Originally published on Forbes
The future is ultimately unknown. There is no more coveted business knowledge than “What's going to happen?” But uncertainty cannot be eliminated, so all we can do is the next best thing: manage it.
Large-scale operations constitute a “game of numbers” involving millions of decisions. You can play this game more effectively by setting odds on individual outcomes. Which transactions should be blocked as potentially fraudulent? Which mechanical parts should be replaced before they break? Which customers should be discounted before they cancel?
input Predictive AI. Learn from the data to measure the probability of each case. There's no magic crystal ball, but there is a next-best solution. It is the probability that each individual is likely to click, make a purchase, lie, die, commit fraud, or become a bad debtor.
GenAI is not suitable. It's newer, sexier, and more advanced, but it doesn't replace predictive AI, it only enhances it. The two are destined to remain inherently different endeavors and disciplines, even as they blend into an integrated technology ecosystem. Here's why:
GenAI is not a replacement for predictive AI
There is always uncertainty. No matter how sophisticated algorithms become, even those involving large language models with trillions of parameters, they generally cannot predict future outcomes with very high confidence. Rather, we can only place odds on the outcome. As algorithms become more sophisticated, predictions improve, but profits decrease. Ultimately, we face limits to how accurately we can predict the behavior of people, businesses, machines, and other types of artifacts.
to Manage uncertainty with predictive AIcompanies must follow A very specific end-to-end paradigm. These projects are “predictive” in nature. I want case-wise odds, not genAI. Their function is to drive many decisions by estimating the odds of a particular case. This requires highly customized predictive AI projects. The main decisions are: three things: 1) What to expect; 2) Indeed must be predicted, and 3) How to use predictions To drive decision making.
GenAI is not well-suited for predictions at such a detailed case-by-case level. Razi Raziuddin, CEO of FeatureByte, who uses genAI to improve predictive AI, is helping the world understand this limitation. “LLM and other genAI models bring value to many business problems,” he told me. “But they are not designed to analyze large-scale tabular data, much less run machine learning algorithms on such data.”
GenAI is built using ML, not for ML. Although genAI is built with more advanced ML methods than those typically used for predictive AI, it does not itself constitute the same kind of “prediction machine.” At the heart of LLM is an ML model that predicts the next word (token) in a sentence. As such, it works well with human language and provides some degree of “reasoning” (according to some definitions). However, using LLM to perform well-defined data analysis tasks, including ML itself, is usually an inelegant overkill and, in fact, usually ineffective.
GenAI cannot run predictive AI projects on its own. To use predictive AI, organizations must explicitly interact with predictive capabilities in three ways: 1) Train an ML model for your immediate prediction goal; 2) Evaluate the model with respect to your prediction goal. The value of business improvement 3) Operationalize the model and use it to predict individual cases and drive decisions accordingly. LLM is not suitable for any of these three stages unless specific modifications are made.
But GenAI helps with predictive AI
Although GenAI does not perform the core analytics of predictive AI; Supports predictive AI project of in various ways. After all, genAI can code, design, and explain. GenAI has been applied to fulfill the following roles: The outspoken co-pilot it unravels, and it Explain how ML models make decisionsand as Predictive AI coding assistantand Prediction function generator.
These developments will integrate predictive AI and genAI within an emerging integrated ecosystem. This works by incorporating predictive AI capabilities within the genAI system. This allows users, for example, to ask conversational AI which customers are at risk of defection and how best to design targeted marketing campaigns to retain them.
The integration of predictive and generative AI has yet to receive its due recognition. said Justin Swansburg, former vice president of DataRobot. point out “There's a lot of opportunity, but so far I don't think it's gotten as much attention as it deserves from an engineering context, output description, and integration into workflows.” [and] Incorporating predictive models as tools. ”
Predictive AI will always have a role in this world, as uncertainty is an indelible aspect of life and business. Only by explicitly incorporating predictive AI capabilities can GenAI systems achieve state-of-the-art capabilities for managing uncertainty. This gives genAI access to a well-established, well-structured paradigm for generating and acting on detailed predictions. When integrated in this way, genAI serves to support and enhance predictive AI.
To learn more about these AI combinations, join me for my presentation, “7 Ways to Hybridize Predictive and Generative AI,” at a free online event. IBM Z day (Streamed live on November 12, 2025 and available on demand thereafter). If your work involves hybrid AI, consider submitting a proposal. I will be speaking at Machine Learning Week 2026.
About the author
Eric Siegel is a leading consultant helping companies implement machine learning and a former professor at Columbia University. He is the founder of the long-running Machine Learning Week conference series, the instructor of the highly acclaimed online course Machine Learning Leadership and Practice – End-to-End Mastery, editor-in-chief of The Machine Learning Times, and frequent keynote speaker. He wrote the best-selling book Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, and Die, which is used in hundreds of college courses. He also wrote The AI Handbook: Mastering the Rare Art of Machine Learning Deployment. Eric's interdisciplinary efforts bridge the stubborn technology-business gap. At Columbia University, he received an Outstanding Faculty Award for teaching a graduate computer science course in ML and AI. He later served as a business school professor at UVA Darden. Eric has also published analytical and social justice editorials. You can follow him on LinkedIn
