AI in marketing

AI in marketing is a broad category that covers everything from genuine capability to well-funded marketing copy. The two look similar from a distance, and the distance is exactly where most tool vendors operate.

The use cases where AI earns its keep are specific, measurable, and grounded in data: predicting which customers are likely to churn, identifying which prospects resemble your best buyers, personalising content and offers at scale. The use cases where it gets oversold are vague, dependent on claims that are difficult to verify, and almost always described with words like "autonomous" and "intelligent" rather than with actual numbers.

This page separates them.

What AI in marketing actually means

AI in marketing refers to the use of machine learning and statistical models to automate decisions, personalise communications, or predict outcomes that would be too slow or too complex for manual analysis.

This is a narrower definition than most vendor descriptions allow. It excludes:

• Rule-based automation with "AI" branding that is just conditional logic
• Language models generating content in bulk without strategic direction
• Tools that claim to "understand" your customers without access to your customer data

It includes:

• Predictive models that estimate the probability of a customer action (purchase, churn, upgrade)
• Segmentation engines that identify customer groups based on observed behaviour patterns
• Personalisation systems that serve different content or offers based on individual-level signals
• Lookalike modelling that finds new prospects who resemble your best existing customers

The distinction is not academic. Tools in the first category produce impressive demos. Tools in the second category produce measurable results.

The AI marketing use cases that work

Predictive segmentation. Statistical models that forecast what individual customers are likely to do next (buy, lapse, respond to an offer) outperform static RFMRFMRecency, frequency, monetary value: a scoring approach that ranks customers by how recently they bought, how often, and how much they spend. A practical bridge from simple rule-based segments to scored segmentation.View in glossary segments when you have enough data to train them. The model does not replace the marketer's judgment; it improves the inputs the marketer acts on.

Personalisation at scale. AI enables one-to-one content and offer personalisation across large customer bases: product recommendations based on browsing and purchase history, send-time optimisation for email, dynamic pricing and offers based on predicted value. The quality of the personalisation depends on the quality of the underlying data. AI does not create insight from nothing.

Predictive customer lifetime valueCustomer lifetime value (LTV)The total gross profit you expect from a customer across the whole relationship, not a single order. It tells you how much you can afford to spend to acquire and keep them.View in glossary. Models that forecast which customers are likely to become high-value, or which high-value customers are at risk, are among the highest-ROI applications of AI in CRM. A business that knows which new customers to invest in heavily, and which existing customers to prioritise for retention, has a structural advantage over one that treats all customers as equivalent.

Lookalike modelling. Finding new prospects who share the characteristics of your best existing customers is one of the core applications of machine learning in acquisition marketing. The model quality depends on the seed: a lookalike built from your highest-value customer segment will outperform one built from your full customer list.

The use cases that get oversold

Autonomous strategy. AI tools that claim to generate marketing strategy, set objectives, or make budget allocation decisions without human direction are, at this stage of development, overreaching. Strategy requires business context that models do not have access to: competitive dynamics, commercial relationships, product roadmap, brand positioning. AI can inform strategic decisions; it cannot make them.

Creative without direction. AI-generated content at scale is genuinely useful for filling in the blanks of a well-structured system: product descriptions that follow a template, email subject line variants for A/B testing, copy adaptations across formats. It is not useful as a replacement for strategic creative direction. The output reflects the quality of the brief, not just the capability of the model.

Insight without data. AI tools that promise to surface customer insight without access to your actual customer data are using generic market data or language model inference as a proxy for your specific customers. Generic data produces generic output. The AI marketing applications that deliver results work on your data, not on a substitute for it.

How to evaluate AI marketing tools

The questions that separate genuine capability from vendor positioning:

What data does it run on? The answer should be specific: your transaction data, your CRM, your behavioural data. Vague answers ("industry benchmarks", "our proprietary model") are a warning sign.

What does the output look like? Ask to see the actual output, not the interface. A predictive churn model should produce a list of customers with churn probability scores. A segmentation tool should produce defined segments with distinguishing characteristics. If the output cannot be described concretely, it probably cannot be acted on.

How do you measure whether it works? Any AI marketing tool should have a clear measurement framework. Predictive models can be tested: make predictions, wait, compare predictions to outcomes. If the vendor cannot tell you how to validate the tool's output, the tool is not designed to be validated.