Don’t Let GenAI’s Glitz Distract From Predictive AI’s Low-Hanging Fruit

Two Different Jobs, One Shared Stage

GenAI creates: text, images, code, conversation. Predictive AI plays the odds: it learns from historical data to estimate the outcome for each individual customer, patient, transaction, or piece of equipment. These are fundamentally different jobs. A camera and a telephone don’t compete for the same task, and neither should these two flavors of AI. In fact, the more advanced organizations are learning to combine them, using each to reinforce the other.

And yet, they predictive AI and genAI do compete – for budget, headcounts, and executive attention. In that contest, genAI’s charm gives it an outsized advantage. It feels closer to science-fiction intelligence than anything that’s come before, and that readily sucks in research dollars and venture capital at a rate predictive AI never enjoyed. The word “AI” itself has narrowed in the public understanding to mean genAI alone. That shift is understandable. It’s also a mistake. My advice to most companies: Invest at least as much in predictive AI as you do in genAI. I know that sounds almost heretical amid today’s frenzy. But predictive AI often represents more accessible untapped opportunities. If you prefer plucking the lowest hanging fruit, that’s often a matter of using predictive AI to improve existing large-scale operations.

What’s Actually Happening Inside The Industry

The press paints a lopsided picture, but the view from inside enterprise data science teams tells a different story. In my work organizing industry conferences, I see close to a 50-50 split between predictive AI and genAI project submissions – not the landslide you’d expect from reading the headlines. Moreover, the eternally-wide range of predictive AI use cases crossing my desk hasn’t narrowed one bit: forecasting electrical grid failures, catching fraudulent insurance claims, flagging likely lease terminations, spotting soiled solar panels before they lose efficiency, predicting which patients will skip appointments, and identifying shoppers about to abandon their carts.

Talk to practitioners directly and you’ll hear the same thing. Data science teams often feel pressure from leadership to chase genAI’s promised revolutionary overhaul of, well, everything. But they privately report that their predictive AI projects are the ones more often moving the needle.

The demand for my book about deploying predictive AI, The AI Playbook, also reflects its enduring importance. Since its 2024 publication, it has received endorsements from Scott Galloway, Charles Duhigg, Mustafa Suleyman, and the CEO of FICO; won multiple awards; become a number-one Amazon category bestseller; been adopted for courses at universities such as Carnegie Mellon, George Mason, New York University, Northeastern, and University of Texas; appeared on several “must-read” lists; led to dozens of keynote addresses; and gained media appearances in the likes of Bloomberg Radio, Fast Company, Harvard Business Review, Los Angeles Times, and over 60 podcasts. (By the way, you can pre-order the paperback edition of The AI Playbook now, and receive free, immediate access to the audiobook.)

Predictive AI Ain’t Going Anywhere – Here’s Why

Predictive AI endures because uncertainty is a certainty. Businesses will always need to make better bets across large numbers of cases – which loan applicant will default, which customer will churn, which part will break down first. That’s a numbers game, played at scale, and it’s precisely the kind of problem predictive AI solves. While genAI is built using machine learning, it is not built to natively do what machine learning was designed for: prediction. Even when a project leverages a large language model to sharpen its predictions, it’s still fundamentally a predictive AI project.

And there’s a twist that makes predictive AI’s future even more secure: GenAI’s own success depends on predictive AI. Large language models hallucinate. They behave unpredictably, which challenges their more ambitious would-be deployments – as customer service agents, analysts, tutors, and virtual assistants. Predictive AI is rapidly emerging as the fix, serving as a reliability layer that flags a language model’s interactions that are most likely to go wrong, and directing human review accordingly. I’m seeing enterprises adopt this hybrid pattern, and I expect it to become p next major growth engine for predictive AI. Given the unprecedented pressure to make “agentic” projects deployment-worthy, this makes for a tremendous killer app. Before long, most genAI deployments will depend on predictive AI as a reliability guardrail that keeps humans in the loop, yet uses those more expensive humans judiciously.

Predictive AI’s Real Obstacle Isn’t Hype – It’s Deployment

If predictive AI has a genuine weakness, it isn’t a lack of relevance. It’s project execution. Despite decades of use, a striking share of predictive AI initiatives never make it into production. The technical modeling usually isn’t the hard part; winning organizational buy-in to actually operationalize is. Deployment means changing how a business operates based on odds rather than certainties – and that’s a harder sell than it might be, largely because it’s still poorly understood by the business stakeholders who have to approve it.

That’s gap in understanding matters the most right now – not a battle for supremacy between two kinds of AI project, but the practical challenge of getting predictive models out of the lab and into daily operations. My book The AI Playbook bridges that gap. It ramps up the business reader on the relevant, accessible semi-technical know-how needed for deep collaboration on predictive AI initiatives, and it presents the gold-standard collaborative practice for ushering those initiatives from conception to deployment.

GenAI earned its spotlight. But predictive AI is still doing the work of making organizations run better, one prediction at a time. Rather than allowing trendiness and glitz to run the show, make informed, practical choices about which kind of AI projects best serve, which ones represent the most easily accessible wins. It will almost certainly turn out to be predictive AI more often than today’s headlines advertise.