When Prediction Becomes Cheap, Everything Becomes a Service

Jun 04, 2026

10 min read

4 views

88%

The strange economics of almost free prediction

What happens when a capability that once required scarce human expertise becomes nearly free?

That is not a hypothetical question anymore. It is the quiet engine behind the next wave of AI disruption. A task once limited by the number of skilled people who could do it, detect patterns, classify signals, or anticipate needs is turning into a cheap, abundant utility. In economic terms, prediction is becoming a commodity. And once prediction gets cheap, whole categories of services can be rebuilt as software.

This is why the opportunity is so large. If software can increasingly perform the predictive work that used to sit inside labor, then the frontier is no longer just replacing isolated tasks. It is converting entire service industries into products, platforms, and automated workflows. That shift is why estimates for services as software reach into the trillions of dollars. The size of the prize is not a mystery once you see the mechanism: the world is full of services whose core value is not the hands that deliver them, but the predictions that guide them.

The deeper surprise is that this is not only a story about cost reduction. It is a story about ubiquity. When prediction becomes cheap, it does not merely make old services cheaper. It makes predictive capability omnipresent, woven into every app, every workflow, every customer interaction, every decision layer. Cheap prediction does what cheap storage and cheap computation did before it: it changes the default shape of the product.

Prediction is not a feature, it is an input

For a long time, prediction was treated like an invisible talent. It looked like intuition, experience, judgment, or expert service. A doctor predicts what test to order next. A salesperson predicts which lead is most likely to convert. A logistics manager predicts where bottlenecks will appear. A credit officer predicts who will repay a loan. These are all forms of pattern recognition applied to incomplete information.

The important shift is to stop treating prediction as a mystical human trait and start treating it as an economic good, something with supply, demand, and cost. Once you do that, the logic becomes clearer. If prediction is an input, then every process that depends on better forecasts can be redesigned around a cheaper version of that input.

Think of steel. When steel got cheaper and more standardized, it did not just make bridges cheaper. It changed what could be built at all. Cheap steel enabled skyscrapers, railroads, ships, and machines that would have been uneconomic before. Prediction is entering the same phase. As the cost drops toward zero, the question is not, “How do we use AI to improve this existing service?” The real question is, “Which services only become viable when prediction is abundant?”

That distinction matters. Many people imagine AI as a clever assistant bolted onto old workflows. But cheap prediction is more radical than assistance. It is an infrastructure layer. It turns previously manual judgment into a reusable primitive. Once that primitive is available everywhere, product design changes from the inside out.

The most important consequence of cheap prediction is not automation. It is recomposition.

Why services become software before industries collapse

When people hear that AI will transform services, they often picture disruption as a clean swap: software replaces a worker. But the transition is messier and more interesting. In many cases, services do not disappear first. They are decomposed.

A service is usually a bundle of at least four things: prediction, coordination, interpretation, and trust. AI is strongest at the first two. It can estimate, classify, route, rank, recommend, and generate. That means the part of the service that was once most expensive, the judgment layer, becomes cheap and scalable. Once that happens, the rest of the service can be reassembled around software.

Consider a few examples:

A call center is not really a building full of people. It is a prediction system for matching customer intent to response.
A junior analyst is not merely producing spreadsheets. They are forecasting, screening, and pattern matching under uncertainty.
A claims processor is not just filling forms. They are estimating legitimacy, prioritizing cases, and identifying exceptions.
A tutor is not just speaking knowledge. They are predicting confusion and adapting instruction in real time.

Each of these can begin as a hybrid, where software handles the routine predictive layer and humans handle edge cases. But as prediction improves, the human role shrinks toward exceptions, escalation, and relationship management. The service quietly becomes software first, with humans retained only where ambiguity, accountability, or emotional nuance still matter.

This is why the estimate of a multi trillion dollar opportunity is plausible. The opportunity is not confined to one market. It spans every sector in which expertise is used to make forecasts that can be digitized. Finance, healthcare, education, insurance, legal work, operations, logistics, customer service, procurement, recruiting, and sales all contain vast amounts of embedded prediction.

The trap is to think in terms of jobs. The better lens is to think in terms of prediction density. The more a service depends on repeated, scalable judgment under uncertainty, the more exposed it is to software transformation.

The hidden shift from scarcity to ubiquity

Cheap prediction has a second effect that is easy to miss: it changes the economics of personalization.

When prediction was expensive, it had to be rationed. You could personalize a medical diagnosis only if a specialist was available. You could tailor a financial recommendation only if a human advisor had time. You could adjust an interface only if a team had enough data scientists. Scarcity meant prediction was reserved for high value cases.

Now prediction can be embedded everywhere. Every website can infer intent. Every app can adapt. Every workflow can route cases based on risk, urgency, or likely outcome. That is what is meant by ubiquity. The cost curve does not just improve margins. It changes the threshold at which prediction gets used.

This creates a powerful but underappreciated dynamic: lower cost expands the set of problems worth solving. The same pattern happened with cloud storage. Once storage was cheap enough, companies kept more data, experimented more, and built products around histories that would have been thrown away before. Cheap prediction does something similar. It makes it viable to forecast, rank, and customize at a granularity that used to be impractical.

