What marketing wants from data
Marketing is the work of getting the right message to the right people at a useful moment. Before digital tools, that meant guesswork supported by surveys and broad categories such as a newspaper’s readership. Today, advertisers have much more detailed information about behavior, and software that can act on it. AI’s role is to find patterns in that information and turn them into decisions: who to show an ad to, which message to use, when to send it, and what to offer.
Where the data comes from
Audience data typically falls into a few groups.
- First-party data. What a business collects directly: purchases, sign-ups, site visits, app activity and email responses.
- Context. The page, topic, time of day or device in use when an ad appears.
- Third-party data. Information assembled by other companies, such as data brokers, about interests or demographics. The practices around cookies, trackers and web tracking are a major part of this story, and rules on them differ by country and have been changing.
The mix used in practice varies across companies and over time as browsers, platforms and laws evolve.
What AI does with it
Segmentation. Instead of dividing customers by a few fixed traits, models can group them by behavior, such as people who browse often but rarely buy, or those who buy only during sales. Each group may then get different messages.
Prediction. Models estimate the likelihood that a given person will click, buy, subscribe or stop using a service. Marketers might use these estimates to decide where to spend budget or to send a reminder to someone who seems likely to drift away.
Lookalike audiences. A company can ask a platform to find people who resemble its existing customers in some measurable way. The platform’s model does the matching.
Recommendations. Content, product and email suggestions often come from the same kind of system behind recommendation algorithms on social platforms and shops: compare what similar people did, and offer what they responded to.
Creative testing. Software can test many versions of a headline or image on small samples, then show the better-performing version more often. Generative AI is increasingly used to draft variations of copy and images, which people typically review before use.
Timing and bidding. Many online ads are sold through automated auctions that take place in a fraction of a second, with software deciding how much to bid for each chance to show an ad.
Why personalization can work, and where it disappoints
Relevant messages are, on average, less annoying and more useful than irrelevant ones. A person shopping for a tent may appreciate a reminder about the one they left in their cart. Businesses like personalization because it can raise response rates while reducing waste.
But there are limits worth understanding:
- Measuring effect is difficult. If an ad is shown to people who were already likely to buy, sales may rise without the ad having caused them. This is the old problem of correlation and causation, and marketers use controlled experiments to try to separate the two.
- Models can be wrong or creepy. A suggestion that seems to know too much can feel intrusive, even if it was built from ordinary data.
- Feedback loops. Showing people more of what they already like can narrow what they see.
- Data quality. Shared devices, outdated records or mistaken guesses can lead to irrelevant or embarrassing ads.
Privacy concerns
The central worry is how much is collected and how it is used. Questions people reasonably ask include:
- What information is gathered about me, and from where?
- Did I understand and agree to this?
- Who else receives it?
- How long is it kept?
- Can I see, correct or delete it?
Sensitive categories, such as health, finances, children or political views, tend to receive extra protection in many legal systems, and there is ongoing public debate about whether inferring such traits from behavior is acceptable. Regulations such as data protection laws in various regions give people certain rights over their data, though the details differ and change. Anyone wanting specifics should check the rules that apply where they live.
There is also a fairness dimension. Targeting can limit who sees offers for housing, jobs or credit, which some laws restrict. Related questions arise in hiring bias and AI ethics more generally.
What individuals can do
- Review privacy and ad settings in the apps and browsers you use; many offer controls to limit personalization.
- Be selective about permissions, such as location or contacts.
- Use browser features or tools that reduce tracking if that matters to you.
- Remember that a free service is often funded by advertising, so a trade-off is involved.
- Treat ads as persuasion rather than information, no matter how well matched they seem.
Basic habits for online privacy go a long way, and none require technical expertise.
The engagement question
“Engagement” in marketing language means attention and interaction: clicks, views, replies, time spent. It is easy to measure, which is both its strength and its weakness. High engagement is not the same as real benefit to the person engaging, and systems optimized only for engagement can encourage habits people later regret. Responsible marketers increasingly look at longer-term measures like satisfaction and repeat business instead.
Summary
AI makes marketing more targeted and more measurable, mainly by predicting who is likely to respond and by automating decisions that used to be manual. The same abilities create real privacy questions and a risk of overclaiming what personalization achieves. Understanding the basic mechanics helps readers see why they are shown particular ads and what choices they have.