Predictive AI still deserves a seat at the budget table

Two kinds of AI, one budget

Generative AI produces text, images, code and dialogue. Predictive AI works with probabilities, assessing likely outcomes for a specific customer, patient, transaction or piece of equipment.

They are not substitutes. A camera and a phone do not compete for exactly the same job; predictive and generative systems should not be forced into the same category either. More mature organizations combine them, using one to strengthen the other.

The funding market has not made that distinction easy to see. Generative AI benefits from novelty: it looks more like a science-fiction intelligence than most earlier software. Research money and venture capital have followed it at a scale predictive AI has never seen.

Inside corporate data teams, however, the picture is less one-sided. The author, who organizes industry conferences, sees applications for predictive and generative AI arriving in roughly equal numbers—about 50-50.

The predictive projects include:

That list is not a fallback category. It is a map of expensive, recurring decisions where even a modest improvement can matter.

The reliability layer

Predictive AI remains useful because uncertainty remains. Companies will always need to estimate which borrower may default, which customer may stop using a service and which component may fail first. Those are numerical decisions at scale—the environment predictive systems are built for.

A project can use a large language model to improve its accuracy and still be fundamentally predictive AI. The model's presence does not change the underlying job.

Generative systems bring a different risk. Large language models can invent information and behave unpredictably, which complicates their use as customer-support operators, analysts, tutors and virtual assistants.

Predictive AI can serve as a control layer around those systems. It can flag interactions with a language model that are most likely to go wrong and send them to a human for review. Companies are already adopting this hybrid structure, and the author expects it to become a major source of growth for predictive AI.

As pressure builds to make AI-agent projects ready for deployment, that role may become particularly valuable. Many generative AI systems may ultimately depend on predictive models as a reliability barrier: a person stays in the loop, but intervenes only when the system identifies a case that genuinely needs judgment.

The missing investment is implementation

I think the market's main mistake is not overestimating what predictive models can calculate. It is underestimating what organizations must change to use their output.

Predictive AI has been used for decades, yet a significant share of initiatives never reaches production. The hard part is often not building the model. It is persuading an organization to act on probabilities instead of waiting for certainty.

That makes predictive AI harder to sell to department leaders who approve projects but still struggle to understand how the systems work. The more interesting question is therefore not which branch of AI attracts more attention. It is whether a company can redesign daily operations around a model's assessment.

The book The AI Playbook, published in 2024, is positioned as a response to that implementation gap. It gives business readers accessible semi-technical knowledge for working productively on predictive AI projects and describes a collaborative process from initial idea to deployment.

Its reception suggests that the subject has an audience beyond data science teams:

The quiet failure in the current AI debate is that attention is being treated as evidence of value. Predictive AI has fewer spectacular demos, but its projects are tied to operational decisions that companies already make every day. Its constraint is not relevance; it is getting the organization to trust and use the result.

That leaves a practical tension. Generative AI may attract the budget because its capabilities are easy to demonstrate, while predictive AI may deliver more dependable gains once it is embedded in the work. The companies that resolve that tension will not choose the more fashionable model—they will fund the decision system their operations can actually absorb.