Most generative AI conversations still revolve around producing marketing faster—emails, subject lines, image variations, campaign versions. That’s useful, but it doesn’t necessarily make marketing smarter. The real opportunity is using AI to explore customer data and find relationships we haven’t thought to ask about.
Target’s famous pregnancy-prediction work is the classic example. Its statistician Andrew Pole analysed transactions and identified about 25 products that, together, could predict whether a customer was pregnant and estimate a due date. The work, later told by Charles Duhigg in The New York Times Magazine in 2012, showed the power of looking at patterns instead of individual purchases.
At the time, that required specialist expertise and heavy technical resources. Today, marketers without a data science background can use tools like ChatGPT or Claude to explore datasets, compare customer groups, build visualisations and interrogate findings. For a D2C brand in India or a global retailer, the technical barrier is falling. The judgement barrier is not.
The 92.9% figure is descriptive, not causal
An independent analysis of the University of California, Irvine’s Online Retail dataset—more than 540,000 transaction records from a UK retailer between December 2010 and December 2011—found that repeat customers made up 65.6% of customers but generated 92.9% of revenue.
That feels like a clear mandate to spend more on retention. But the stat only describes how revenue was distributed. It doesn’t explain why customers returned, whether a particular retention programme drove loyalty, or which customer experiences marketing can realistically influence. Repeat buyers buy more by definition, so their higher revenue share is partly structural, not a proven strategy.
This is the trap AI-assisted analysis makes easier to fall into: an accurate figure can still produce a misleading conclusion if we treat correlation as explanation.
Pattern, hypothesis, test
AI can surface associations, suggest competing explanations and write analytical code. Behavioural science can help marketers form plausible explanations. Disciplined experimentation can then show whether a change actually shifts the behaviour.
- Use AI to ask different questions: what distinguishes returned customers from one-time buyers?
- Turn each discovered relationship into a behavioural hypothesis: are they more engaged, or does discovery itself build confidence?
- Run a small test before you scale a campaign or automation.
What this means for marketing teams
The edge is moving from finding a correlation to deciding whether it matters. As analysis gets cheaper, more teams will see interesting patterns. Those that challenge assumptions, interrogate findings and test before scaling will get more value from the same data. AI still misreads fields, miscalculates, ignores confounders and can confidently explain relationships that don’t exist—so validation and specialist input remain necessary. But the marketer’s new job is to pair pattern discovery with behavioural understanding and measurement.
Source: MarTech
