Who Is Chairing Your Decision Cycle?
Why Agentic Orchestration Needs More Than Faster and Smarter Planning
Supply chain planning technology is evolving fast.
Forecasts are becoming more responsive. Simulations are faster. Optimization is becoming more granular. Re-planning is moving from monthly to weekly, daily and increasingly event-driven cycles. Advanced planning platforms are extending beyond traditional planning into concurrent planning, decision intelligence and agentic capabilities.
This is an important evolution in supply chain planning and decision making.
But it raises a different question:
What exactly are we preparing for agentic orchestration?
Because a collection of fast and smart planning tools is not yet a decision system.
From planning capability to decision-system maturity
Many organizations are still working to improve planning maturity—at very different scales and levels of sophistication: better forecasts, stronger optimizers, faster simulations, more granular scenarios and shorter re-planning cycles.
All create value.
However, technology capability is moving faster than implementation maturity and organizational mindset in many cases.
An advanced APS/ADP platform can therefore be capable of supporting much more while still being consumed primarily as:
Planner → APS/ADP → Simulation → Optimization → Re-plan
The planner becomes dramatically faster and smarter.
But the harder question is not whether these capabilities exist. It is whether they are being designed as one connected decision architecture—or implemented as increasingly sophisticated capabilities around the planner.
Because what happens when the decision leaves the planner's desk?
A promotion assumption changes. Finance revises an outlook. Commercial changes a market commitment. Operations identifies a capacity or execution constraint. An execution event invalidates an assumption used only hours earlier.
The planner may have a sophisticated optimization engine while the broader business decision continues through meetings, spreadsheets, documents and other functional systems.
Each part may work. But the decision itself can remain fragmented.
There is a subtle difference between making a plan faster and making a decision system work better.
In a decision-centric architecture, optimization and control-tower capabilities should be decision engines—not the architecture itself.
And that is why the decision cycle deserves at least as much attention as the planning technology within it.
The decision cycle is bigger than the planning cycle
I believe the evolution needs to be designed around the end-to-end business decision cycle, rather than individual planning applications.
One way of looking at that cycle is:
IBP → S&OP → S&OE → Manage Exception → Re-plan → Execute → Outcome → Learn
But this is not a linear planning process.
IBP, S&OP and S&OE represent different decision horizons and governance mechanisms within one connected cycle.
IBP establishes strategic direction and longer-term priorities. S&OP translates them into tactical cross-functional trade-offs. S&OE operates closer to execution, testing whether current assumptions remain valid. Exception Management identifies situations requiring intervention. Re-plan evaluates alternatives. Execution converts decisions into action. Outcomes tell us what happened. Learning feeds that experience into the next decision.
And the loop must work backwards as well as forwards.
Execution feedback may re-trigger S&OE. A recurring cross-functional issue may move into S&OP. Only a genuinely structural issue should challenge assumptions at IBP level.
Exception Management also has another important role in an agentic environment.
No planning or AI system will perfectly anticipate every future condition—at least not anytime soon. Some situations will fall outside modeled assumptions, established rules or an agent's delegated authority.
That is not necessarily a system failure.
Exception Management can therefore become a governed point where human judgment deliberately re-enters the decision cycle when uncertainty, materiality or risk exceeds established guardrails.
This is more than faster re-planning.
It is decision governance across horizons.
Where do agents fit?
Once the decision cycle is visible, agent roles become easier to define.
But before asking:
“What can an AI agent do?”
I would first ask:
What is the decision agenda?
Do we have the process and role definitions required to run it?
Only then would I ask:
“Which work inside this cycle should be performed by a human, an agent, an optimization model or an execution technology?”
A Signal Agent might detect material deviations. A Response Agent might diagnose exceptions and develop alternatives. A Goal Agent might translate business objectives into decision criteria and evaluate scenarios against objectives and constraints. An Alignment Agent might surface cross-functional conflicts and trade-offs. An Actuation Agent might translate an approved decision into execution. A Learning Agent might connect outcomes back to the decisions and assumptions that produced them.
The names themselves are not the point.
The point is to define what responsibility is being assigned to AI—and how that responsibility connects to the business decision agenda.
Governance before autonomy
Agent autonomy is only one design dimension. The governance interaction model is another.
Some organizations may deliberately run IBP, S&OP or S&OE through structured decision forums—physical, virtual or global—because their business, decision rhythm, risk profile or governance model requires collective judgment.
Others, with sufficiently mature processes and systems, may allow more decisions to progress through system-triggered, event-driven workflows, bringing people together mainly by exception.
Neither model is inherently more advanced. System-triggered governance is not a maturity goal in itself. Without mature processes, systems and decision logic, removing the meeting does not create orchestration—it may simply remove the governance.
But whichever interaction model is used, decision rights still need to be explicit.
Before assigning the agent, assign the decision right.
Agent capability and decision authority are not the same thing.
This is where traditional operating-model disciplines such as decision rights and RACI become highly relevant to agentic design.
Who is Responsible? Who is Accountable? Where may an agent recommend or act? When must a human approve? When should a decision escalate?
Human–Agent RACI should change with decision context, materiality, risk and maturity.
