Don’t let the robots take over: How to use AI in supply chain planning without losing control

If you’ve ever watched I, Robot, you know the plot: the robots were supposed to help… right up until they started running the world. That’s the image Søren Hammer Pedersen opens with in Roima’s S&OP MasterClass. Half joke, half warning. Because if you look at LinkedIn, events, and the conversations happening across supply chain planning right now, it can feel like AI is taking over everything.

And maybe that’s not entirely wrong.

The question isn’t whether AI is coming into planning. It already is. The question, rather, is: how do you balance automation with human judgment so you actually gain value without letting a black box make critical decisions at scale?

Søren is joined by Benjamin Obling, who has spent 16 years working with PERITO IBP at Roima and is responsible for onboarding and developing clients on the platform. And their conversation lands on a clear message: AI is powerful, but the companies that win will be the ones who use it with common sense and intent.

Benjamin is the first to admit that the answer can sound trivial: there is huge potential, and everybody can see it. But he frames it in a way that makes the stakes very practical.

There are plenty of tasks in the supply chain that we want automation to take over, because humans shouldn’t spend their careers pushing data around in Excel. If AI can do those tasks better and faster, planners can spend their time on work that is more strategic and, frankly, more fun.

But that’s also where the risk shows up.

The danger that AI will take over in some dramatic sci-fi way isn’t really it. The danger is more mundane: AI can do bad things very fast if you don’t have a grasp of what it’s doing.

That’s why people should care now. Not because AI is new, but because it’s moving from “cool technology” into the actual decisions that shape inventory, service, and customer value.

When Søren asks where AI really makes a difference in practice, Benjamin points to the areas where the supply chain has always felt the weight of manual effort and imperfect decisions.

AI is already being used to improve and automate:

• Demand forecasting (accuracy and automation)
• Inventory planning (including settings that influence safety stock)
• Master data follow-up and predictions
• And more broadly, optimization opportunities across planning

In other words, it’s no longer just a chatbot helping people create presentations or summarize meetings. It’s moving deeper into the planning engine; into the numbers that guide purchasing, production, and inventory strategy.

And that’s where governance suddenly matters.

A natural assumption is that if AI improves forecast accuracy, then it’s automatically good. Benjamin’s point is more nuanced: accuracy isn’t the only goal – stability matters too.

In the “old world,” many planning approaches were predictable. If you calculate a rolling average, you get the same result every time. It might be boring, but it’s stable. Many people grew up believing computers are never wrong because they are consistent.

AI breaks that mental model.

Benjamin describes AI as having built-in randomness and as a “black box by definition.” That means predictions can change in ways that you don’t feel intuitive. The risk is that AI will interpret noise as signal, and then produce forecasts, safety stocks, or purchase proposals that go completely out of line.

That’s the moment where “AI is amazing” turns into “AI is scary.”

Not because it can’t deliver value, but because without guardrails, you can end up with something unstable that makes extreme decisions at scale.

So where do you start?

On one side, list the possibilities: demand forecasting, safety stock, supply planning, optimization, and even network-level decisions. On the other hand, look at your company’s challenges.

Do you have a forecasting problem? Do you believe you can improve accuracy? If yes, what is the benefit, and what’s the business case?

Benjamin predicts that the AI hype will cool quickly because companies will start demanding results instead of buying something just because it has the letters “AI” attached. Søren jokes that some companies might not just want one AI initiative; maybe they want two. But the point stands: the starting point is impact.

Build a business case by asking:

• What workforce can we release from manual tasks?
• How much accuracy can we improve?
• What is the value of that change?

It sounds basic, but it’s the anchor that keeps you from automating things that don’t matter.

Søren pushes the conversation beyond the planning silo. If AI improves forecasting, shouldn’t that ripple into other areas too?

Benjamin gives a concrete example. If you have a better forecast, you’re obviously using it in purchasing and production planning. But you could go further: inbound deliveries in the warehouse. If you can predict what will be picked later, you can decide where to place goods when they arrive – reducing handling and improving efficiency.

