"AI" in marketing tools covers three different things. Knowing which one you are being sold is the first skill a marketing leader needs.
1. Predictive AI: scores, forecasts and optimisation
Models trained on past data estimate what will probably happen: which customer is likely to buy or churn, which audience will respond, how to split a media budget, what bid to place. Ad platforms have done this for years. It is only as good as the data it learns from, and it optimises for the goal you give it — which may not be the goal you meant.
2. Generative AI: drafts, images, video and summaries
Large language models and image or video models create new content from a prompt: copy, product images, video cuts, translations, research summaries. They are fast and cheap per piece, and they can be confidently wrong, off-brand or legally risky (claims, likeness, copyright). They don't know your brand unless you tell them.
3. Agentic AI: taking actions
An agent is an AI model connected to tools: it can build an audience, write variants, launch a test, read the results and shift budget, with little or no human step in between. Ad platforms and marketing clouds are moving fast in this direction. Agents multiply both value and risk, because mistakes go out at machine speed.
Where marketing teams use it today
Gartner's survey of 418 marketing leaders (July–September 2024) found 27% of CMOs reported limited or no generative AI adoption in campaigns; among adopters, 77% used it for creative development. Creative and content are where most teams start; strategy, measurement and agents come later.
A simple test for any AI pitch
- Which kind is it — predict, generate or act?
- Which marketing number should move — cost per asset, speed, conversion, share of AI answers — and against what baseline?
- What does a human check before it reaches a customer?
If a vendor can't answer the second question in your numbers, you are buying a demo, not results.