Predictive AI uses past data to estimate what's likely to happen next: which accounts are likely to buy, which customers are likely to leave, or which ad is likely to get a response. Generative AI produces new content, such as text, images or code, based on patterns in the material it was trained on. Predictive AI gives you a score or a forecast. Generative AI gives you a draft. They solve different marketing problems, and some work needs both.

Since generative AI tools became widely available, "AI in marketing" has often come to mean writing tools. Predictive AI has been part of marketing for much longer, in lead scoring, churn models, ad bidding and product recommendations. Knowing which kind you're dealing with tells you what it's good for, what data it needs and how to check it.

What Predictive AI Does in Marketing

Predictive AI learns the link between past inputs and known outcomes, then applies it to new cases. It's usually built with machine learning trained on historical data where the result is already known. Common marketing uses include:

• Lead and account scoring: ranking leads or accounts by how likely they are to buy.
• Churn prediction: flagging customers who are likely to cancel or not renew.
• Forecasting: estimating pipeline, demand or revenue for the coming months.
• Media buying: automated bidding systems predict which impression or click is most likely to lead to the result you've asked for.
• Recommendations: suggesting the next product, content or offer a customer is most likely to respond to.

Predictive AI needs data with known outcomes, enough examples to learn from, and a pattern that stays reasonably stable over time. That's harder in B2B than in consumer markets. For example, a business that closes 40 deals a year has only 40 examples of a won deal, and buying cycles can run for months, so a model learns slowly and may pick up patterns that don't hold.

What Generative AI Does in Marketing

Generative AI produces new material in response to a prompt. Large language models generate text by predicting likely next words, based on patterns in very large amounts of training text. Image and video models work in a similar way with pictures. Common marketing uses include:

• Drafting: first drafts of emails, ads, landing pages, posts and briefs.
• Variations: versions of a message for different segments, roles or formats.
• Summarising: condensing research, call transcripts, reviews or reports.
• Images and video: concepts, mock-ups and production assets.
• Answering questions: chat assistants on a website or inside a team's own tools.

Generative AI produces text that reads well, but it isn't checking whether the text is true. In Moffatt v Air Canada (2024), a website chatbot told a customer he could claim a bereavement fare after he'd travelled. The airline's own policy said otherwise. British Columbia's Civil Resolution Tribunal held the airline responsible for what its chatbot said, rejected the argument that the chatbot was responsible for its own actions, and ordered it to pay damages.

The Difference at a Glance

Where Each One Goes Wrong

A predictive model can be accurate and still point you at the wrong customers. Eva Ascarza tested this with retention campaigns in the Journal of Marketing Research (2018). Targeting customers by how much a retention offer would change their behaviour reduced churn more effectively than targeting customers by how likely they were to leave. Some high-risk customers leave whatever you do, and some low-risk customers stay anyway. The same applies to lead scoring: accounts most likely to buy may buy without your campaign, so a high score doesn't prove your marketing caused the sale.

Generative AI goes wrong in a different way. It produces wrong facts, invented sources and generic copy with the same confidence as good work. It's also a poor fit for prediction. Gartner rates generative AI as "hardly useful" as the main technique for prediction and forecasting, planning and decision intelligence, and points to non-generative machine learning for prediction and optimisation techniques for planning. So a chatbot isn't the right tool for forecasting next quarter's pipeline, even if it gives you a number.

How to Choose Between Them

Start with the decision you need to make, not the tool. Ask what output you need:

• A number, a ranking or a yes/no about who, what or when: predictive AI, if you have enough data with known outcomes.
• Words, images or a summary: generative AI, with a clear brief, good source material and a person checking the result.
• Both: for example, a predictive model ranks accounts showing signs of buying interest, generative AI drafts outreach for each role on the buying committee, and a marketer checks every message before it goes out.

Gartner also suggests combining generative AI with other techniques where it helps. The order matters: decide who to reach and why with evidence, then use generative AI to help produce what they'll see.

FAQ

Is ChatGPT predictive AI or generative AI? Generative AI. It produces text by predicting likely next words, but it isn't built to forecast business outcomes from your data. Lead scoring, churn models and demand forecasts use predictive models trained on your own results.

Is predictive AI the same as machine learning? Predictive AI is usually built with machine learning trained on past outcomes. Machine learning also underpins generative AI, so the useful distinction is what each one produces: a prediction or new content.

Can generative AI forecast sales or pipeline? It can produce a forecast-shaped answer, but Gartner rates generative AI as hardly useful as the main technique for prediction and forecasting. Use statistical or machine learning models built on your own data, and use generative AI to help explain the results.

Which should a B2B marketing team start with? Whichever matches the problem. Generative AI is quicker to try because it doesn't need your historical data. Predictive AI is worth it when you have enough deals or customers with known outcomes, and a decision that a better ranking or forecast would change.

Start With the Decision, Then Pick the Tool

Predictive AI estimates what's likely. Generative AI creates something new. Neither one decides who your customers are, what you should offer them or why they should choose you. B2B Marketing Fundamentals covers the strategy decisions that come before any tool or campaign, and ends with a one-page growth strategy for your own business.

Sources

• Leinar Ramos, "When Not to Use Generative AI", Gartner. Source of the rating of generative AI as hardly useful as the main technique for prediction and forecasting, planning and decision intelligence, and of the recommendation to use non-generative machine learning, optimisation or simulation instead, or to combine techniques. gartner.com
• Eva Ascarza, "Retention Futility: Targeting High-Risk Customers Might Be Ineffective", Journal of Marketing Research, 2018. Source of the finding that targeting retention by expected response reduces churn more than targeting by risk of leaving.
• Moffatt v Air Canada, 2024 BCCRT 149, British Columbia Civil Resolution Tribunal. Source of the chatbot liability example. civilresolutionbc.ca