As part of our always-on listening through our event pulse with Play Verto we know that AI in marketing is top of mind for our UAE Members. This piece looks at how using AI without critical oversight risks optimising biases which can lead to brands confidently making wrong decisions.
AI in marketing works. The problem lies in the system of assumptions we ask it to optimise.
When optimisation becomes normalised bias
Partly because of hype, and partly because it remains a genuinely interesting topic, AI continues to fuel an almost constant stream of conversations, both online and offline, a continuous flow of positions, reflections and strategic rationalisations. There is nothing inherently wrong with this. The issue arises when part of the debate begins to normalise the idea that algorithmic bias does not distort outcomes, but optimises them.
This is where the fracture emerges. In this context, bias is not perceived as an error, but as a signal that something is working. When optimisation that works starts reinforcing existing prejudices, we end up with models analysing other models: synthetic systems, scalable, coherent and often complacent.
The moment brands stop questioning insight
It is at this point that brands make their most significant mistake. They accept that synthetic consumer insights confirm their own vision and expectations, without questioning the context in which trends, motivations and behaviours can change rapidly. The result is the ability to make the wrong decisions with a high degree of confidence.
When insight reinforces narratives instead of challenging them
The complacency generated by insights that appear solid, because they are supported by models and data, begins to influence how consumers are defined, how their habits are interpreted and how their future trajectories are shaped, without properly considering the medium to long-term impact of these decisions. This is a challenge that future marketers will need to face with clarity and critical thinking.
Basing analysis on data that is generated and interpreted outside of structured contexts and clear guidelines reinforces the dominant narrative. Insights that appear objective are, in reality, culturally and commercially oriented, because they reflect what the system has been trained to see and value.
Bias across the marketing value chain
This mechanism does not apply to a single area. It can be observed, with similar effects, across the entire marketing value chain, from audience segmentation to media budget allocation. How much is AI truly optimising for the future, and how much is it simply confirming the current status quo?
If a platform performs well in a market, the system will recommend using it. This is a rational decision, but not necessarily a strategic one. What is often missing at this stage is a broader perspective, one that questions not only where to invest, but why and for how long.
Retaining strategic control
To prevent algorithmic bias from taking over, what I try to do is work on how the system is instructed, reducing external noise and favouring broader time horizons over short-term logic. This is not a definitive solution. AI will still tend to comply with the instructions it receives and return outputs that are coherent with the boundaries imposed.
But the point is not to educate AI, it is to train ourselves not to let it take control of our strategies.
5 Key Points
AI works by optimising assumptions, not questioning them
It performs efficiently within the boundaries it is given, even when those boundaries are flawed.
Algorithmic bias often looks like successful optimisation
Because it improves short-term performance, bias is rarely perceived as a problem.
Synthetic insights tend to confirm internal narratives
Models analysing other models reinforce expectations instead of challenging them.
Bias scales across the entire marketing system
From insight to media allocation, optimisation systematically reinforces the status quo.
The real risk is losing strategic control to optimisation.
AI should support decision-making, not replace strategic judgement.
3 Take aways
AI makes decisions more robust, not necessarily more accurate
When optimisation is mistaken for truth, the risk is making the wrong decisions with great confidence.
Algorithmic bias is not an anomaly, but a systemic effect.
It originates in initial assumptions and propagates across the entire marketing value chain.
Responsibility remains with the marketer, not the model
Governing AI means maintaining critical control over the assumptions that shape insights and strategy.
2 Action items
Make assumptions explicit before optimising
Before using AI generated insights, clearly define which assumptions the system is built on and which ones must be challenged rather than reinforced.
Introduce a strategic counter-check to AI outputs
For every AI driven recommendation, ask what decision would change if the insight were wrong, and use real world signals to validate it beyond model coherence.
"The risk is making the wrong decisions with great confidence."
