhead, allowing AI adoption to scale without creating brand, compliance or customer trust, on the contrary, is a huge risk.
What Is Stopping Most Marketing Teams From Using AI Effectively?
The skills gap is real — and significantly understated in most conversations.
Salesforce-adjacent research found that
- 39% of marketers are unsure how to use AI safely,
- 43% do not know how to get maximum value from it,
- 54% believe training is important,
- But, 70% say their employer provides none.
Those numbers reflect the missing commercial judgement:
Knowing when AI output is good enough, when it needs to be improved, when it should not be used, and what the downstream risks look like.
This matters most in customer-facing work and explains why senior leaders' judgements are essential. Proper and curated training needs to go beyond tool use. It needs to cover review and validation skills, commercial and strategic context, risk awareness, and when to escalate rather than proceed.
Before that process and governance is in place - a junior marketer who does not yet have the experience to decide whether a campaign message is strategically sound should not be approving AI-generated content at speed with no review process, as simple as that.
What Due Diligence Should a CEO Apply Before Expanding AI in Marketing?
Before expanding AI use in marketing, the right leadership questions are about readiness, not ambition.
Is our customer data actually in a usable state?
Check if your data is clean, connected, consistent and reliable enough to power AI-assisted decisions. If the answer is uncertain, fixing the data foundation is the first investment.
Do we have a governance framework, even a simple one?
What can AI be used for, what data can be entered into AI tools, which outputs require human review, and who holds final accountability? These questions do not require a heavyweight policy document. Have someone senior enough to make the call and communicate it clearly.
Are we measuring productivity or commercial impact?
Time saved is easy to count. Whether AI is actually improving the quality of customer understanding, proposition clarity, campaign performance, cost savings or incremental revenue is harder — and more important.
Where is human judgement non-negotiable?
In any business, parts of the customer relationship should not be delegated to an automated process. Knowing where those lines are, and making them explicit, is a leadership responsibility.
What is the biggest risk of using AI in marketing without a governance framework?
Brand inconsistency and compliance exposure are the most immediate risks. In regulated sectors, using AI in customer-facing communications without a review process can breach FCA or ICO expectations around explainability and fair treatment. Beyond regulation, the subtler risk is that AI quietly erodes the brand voice and customer trust that took years to build.
Why does more AI content sometimes make marketing performance worse?
Because volume is not the same as relevance. When teams produce more content without a sharper proposition or better customer insight underpinning it, they create more noise in already crowded channels. AI amplifies whatever strategy is already in place. If the strategy is weak, it amplifies the weakness.
How do we know if our marketing data is good enough for AI?
Ask four questions:
- Is our customer data unified across CRM, campaign, sales and product systems? I
- Is it regularly maintained and accurate?
- Is it accessible to the tools we want to use?
- And do we have consent and compliance controls in place?
If any of those answers is no or uncertain, that is the first investment to make.
What governance model should a mid-sized business put in place for AI in marketing?
It does not need to be complex. Define: what AI tools are approved for use, what data can and cannot be entered into them, which outputs require human review before going to market, and who holds final sign-off. A one-page policy agreed at leadership level is enough to start. The important thing is that it exists and is communicated.
Catherine Mak is Founder of ZILU consultancy She works as a Fractional CMO with growth-stage businesses, scaleups and organisations navigating AI adoption in their marketing functions. Clients include Citibank, Bupa Global and PEI Group.tart writing here...