Why Agentic AI is finally working in supply chains

For the past several years, artificial intelligence has been positioned as the solution to nearly every supply chain problem. Yet despite heavy investment, many organizations remain stuck in pilot mode, struggling to translate AI ambition into measurable operational value. MIT research last year found that the majority of AI projects failed to deliver the expected return on value.

The reasons are many, but those struggles are now giving way to a more disciplined approach to Agentic AI adoption. That approach is grounded in connected planning, decision history, and human-in-the-loop execution. Rather than layering AI tools on top of fragmented processes, leading organizations are embedding AI directly into the way supply chain decisions are made, tested, and revisited.

From AI enthusiasm to planning reality

A common failure pattern has become familiar across industries. Executives mandate AI adoption, systems are deployed, and expectations rise; only to fall short when the technology fails to solve the underlying business problem.

The issue, practitioners say, is not that AI lacks potential. It is that many initiatives are disconnected from core planning processes.

Kent Esslinger, senior director of industry solutions at o9 Solutions, told Supply Chain Management Review that the problem of adopting AI agents that “if it’s not enhancing the current process, there’s no value of having an agent on the side doing something.”

Why Integrated Business Planning has become foundational

Integrated Business Planning (IBP) is helping drive a shift in the adoption curve. Long treated as a periodic financial reconciliation exercise, IBP is evolving into a continuous, scenario-driven discipline that connects commercial decisions with supply chain realities in real time.

Esslinger, speaking at the recent NRF Retail Big Show in New York City, said this disconnect is precisely what next-generation IBP is designed to solve.

“We call it … next-gen IBP,” he said. “You have demand planning, supply chain planning; all these functional plans that have to come together to make that true integrated business plan that goes from commercial through to supply chain.”

What distinguishes newer IBP approaches is not just tighter alignment, but the ability to test decisions before committing to them within a shared planning environment.

Decision memory: the overlooked enabler of AI

One of the less visible barriers to effective AI adoption is the absence of decision memory. While companies have invested heavily in data lakes and dashboards, few maintain structured records of why planning decisions were made, under what assumptions, and with what expected outcomes.

“What people are used to is, ‘I have all these data sources, then offline I go make my plans and my meetings,’” Esslinger said. “It’s very hard to then go back a month later, a season later, and see what actually happened. Did we make the right decision?”

Agentic AI changes that equation when it is integrated inside the planning system itself, he explained. When decisions are modeled, executed, and later reviewed within the same environment, AI can begin to learn from outcomes, not just create correlations.

That learning loop enables what Esslinger described as “post-gaming”—the ability to revisit past decisions, analyze root causes, and apply those lessons forward across the supply chain.

The digital twin as the missing link

While large language models have demonstrated impressive conversational capabilities, supply chain leaders are learning that reasoning alone is not enough. Physical networks impose constraints that must be respected if AI recommendations are to be actionable.

“When it hits the supply chain world, you need a very structured model of how the supply chain actually operates,” Esslinger explained. “What constraints exist, what flow paths are allowed. You need that core underlying data model or digital twin that can be paired up with the LLM.”

Without that foundation, AI-generated insights risk becoming theoretically interesting but operationally irrelevant. With it, agentic systems can test scenarios that reflect real-world feasibility, not just statistical possibility.

Where Agentic AI is delivering value today

Rather than sweeping transformation, the most effective Agentic AI deployments are targeting specific, high-friction planning tasks.

Inventory root-cause analysis is one example.

“You can’t make recommendations at a high level where you need to get very granular very quickly on exactly why the inventory was wrong or in the wrong place,” Esslinger said. “That’s where the agents are super effective.”

Anjali Burkins, senior director of North American retail at o9, pointed to in-season decision-making as another area of impact.

“They’re spending so much time getting the facts and the root cause of what happened,” she said. “All that is resolved through the agent, and then you can make a decision based off the recommendations.”

In these cases, AI is not replacing planners, it is removing the friction that prevents them from acting quickly.

PepsiCo and the role of planning maturity

Large consumer products companies offer a useful lens for understanding why some AI initiatives scale while others stall. Organizations such as PepsiCo, which operates highly complex, global supply networks, have spent years maturing their integrated business planning capabilities before layering advanced analytics and AI on top.

The lesson is that AI amplifies planning discipline, it does not substitute for it. Companies that lack aligned commercial, supply, and financial planning structures struggle to extract value from AI, regardless of model sophistication, Burkin and Esslinger noted.

Human-in-the-loop is not a compromise

One of the most persistent misconceptions about AI in supply chains is that automation implies workforce displacement. In practice, the opposite is occurring.

“As the supply chain becomes more autonomous, the function of the supply chain professional has evolved toward exception management,” Burkins said. AI handles routine, repeatable decisions, while humans focus on judgment, trade-offs, and strategic oversight.

Importantly, agentic systems are being designed to ask clarifying questions rather than assume intent.

“The agent will actually prompt them back with the right clarifications required,” Esslinger said. “You may say, ‘What happens if demand increases by 10%?’ There’s a lot of things that need to happen to run that scenario” and the agent is trained on asking the right questions to narrow down the possible scenarios.

This interaction model lowers the barrier to advanced planning without requiring users to become AI specialists.

Why this moment feels different

What distinguishes today’s agentic AI efforts from earlier waves of supply chain automation is not novelty, it is integration. AI is no longer being asked to operate independently of planning systems, organizational processes, or human decision-makers.

Instead, it is being embedded where decisions already happen.

“Anything that can expedite or increase velocity for your planning is basically how that’s getting embedded,” Esslinger said. “Not a standalone AI product that you hope people use.”

As volatility persists and margins tighten, that practical orientation may prove decisive. Companies that treat AI as a decision accelerator, grounded in connected planning and real operational constraints, are beginning to move beyond experimentation.

SC
MR