At IBP level, agents may be analytical and challenge partners while strategic judgment remains human-led. Within S&OP, S&OE and re-planning, humans and agents may work more closely together. Closer to execution, sufficiently mature and governed processes may move toward human-on-the-loop, with agents operating within predefined authority and guardrails and humans intervening by exception.
Neither governance model removes the need for a structured decision agenda, explicit decision rights and an integrated decision log.
The interaction model may change. The decision discipline should not !!!
Connecting the decision—not just the systems
Even strong Human–Agent governance is insufficient if every function continues to operate within a different decision context.
Supply Chain decisions rarely belong only to Supply Chain.
A promotion changes demand. A commercial commitment changes priorities. A financial assumption changes the acceptable trade-off. An operational constraint changes feasibility. A supply decision affects revenue, margin, working capital and service.
Commercial, Finance, Supply Chain and Operations may use different applications—or even different domain control towers—but they are participating in the same business decision context.
That requires more than connected data.
It requires the ability to preserve decision lineage across functions and systems.
Suppose an S&OP decision changes a consensus forecast. Supply Chain re-plans. Commercial modifies a promotion. Finance updates its outlook. Operations changes execution.
Depending on the governance model, some of these decisions may be made in a structured forum; others may be system-triggered actions.
Either way, the decision chain should remain connected.
Can the organization later reconstruct:
Which assumption triggered the decision? Which plan version was considered? Why was one scenario selected? Which subsequent assumptions changed? What action followed—and what actually happened?
At its simplest:
Plan Version ↔ Decision ↔ Assumption ↔ Action ↔ Outcome
Not every system needs to contain everything. The architecture can remain distributed.
But the semantic relationships between those objects need to survive across systems.
Connected transactions do not automatically create a connected decision system.
Semantic linkage should preserve the lineage of the business decision, not merely the technical integration between applications.
From decision history to institutional memory
This also changes what a decision log can become.
A mature decision architecture increasingly connects:
Connected Plans ↔ Connected Decisions ↔ Connected Outcomes ↔ Institutional Memory
Connected Plans tell us what we expected.
Connected Decisions tell us what we chose—and why.
Connected Outcomes tell us what happened.
Institutional Memory connects them over time.
Instead of asking:
“What did we decide last time?”
the organization can ask:
“Under similar conditions, what did we decide, why did we decide it, and what actually happened?”
For agentic AI, that distinction matters.
Giving an agent access to historical documents gives it information.
Connecting:
Trigger → Assumption → Scenario → Decision → Rationale → Action → Outcome
gives it decision context from which learning becomes possible.
Decision history is not institutional learning until the decision can be connected to its context and outcome.
An agent that can retrieve history has memory. An agent that can relate context, rationale, action and outcome begins to have something much closer to decision intelligence.
So, what needs to sit underneath?
A connected decision cycle cannot operate on technology alone. It needs three complementary foundations.
Business & Operating Model — strategic direction, objectives/OKRs, decision rights, governance, roles, KPIs, alerts and triggers.
Decision & Optimization Model — decision objectives, constraints, trade-offs, scenarios and optimization logic that remain calibrated to changing business priorities.
Data, Semantics & Technology — master data, ontology, interoperability, analytics and semantic linkage that preserve decision context and lineage across systems.
A model can solve the mathematical problem perfectly and still support the wrong business decision if its objective no longer reflects the business strategy.
This is where the Goal Agent can help maintain alignment across:
Strategic Objectives → Decision Objectives → Constraints & Trade-offs → Optimization Logic
Together, these foundations support something larger: an operating model that connects decision context, governance, planning, execution and learning.
The picture is not a technology stack. It is an operating model for how decisions move.
Cross-functional context defines what matters. The decision cycle connects strategic intent to execution and learning. Agentic orchestration works across that cycle. Human–Agent governance determines authority and intervention. Execution turns decisions into outcomes, while the connected decision system preserves the lineage between plans, assumptions, decisions, actions and results.
The foundations make the system possible. The connections make it work.
Conclusion: So, who is chairing your decision cycle?
Perhaps the answer is not simply “the agent.”
Some decisions will remain human-led, some agent-supported and others increasingly agent-led within defined guardrails. Likewise, some will continue through structured decision forums while others flow through system-triggered workflows.
What matters is not whether the chair is human or digital.
What matters is whether the decision system knows who—or what—should chair which decision, under what authority, and when.
That is where the chair metaphor becomes useful. Chairing is about orchestration and governance—not automatic decision authority.
A good chair identifies what requires attention, brings the right context and participants together, challenges assumptions, makes alternatives visible, records the decision, triggers what happens next—and ensures the outcome comes back into the cycle.
So before asking:
“How much faster can we generate a plan?”
I would ask:
“Do we have a connected decision cycle where humans and agents can operate together—and agentic orchestration can actually work?”
Can Commercial, Finance, Supply Chain and Operations operate from a shared decision context?
Can execution re-trigger the appropriate decision horizon?
Have we defined Human–Agent decision rights, governance mechanisms and guardrails?
Can decisions remain connected to their assumptions, actions and outcomes—and can those relationships inform the next decision?
Because the real transformation is no longer only about making the planning process smarter.
The unit of transformation is becoming the decision cycle.
Faster planning ≠ Decision orchestration.
Connect the decision loop before you orchestrate it.
Comments