That’s a powerful idea: AI predictions don’t only improve planning outputs; they can be reused across functions if there’s a business case.

And it’s also a reminder that not every prediction matters. Some predictions change nothing. Others unlock operational savings. Your job is to find places where a better prediction changes what you do.

Many companies have the first steps in place. They accept that AI is a black box. They build business cases, and they choose priorities.

What’s missing are the last two steps:

1. How do we guard AI?
2. How do we design the human role?

Guardrails are in two layers.

At a very practical level, you can fence predictions in. For a forecast, you might limit how much it can increase or decrease, so it doesn’t go “completely nuts” compared to what’s historically plausible.

At a more strategic level, you define the playing field of AI. For example, in the product portfolio: what should be stocked, and what shouldn’t? AI can propose moving SKUs between make-to-stock and make-to-order. It’s a commercial decision – It’s your company’s value proposition to customers.

AI can provide excellent input, but humans must decide what the company wants to offer “off the shelf” versus what customers can wait for.

And this is why “it’s just calculating safety stocks” is a dangerous sentence. Safety stock decisions shape inventory strategy and customer promise. That should never be handed over without human intention and governance.

The beauty and horror of automation is that you can create structured, scaled errors very fast. Nobody wants that.

Søren asks the question many planners are thinking: when do you trust AI, and when don’t you?

Benjamin frames it as “the predictability of the prediction”, meaning how much the output changes when the inputs change.

If you have an area that’s easy to predict, letting AI take over is relatively safe. It will likely do a very good job most of the time.

The tricky cases are the classic planning nightmares:

• lumpy demand
• slow movers
• sporadic spikes that might be noise or might be a new level

In those cases, AI might interpret temporary spikes as structural changes and dramatically inflate the forecast or safety stock. That’s where you need judgment.

But again, guardrails change the game. If governance ensures that “big money” decisions require approval – like purchases above a certain euro threshold – then AI can still work quickly without running wild. If it’s a cheap item like O-rings, maybe you don’t care. If it’s an expensive, long lead-time component, you absolutely do.

The final part of the conversation turns from daily planning into monthly governance: S&OP.

Benjamin is quite optimistic. AI enables you to run the process with higher quality:

• better forecasts
• better capacity simulations
• faster what-if scenarios
• faster analysis and decision material (including comparisons between scenarios)

The workload reduction is significant. In demand planning specifically, Benjamin says manual work around the forecast can drop by 30% up to 70–80% when AI models are used.

That changes the nature of the role. Planning becomes less about creating numbers and more about interpreting, controlling, and using them. Some of the day-to-day work may become alert-driven: approve, reject, and review exceptions. That might sound trivial, but the trade-off is big: time is freed up for scenario thinking and decision preparation.

“What if we change service levels?”

“What if we open a new production line?”

“What if demand in Germany increases by 10%?”

That is a different job than downloading data from SAP and working in spreadsheets.

But Søren adds a critical point: leadership may be aware AI is a black box, and that can create fear. So, communication matters. Companies need to explain how AI is controlled, fenced, and validated, or they’ll constantly backtrack with “that can’t be right.”

So, don’t give up on common sense. Black box doesn’t mean no business case, no explanation, and no governance. You may not understand the deep learning network inside, but you can understand:

• what you put into it
• how you validate outputs
• how you fence it
• how alerts and workflows build trust

When Søren asks where this is going, Benjamin is honest: it will change dramatically, and some of what they’re saying will likely be partly wrong in two years. The practical advice is therefore not to build five-year payback assumptions in a space that could look different in two.

Instead:

• build business cases with a short time horizon
• improve fundamentals like master data
• automate what you can now to free up resources
• and use those freed resources to adapt to what comes next

Be quick on your feet. Harvest results fast. Because AI won’t wait.

And the goal isn’t to treat it like magic, but rather to use its power while keeping humans in control of what really matters.