Moving Toward Autonomous Planning Without Losing Sight Of ROI

Direct Source Verification: This story is aggregated from Forbes (forbes.com). Full reporting rights and copyright belong to the primary publisher.
As businesses give AI more authority, they need to understand what those decisions will cost, what they will improve and how to measure the results.

gettyAlex Koshulko, Ph.D, is a supply chain planning expert with 10+ years of experience. CEO and co-founder of Streamline.

What if your AI agents hit their KPIs and still cost your company $330,000?

In a simulation published by Supply Chain Management Review, five AI agents represented different supply chain functions. Four improved their performance metrics, yet the simulated company’s EBITDA fell by approximately $330,000 compared with a scenario where their objectives were aligned. Each function pursued its own targets at the expense of the overall business.

For me, this exposes one of the challenges in autonomous planning. An agent can perform its assigned task successfully while making decisions that hurt the company’s financial performance.

As businesses give AI more authority, they need to understand what those decisions will cost, what they will improve and how to measure the results.

Ask a planner why they cut last week’s purchase order by 20%, and you may hear reasons that the planning system doesn’t fully capture. An AI agent reviewing the same order needs the numbers and the context behind them.

Companies preparing for autonomous planning have good reason to start with data management. McKinsey found that approximately 65% of advanced planning system programs failed to achieve their expected ROI, with poor data management among the five leading causes.

Fixing the data addresses part of the challenge. But how do you give an agent the knowledge planners use to interpret it and assess the consequences of their decisions?

The solution might be a contextual layer that brings together ERP data, input from planners and simulation results that show how different planning choices could affect operations and finances. Building it requires mature processes, but the work can improve them.

The goal is to give agents enough context to evaluate the financial consequences of their actions before execution, with people responsible for setting objectives and constraints. But even then, forecasts and simulations cannot guarantee the outcome. Business results must determine when agents receive greater authority.

A planner may spend time exporting reports, reconciling figures and obtaining approvals because the systems involved cannot share information directly. An AI agent with access to the relevant data and business context can handle parts of that work without following the same sequence of steps.

Before delegating a workflow, ask: if an agent had direct access to the right information, would we design the process the same way?

Gartner predicts that cross-functional multiagent systems will execute 35% of enterprise workflows with minimal human approvals by 2030, up from 3% in 2025. As companies delegate more work to agents, they need to distinguish manual steps imposed by system limitations from controls that protect business outcomes.

Data transfers and duplicate checks may no longer require human effort, but approvals tied to financial exposure, customer commitments or difficult-to-reverse decisions serve a purpose. These distinctions should shape the workflow before an agent takes over its execution.

I suggest beginning with a narrow pilot where the consequences of a wrong decision are manageable and can be detected quickly. Replenishing a stable product is a more practical starting point than allocating scarce inventory among strategic customers during a disruption.

The pilot should show whether the business outcome improved against a baseline. Before expanding its authority, the company should test how the agent responds when demand, supply conditions or business constraints change.

The level of autonomy needs to reflect the risks of each decision. A company might allow an agent to handle replenishment within defined limits, require planner approval for higher-value orders and keep strategic allocation decisions under human control.

Gartner’s forecast illustrates how limited autonomous decision-making is expected to remain. While 83% of surveyed organizations had already invested at least $3 million in planning automation, including AI, Gartner expects only 5% of organizations implementing supply chain planning automation to allow systems to make at least 10% of planning decisions autonomously by 2030.

When financial performance improves after an AI rollout, how much of that improvement can we credit to the agent? Inventory may have fallen because demand weakened or supplier terms changed. Both factors need to be accounted for when calculating the agent’s contribution.

It’s also worth looking at the impact on working capital and operating costs. Lower inventory can free up cash, but if it leads to emergency shipments or missed orders, those consequences belong in the assessment, too. So do the costs of deploying and operating AI.

A purchasing decision may take weeks to reveal whether it protected service levels or created excess stock. During that time, an agent could make dozens of similar decisions before the business sees the consequences. Evaluate ROI across the relevant purchasing and replenishment cycles, allowing enough time to measure the financial outcomes of those decisions, including the cost of correcting errors.

Gartner predicts that by 2029, 60% of supply chain planner roles will shift from operational execution to strategic oversight as autonomous planning systems take over routine decisions.​ As a result, I expect planners to spend more time setting the boundaries for autonomous execution. They will define spending limits, decide when human intervention is necessary and review results across planning cycles.​

A planner who once approved individual purchase orders may soon oversee hundreds of decisions made by AI agents. In our practice, for example, a North American CPG manufacturer increased the number of forecast edits by more than 100 times through the use of AI agents. This illustrates how agents can expand the scale of planning activity, but it also makes oversight more important. Planners will need to recognize when established rules require adjustment, whether because supplier conditions have shifted, customer priorities have changed or financial results no longer justify the current approach.​

AI agents will not eliminate the need for human judgment in supply chain planning. While they have tremendous potential in the supply chain, ​their value will depend on how effectively organizations combine autonomous execution with human oversight.​​​​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?

Original Source
https://www.forbes.com/councils/forbestechcouncil/2026/11/20/moving-toward-autonomous-planning-without-losing-sight-of-roi/
Visit Forbes ↗
SHARE STORY:
𝕏 f in

Related Coverage in Business