What type of AI should companies focus on?Generative AI that generates text, computer code, images, video, or other content; or Is it predictive AI for advertising, marketing, fraud detection, risk management, medical treatment, and other established operations?
People who ask this question may be missing the point. It’s like when you’re trying to decide which of two people to marry. If it falls within an either/or box, that means neither option is clearly enforceable yet.
Rather than selecting an attractive technology and then looking for a problem (also known as solutionism), industry leaders can do so by first identifying the key problem and then finding the best way to solve it. My advice is to start with your value proposition. “AI strategies fail because AI is a means, not an end,” said Mifnea Moldoveanu, a business professor at the University of Toronto. This makes as much sense as asking, “Do you have an Excel strategy?” ”
Instead, ask yourself where are the best opportunities to improve your operations. And how can technology help you pursue that opportunity and secure operational wins?
Predictive and generative AI
Generative AI and predictive AI serve completely different purposes. Generative AI is intended to perform tasks currently handled by humans. In contrast, predictive AI pursues less ambitious but often more significant goals. It’s about streamlining a company’s biggest operations, the very processes that have already evolved to become codified. So, while generative AI may seem more impressive and interesting, predictive AI often delivers significant bottom-line improvements in enterprise efficiency.
Many companies would benefit from redirecting generative AI’s disproportionate attention to predictive AI.
Predictive AI (also known as predictive analytics or enterprise machine learning) improves the performance of almost any type of existing large-scale operations across departments, including marketing, manufacturing, fraud prevention, risk management, and supply chain optimization. It is the technology that companies use to . Learn from data to predict outcomes and actions, such as who clicks, buys, lies or dies, which vehicles need maintenance, and which transactions turn out to be fraudulent. Masu. These predictions determine who to call, mail, approve, test, diagnose, warn, investigate, jail, date, and treat in millions of cases every day. Drive business decisions.
Predictive AI has three main advantages over generative AI.
1. Predictive AI often provides higher returns than generative AI. Predictive AI improves a company’s largest processes and therefore has the potential to have the biggest impact on company efficiency.
A mature organization is one that streamlines key activities and establishes them as mechanistic, systematic processes made up of many individual decisions. Given the homogeneity, such processes are ripe for predictive optimization.
Therefore, predictive AI brings high returns and improves customer experience. UPS saves an estimated $35 million annually by predicting tomorrow’s deliveries and optimizing package delivery plans. A midsize bank could save $16 million annually by predicting which payment card transactions are fraudulent. In a marketing campaign, he can increase profits five times by predicting which customers will buy.
This is the original AI and one of the established enterprise use cases for machine learning with decades of proven results.
Although it is old, predictive AI is not. Most of the money is still there. The predictive AI market is predicted to reach $64 billion in 2025, but spending on generative AI is still less than 7% of last year’s forecast AI spending, and other estimates put it at less than 4%. It is estimated that. However, the potential of predictive AI is still largely untapped. One reason for this is that companies are still struggling with how to effectively move from development to operations. There are many opportunities.
2. Predictive AI can operate autonomously, whereas generative AI typically cannot. In many deployments, predictive AI can reliably drive decisions without human intervention, while generative AI supports human processes rather than automating them.
Enterprise use cases for generative AI, such as helping create marketing creative or improving the efficiency of code, require that each output (all assertions, suggestions, inferences, statements, segments of computer code, and draft documents) You need a human in the review loop. Generate. Generative AI is in a position to take on consequential tasks by humans, activities that attract the eye of surveillance, as computers require high levels of performance to operate without continuous human supervision. If a machine writes something that would normally be written by a human, you cannot blindly trust it.
In contrast, many applications of predictive AI can leverage the immense value of full autonomy by adopting more permissive features. The banking system will instantly decide whether to authorize the charge to your credit card. Websites make instantaneous decisions about which ads to show, and marketing systems make a million yes/no decisions about who to contact. The same goes for political campaign analysis systems. E-commerce sets a price for each purchase, from airline tickets to flashlights. The safety system determines which bridges, manholes, and restaurants to inspect. No humans are involved in these specific decision-making steps.
3. Predictive AI is much cheaper and has a much smaller footprint than generative AI. Machine learning models needed for predictive AI projects are typically orders of magnitude lighter than models for generative AI.
Large language models (generative AI models that generate text and code) typically consist of 100 billion to 1 trillion parameters and are often trained over billions of pages. In contrast, predictive AI models often consist of only tens to hundreds of parameters (rarely more than a few thousand) and are typically trained with fewer than 100,000 learning cases. You can train on your laptop.
This difference stems from a fundamental difference in purpose. Generative AI strives to create content that relies on as deep an “understanding” of human concepts as can be squeezed out of a machine. In contrast, predictive AI seeks to predict inherently unpredictable outcomes such as human behavior, such as who will click, buy, lie, or die. No matter how complex a model is, there is an upper limit to how accurately it can predict such things. You generally cannot expect to achieve a “crystal ball” level of confidence for such predictive goals. Instead, predictive AI provides predictions that are better than guessing, which is usually more than enough to make a meaningful improvement to your bottom line. Relatively lightweight models are usually suitable for predictive AI projects, as you face diminishing performance benefits as the complexity of the predictive model increases.
Do more with predictive AI as you deploy generative AI
It is no surprise that the glamor and novelty of generative AI is attracting attention. Never before have we seen a computer behave so humanly. However, the focus on generative does not preclude greater adoption of predictive.
On the other hand, generative AI isn’t all hot air. Having a computer generate the first draft of your text or code can be truly amazing and worthwhile, as long as you recognize the need for humans to review each draft. On the other hand, I don’t think we’ll see AI that can fully automate jobs anytime soon. It is just a tool, not a replacement for humans.
The commonality between generative and predictive AI—the core principles of machine learning—is buried deep within. Therefore, this is not a binary competition or a zero-sum game. They shouldn’t compete like ski resorts and water parks. Ideally, organizations should approach each operational problem using the most appropriate technology to create the most value.
follow me twitter Or LinkedIn. check out My website and other works can be found here.
