4 hours ago
3 Predictions For Predictive AI In 2025

Originally published in Forbes

GenAI’s complementary sibling, predictive AI, makes fewer headlines these days, but its vital role in achieving operational gains only continues to expand. Moreover, many thought leaders and AI evangelists contend that predictive AI delivers greater returns than genAI.

Here are three ways predictive AI will advance in 2025.

1) The emergence of hybrid predictive AI / genAI systems

Predictive AI and genAI need one another. Many capabilities can only materialize by combining the two in innovative ways.

For one, predictive AI has the potential to do what might otherwise be impossible: Realize genAI’s bold, ambitious promise of autonomy—or at least a great deal of that often overzealous promise. By predicting which cases require a human in the loop, genAI systems could gain the trust needed for them to be unleashed.

For example, consider a marketing campaign using genAI to personalize content for each customer. The content might be 95% reliable, meaning that 5% of customers would receive false or otherwise problematic content. This could make the project as a whole unviable, resulting in it simply never deploying. But if predictive AI flags for human review the, say, 15% of cases most likely to be problematic, this might decrease the rate of problematic content reaching customers to an acceptable 1%, while achieving 85% genAI autonomy. That is, for 85% of cases, genAI could do its thing with no human in the loop. In light of genAI’s audacious buildup, enterprises will soon begin turning to this approach as a necessary means to achieve a substantial piece of genAI’s promise.

In research, this use of predictive AI is called hallucination detection, but it hasn’t yet made a dent outside the lab. Expect to see a few startups soon.

There are several other ways in which marriages between predictive and generative AI are beginning to form, including:

• GenAI helps data scientists code predictive AI projects.
• GenAI chatbots help explain predictive AI to stakeholders. When you show an LLM the evaluation and valuation of predictive AI—such as by way of a profit curve—and prompt it with background information such as this article about how predictive AI works, it responds well to various queries and provides accessible explanations, as I’ll cover in a future article.
• GenAI models are used to predict.

For more, see the 2025 edition of the machine learning conference I chair, which includes a track devoted to predictive/generative AI hybrids.

2) The emergence of “ML valuation”—to ensure predictive AI captures business value

Whether combined with genAI or not, predictive AI projects must be valuated: The estimated business value they would deliver in deployment must be calculated. But to date, they usually are not valuated.

Imagine you’re developing a rocket. If you don’t stress test it—in its intended usage (via simulations, wind tunnels, etc.)—then its launch will be a shot in the dark, if not entirely scrubbed by sensible decision-makers.

That’s the status of most enterprise ML projects, aka predictive AI. The industry hasn’t matured to the point where such pre-launch stress testing is commonplace. As a result, most predictive AI deployments are scrubbed. Predictive AI’s great potential remains intact and many projects achieve it—but many more do not.

For business applications of predictive AI, this means forecasting business metrics like profit and savings. These estimate the business value that will be realized by deploying a predictive model to improve operations. Most projects only calculate technical metrics—precision, recall, AUC, F-score and other arcane measures—that reveal little to nothing about the potential business value. That is, they evaluate but don’t valuate. Accordingly, most projects fail, which brings us to…

3) The wide-scale adoption of bizML, the playbook for AI initiatives

Outside Big Tech and a handful of other leading companies, predictive AI initiatives routinely fail to deploy, never realizing value. What’s missing? A specialized business practice suitable for wide adoption.

Last year, I presented bizML, the gold-standard, six-step practice for ushering predictive AI projects from conception to deployment. This disciplined approach serves both sides: It empowers business professionals and establishes a sorely needed strategic framework for data professionals.

Predictive AI is a lot older than genAI—yet it’s still immature as an industry. The core number crunching is sound, but the divide between tech and business, including the lack of valuation in terms of business metrics, usually deters operationalization. The right methodology is out there, but most business professionals are not aware that a specialized practice is even needed in the first place, much less privy to any particular formalization. That’s why I branded the industry’s best practices with a hopefully catchy buzzword, bizML.

This year, expect to see a jump in the rate of successful predictive AI deployments as organizations adopt these best practices, bringing together business and data professionals to collaborate deeply across each project, end to end. In a nutshell, bizML ensures that the team jointly establishes three things: 1) what ML is called upon to predict, 2) how well it predicts—and how valuable those predictions will be—and 3) how the predictions are acted upon to improve operations.

It will be an exciting year as these developments bring predictive AI back into the spotlight and further amplify its value. These advances represent the much-needed professionalization of predictive AI as a field.

About the author
Eric Siegel is a leading consultant and former Columbia University professor who helps companies deploy machine learning. He is the founder of the long-running Machine Learning Week conference series, the instructor of the acclaimed online course “Machine Learning Leadership and Practice – End-to-End Mastery,” executive editor of The Machine Learning Times and a frequent keynote speaker. He wrote the bestselling Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die, which has been used in courses at hundreds of universities, as well as The AI Playbook: Mastering the Rare Art of Machine Learning Deployment. Eric’s interdisciplinary work bridges the stubborn technology/business gap. At Columbia, he won the Distinguished Faculty award when teaching the graduate computer science courses in ML and AI. Later, he served as a business school professor at UVA Darden. Eric also publishes op-eds on analytics and social justice. You can follow him on LinkedIn.