A simple analogy helps. Imagine a restaurant where every table has a smart host who predicts what each guest wants before they ask. When prediction is scarce, only VIPs get that treatment. When prediction is cheap, the experience becomes standard for everyone. The restaurant is still a restaurant, but its economics, staffing, and customer expectations have changed completely. The value no longer comes from merely serving food. It comes from anticipating need.

This is the subtle but profound shift: software no longer just stores or displays information. It begins to anticipate behavior. When that happens, product quality becomes less about static functionality and more about adaptive intelligence.

The real moat is not prediction, it is what you do with it

If prediction becomes abundant, then prediction alone stops being the competitive advantage. That is the next strategic misunderstanding to avoid.

A world with cheap prediction does not mean all businesses become equal. It means the value migrates. Once the predictive layer is widely available, the differentiator becomes the system built around it: access to data, speed of feedback, trust, integration into workflows, and the ability to convert predictions into action.

This is the crucial mental model: prediction creates options, but execution captures value.

Think about lending. If you can predict default risk better than your competitors, you have an edge. But if everyone can access similar predictive models, the real moat shifts to underwriting discipline, customer acquisition, capital costs, and regulatory relationships. Or think about healthcare. Better prediction of patient risk is useful, but the winner may be the organization that can actually intervene earlier, coordinate care, and maintain trust.

This means the most durable companies in the AI era may not be the ones with the fanciest models. They may be the ones that can embed prediction into a complete operational loop. The loop looks like this:

Sense: collect signals from users, processes, or environments.
Predict: estimate what is likely to happen next.
Decide: choose an action based on that estimate.
Act: deliver the intervention, recommendation, or service.
Learn: observe the outcome and improve the loop.

The competitive advantage lies in shrinking the time between prediction and action. A model that predicts churn is useful. A system that predicts churn and automatically triggers the right retention offer at the right moment is much more valuable. In other words, prediction is necessary but not sufficient. The real business is the closed loop.

This is why services as software is such a powerful idea. It is not simply about digitizing labor. It is about turning expertise into a repeatable feedback system.

In the AI economy, the winners are not those who predict best in isolation. They are those who operationalize prediction fastest.

A practical framework: where software will eat services first

Not every service is equally vulnerable. Some rely heavily on empathy, negotiation, physical presence, or moral accountability. Others are fundamentally predictive and therefore more easily converted into software. A useful way to think about the transition is to map services along two axes: predictive content and exception cost.

High predictive content, low exception cost: These are the first to go software native. Examples include lead scoring, content ranking, simple underwriting, basic support routing, inventory forecasting, and routine document review.
High predictive content, high exception cost: These become hybrid systems. AI handles the common cases, humans handle the edges. Examples include medical triage, fraud detection, and legal intake.
Low predictive content, high relational value: These resist full automation longer. Examples include therapy, executive coaching, complex negotiation, and certain forms of care.
Low predictive content, low repeatability: These are messy and local, less amenable to software conversion in the near term.

The important insight is that the first wave of disruption will not target every human role equally. It will target the parts of roles that are easiest to standardize and predict. That is why companies should not ask whether a job is “safe” in the abstract. They should ask which part of the job is actually a prediction engine.

This also helps explain why many AI products feel deceptively small at first. A tool that merely improves classification or ranking may look incremental. But if it sits at the center of a service workflow, it can rewire the economics of the entire system. A better lead scorer changes sales. A better triage engine changes care delivery. A better matching layer changes marketplaces. In each case, the software is not just a feature. It is the new service architecture.

Key Takeaways

Treat prediction as an input, not a magic trick. If it can be priced, it can be redesigned into products and workflows.
Look for services with high prediction density. The more a service depends on repeated judgment under uncertainty, the more likely it is to become software.
Do not confuse cheap prediction with durable advantage. The moat moves to data, workflow integration, feedback loops, and trust.
Design for closed loops. The real value comes when prediction is connected to action, not when it remains a standalone score.
Audit your own work for predictive tasks. Ask which parts of your role are ranking, classifying, forecasting, or routing. Those are the parts most likely to be transformed first.

The future belongs to systems that anticipate

The biggest change brought by AI may not be that machines become more intelligent in the human sense. It may be that anticipation becomes ambient. Prediction, once scarce and specialized, will be everywhere, quietly shaping interfaces, workflows, and decisions.

That has a profound implication. When prediction is cheap, value shifts away from merely answering questions and toward organizing action around likely futures. A company is no longer just a provider of services. It becomes a system for sensing what is coming, deciding what matters, and responding before the need is fully visible.

That is why this moment is bigger than automation. Automation removes labor from tasks. Cheap prediction changes the structure of services themselves. It turns expertise into software, uncertainty into a design problem, and responsiveness into the new premium experience.

The deepest reframe is this: the future of AI is not about replacing what humans do best. It is about making prediction so abundant that the economy reorganizes around it. Once that happens, the real question is no longer who can predict. It is who can build the most useful world around prediction.