From Marble Run to Pinball: Why Visibility and AI Change the Game

AI Isn’t That New—But It Behaves Differently

AI is widely seen as a step change in how technology works. That framing makes sense, but it can lead to mistrust about what it can do and missteps in what it should do. In reality, much of what we call AI today is built upon systems, workflows, and data structures that organizations have been developing and refining for decades

Robotic process automation (RPA), workflow management, automations, and rules engines have long been used for years to streamline operations across manufacturing, automotive, and retail, among other industries.

What’s changed isn’t that systems suddenly became “smart.” It’s how all the innovation has come together to become one AI-driven, powerful force.

Looking back, traditional automation has followed a fixed path—each step defined in advance, each outcome predictable. In contrast, AI systems work differently. Instead of carrying out a pre-defined sequence, they collect data, evaluate multiple inputs, and present potential solutions.

That shift — from following rigid rules and executing each action just so to following a myriad of paths to potential solutions — is what separates traditional automation from AI.

The Marble Run: Systems That Follow a Fixed Path

I like to use a simple analogy to explain the way that smart technology like AI is evolving.

Traditional automation works like a marble run. You design the path from beginning to end; every turn, every drop, every outcome is predetermined. When you release the marble, it follows that path exactly as intended. You can’t really see what is happening inside the run, but as long as there aren’t any major issues or cracks, it delivers a very predictable outcome. The marble emerges successfully at the end, although you can’t trace the path of that success.

The path is invisible because automation has eliminated the manual processes along the way; AI completes very specific incremental tasks, whether guided by a human or completely “unassisted.” For example, surfacing a variance in a budget might generate an action and then another action, and so on.

Marble-run systems are everywhere. Consider inventory reconciliation that compares quantities across ERP, WMS, and on-hand records. Or order routing and fulfillment, where you might send orders to the nearest warehouse, the nearest warehouse with available inventory, or pre-defined shipping zones.

These processes can be effectively automated because they are structured and repeatable. The inputs are known, the rules are clear, and the output is consistent. The system executes exactly as directed, in exactly the way it was designed, eliminating the time a person needs to spend managing it.

The Pinball Machine: Agile Systems That Leverage AI

AI and smart technology behave differently, similar to a pinball machine. You define the inputs and rules that initiate the action. But once the ball is in play, the system has many decision points: Depending on the strength of the launch and the exact angle at which the ball hits a bell or a whistle, the system determines a different path forward.

A large assortment of information is gathered along the way to produce the most relevant outcome. Similarly to that, the pinball machine reacts to the conditions, interactions, and inputs in real time. Was the game board located in a humid environment? Is the temperature high or low? Is the person launching the ball an adult or a child?

The outcome isn’t fixed — it’s influenced by a range of factors. Nevertheless, the ball always returns to the sender, just like AI analyzing data and delivering recommendations.

Now, let’s look back at our previous examples of marble-run systems.

Inventory reconciliation may have clear inputs and rules, but inventory prediction requires much more sophisticated demand forecasting with a wide range of inputs, including historical sales, current inventory levels, lead times, and external factors such as industry trends, interest rates, and even weather patterns.

AI can handle the complexity that arises as each input potentially bounces the “pinball” in a different direction. While a “marble run” tool  can only deliver the information via one output, the “pinball machine”-style solution gathers and delivers information at every touchpoint; it also reports back to the user in real time, with the same second-by-second, glass-like visibility we associate with the pinball machine’s instant scoreboard and flashing lights.  

Last-mile delivery is another ideal pinball-machine situation: Route optimization and predictive arrival times require a host of constantly changing inputs. Modern technology that leverages AI can solve these problems — all while providing the visibility you need to see what’s going on at every step. You can’t do that with a marble run! In many cases, you might have multiple balls in the run at the same time and depending on how they interact with each other, the outcome might be different.

In all cases, the system does not follow a fixed set of steps. It is continuously evaluating multiple variables and determining the best possible outcome in the moment, reacting to external conditions as it goes. There is no single correct path; there are only better or worse solutions based on available information at a particular moment in time. The scalable nature of AI allows the system to continuously adjust as the pinball moves.

Pinball Machine Systems Create New Friction

There is a natural tendency to expect AI to behave like traditional automation. Teams want consistent outputs, repeatable results, and full control over the process. But AI is not designed to be static, and our education and curiosity needs to evolve along with it. As a society, we are all still learning what AI can do and where to use it best.

Because it is making decisions based on multiple variables, the same input may not always produce the same output. That variability is not a flaw; it is a fundamental characteristic of how these systems work.

Still, it creates friction. Teams question why results differ. Leaders hesitate to trust the outputs and are concerned about information security. The pricing team may expect the same inputs to always produce the same price recommendation, only to observe AI suggesting different options based on subtle changes in demand signals.

In some cases, organizations try to force AI into rigid, fixed-input workflows where it doesn’t belong. Returns processing is an example where retailers expect the same output every time, rather than a more realistic, probabilistic output that weighs changing signals. When teams expect deterministic behavior from a probabilistic system, trust erodes quickly.

The issue is not with the technology itself. It's the expectation of how that technology should behave. When adopting smart technology like AI, companies need to deploy each new solution using the equivalent of a pinball machine’s game board: letting the ball move freely within the boundaries of guardrails they have set.

Deploy the Right System for the Right Job

Understanding this distinction helps clarify the use cases where each type of system is most appropriate.

Traditional, rule-based systems work best when processes are stable, repeatable, and built around constraints that can be clearly defined. They excel at tasks that require consistency and precision; frankly, they are also ideal for the jobs that no human enjoys doing over and over.

AI-driven systems are better suited for situations where judgment is required, where data inputs are diverse, complex, and/or unstructured, and where there is no single correct answer. This is why it’s important to keep humans looped in and empowered to make crucial decisions based on factors that technology can’t precisely discern: things like tone, emotion, feelings, or body language.

However, organizations shouldn’t have to choose between traditional automation and AI. They can leverage both without having to become AI experts. Instead, companies can select a technology solutions partner with years of industry-specific experience, freeing them up to focus on what they do best: running their business.

With AI, The Foundation Matters Even More

AI may be game-changing, but one thing remains constant: AI systems still depend on a strong foundation.

A system making probabilistic decisions still requires clean, structured data, defined processes, and clear objectives. Without that foundation, the quality of those decisions quickly degrades.

Poor data leads to poor outcomes. Unclear processes introduce inconsistency. Lack of context reduces reliability.

Imagine inserting 15 balls into a pinball machine at the same time. Even though you have visibility into where each of them travels, there is no way to make a connection on the pattern or output they create unless you play it back in slow motion to evaluate the inputs and outputs at every turn. Today, AI can help you make these connections quickly and efficiently.

AI Success Starts with the Right Expectations

AI is not a replacement for automation. It's a powerful technology with complementary strengths. When it produces results that are different, that means it is operating as designed.

Those technology solutions providers who succeed with AI will not insist that their customers apply it everywhere. They will understand the difference between tasks that follow a predefined path, versus more complex use cases that require unstructured data gathering before they can propose outcomes.

They will have deep industry knowledge, support your use cases, and understand the value of collaboration to achieve measurable results. The difference between those in the know and those who are merely jumping on the AI hype train is not always easy to see, and the consequences of choosing the wrong path can be scary.

That’s why, when you invest in a partnership you trust, you gain confidence and clarity that comes from walking boldly into the forefront of change.

Learn how Epicor AI solutions can make every day easier for your business.

Follow me on Linkedin for more content like this.