Prediction can tell a company what customers are likely to do. It cannot, on its own, tell the company what to change.
A customer enters an online shop, adds a product to a basket, reaches checkout and leaves. The system records the actions it can observe and may estimate how likely customers with similar behaviour are to abandon the purchase. What it does not necessarily know is why.
Perhaps the price felt too high. Perhaps unexpected delivery charges appeared at checkout. Perhaps the customer did not trust the payment process or simply did not see enough value in the product. Or perhaps something interrupted the purchase. The observable outcome is the same; the reasons may be entirely different.
Over 15 years in marketing, I have seen how easily numbers become explanations. Sales fall and we conclude the price is too high. Visitors abandon shopping baskets and we decide they were never serious buyers. One message generates more engagement and we say we have discovered what consumers want. In each case, the data captured an outcome; we supplied the cause.
That distinction matters even more as artificial intelligence becomes better at recognising behavioural patterns. A 2026 methodological study in Scientific Reports, using controlled simulations, illustrates why predictive success should not be treated as causal explanation: a model can predict accurately by exploiting correlations without identifying the causal mechanism behind the observed outcome.[1]
If the prediction is accurate, why should a business care about the cause? Because prediction does not tell it what to change.
If a customer does not buy because the price is genuinely too high, the company may need to reconsider its pricing or offer. If the customer does not understand the product’s value, the problem may lie in communication. If trust is missing, the business may need greater transparency. If checkout is confusing, the problem may lie in the customer experience.
The predicted outcome is the same: the customer is unlikely to complete the purchase. The appropriate business response is not.
Prediction tells a business what is likely to happen. Understanding helps it decide where to intervene.
This is an important limit of prediction: identifying a likely outcome is not the same as establishing its cause.
A useful discipline is to move through a simple sequence:
Stimulus → response → pattern → hypothesis → test → decision.
A single response is not yet a pattern; repetition is what makes it one. The crucial step is testing the hypothesis, because a pattern can be real while the explanation attached to it is wrong.
Consider one of the most common things a business hears from a customer: “It’s too expensive.”
Sometimes that statement is literal: the customer cannot or does not want to pay the price. In other cases, it means: “I don’t see why it is worth that much.”
Those are different problems. If affordability is the barrier, the pricing or offer may need attention. If perceived value is the problem, reducing the price may sacrifice margin while leaving the underlying communication weakness unresolved.
There may be other explanations. The customer may prefer a competitor, may be postponing the decision or may simply be declining politely. The point is not to infer a hidden motive from one sentence. It is to treat the first explanation as a hypothesis rather than a conclusion.
Understanding behaviour, then, is not an abstract exercise in psychology. It has economic value because it helps a company identify where intervention is actually needed.
The issue becomes more significant as personalisation grows more sophisticated. A business may learn which customers respond more strongly to urgency, which are more likely to buy and, potentially, which appear willing to pay more. Behavioural prediction can therefore shape not only which advertisement a person sees, but how an offer is presented.
In Cyprus, that possibility already has a legal dimension. Article 17(1)(ε1) of Consumer Protection Law 112(I)/2021, as amended, includes among the pre-contract information requirements for distance and off-premises contracts an obligation, where applicable, to tell the consumer that the price has been personalised on the basis of automated decision-making.[2] Article 6(1)(ea) of the European Union’s Consumer Rights Directive contains the corresponding requirement.[3]
Neither provision bans personalised pricing. Where the disclosure obligation applies, it requires the consumer to be told that the price has been personalised in this way.
European data protection law adds another layer. The General Data Protection Regulation defines profiling as automated processing used to evaluate aspects of an individual, including economic situation, preferences, interests and behaviour. It also gives individuals the right to object to processing for direct marketing, including related profiling. Article 22 addresses the narrower category of decisions based solely on automated processing that produce legal effects or similarly significant effects, subject to specified exceptions and safeguards.[4]
The broader point is that the ability to know more about a customer does not settle how that knowledge should be used.
There is another limit that law alone cannot define: trust. Research published in the International Journal of Research in Marketing, based on nine controlled experiments involving 2,887 participants, found lower brand trust when consumers believed they were disclosing information to artificial intelligence rather than to a human. The effect was partly explained by expectations that the information would reach a larger audience, increasing participants’ sense of exploitation.[5]
This does not mean consumers distrust every use of artificial intelligence. It shows that how information is collected and perceived can affect trust.
A business may infer that a customer responds strongly to urgency, is likely to buy or appears willing to pay more. But the ability to make that inference does not answer whether acting on it will strengthen the relationship or damage it.
For years, marketers have tried to collect more data in order to understand consumers better. The risk now is assuming that more data automatically produce deeper understanding.
A business may know which message generates more clicks without knowing which one builds more trust. It may know who is more likely to buy without understanding what that person actually needs. It may accurately predict behaviour without understanding the human being behind it.
Perhaps good marketing in the age of artificial intelligence requires less certainty and better questions:
What exactly was observed?
What happened immediately before it?
What evidence would show that our explanation is wrong?
Can the hypothesis be tested?
Once the business understands the behaviour more clearly, the final question is not simply how to use that knowledge, but how to use it in a way that creates value without undermining consumer trust.
Artificial intelligence can help businesses see more, identify patterns faster and predict behaviour with greater accuracy. But the real advantage may belong not to the company that knows the most about its customers, but to the one that understands what its data mean, what they cannot establish and how that knowledge should be used.
Source notes
[1] Pablo Neirz, Héctor Allende and Carolina Saavedra, “Epistemic frontiers and the distinction between causality, information, and predictability in pattern recognition”, Scientific Reports, 2026. The paper uses controlled simulations and explicitly distinguishes predictive success from causal or mechanistic understanding.
[2] Republic of Cyprus, Consumer Protection Law 112(I)/2021, as amended, Article 17(1)(ε1). The current consolidated text lists the disclosure of personalised pricing based on automated decision-making among the information that must, where applicable, be provided before conclusion of a distance or off-premises contract.
[3] Directive 2011/83/EU on consumer rights, Article 6(1)(ea), which contains the corresponding European requirement to inform the consumer, where applicable, that a price has been personalised on the basis of automated decision-making.
[4] General Data Protection Regulation (EU) 2016/679. Article 4(4) defines profiling; Article 21(2)–(3) covers the right to object to direct-marketing processing and related profiling; Article 22 concerns solely automated decisions producing legal or similarly significant effects, subject to its exceptions and safeguards.
[5] Deniz Lefkeli, Mustafa Karataş and Zeynep Gürhan-Canli, “Sharing information with AI (versus a human) impairs brand trust: The role of audience size inferences and sense of exploitation”, International Journal of Research in Marketing, Vol. 41, Issue 1, 2024, pp. 138–155. The research reports two pilot studies and nine controlled experiments involving 2,887 participants.