Operations decision library
Twenty-seven operational decisions agents should not own alone.
This library began with sixteen LinkedIn use cases and now covers twenty-seven operational decisions across the ASCM SCOR Digital Standard. Every case starts with the decision, the people who live with its consequences, and the boundary an agent must respect.
Explore the interactive relationship map
A better plan makes the trade-offs visible.
Agents can reconcile signals, simulate scenarios, and surface conflicts. They cannot turn a forecast, segment, or scenario into an organisational commitment by themselves.
S&OP agents prepare the trade-off, they do not settle it
Agents can assemble the scenario pack, reconcile demand and supply views, and expose the gap before the meeting. The consensus number is still a management commitment.
Decision owner: S&OP / IBP leadership with supply, demand, and production planning
Agent boundary: The agent may model and reconcile. It may not publish a consensus plan without the S&OP owners.
Segmentation shows agents where not to generalise
Products and customers need different policies. The segmentation has to reflect operating reality, not a tidy clustering exercise.
Decision owner: Supply chain, commercial, and customer-service leaders
Agent boundary: The agent may cluster and propose. Segment policy and service promises stay with commercial and operations owners.
Risk agents need escalation rules
A long risk list is not management. Owners need thresholds, response options, and a clear point where the agent stops.
Decision owner: Risk, procurement, and supply chain leaders
Agent boundary: The agent may watch and rank exposure. Mitigation spend and supplier consequences need a human mandate.
Network-design agents simulate options, not commitments
Models can compare capacity, cost, resilience, service, and emissions. Leaders still own the irreversible choices.
Decision owner: Network strategy and supply chain leaders
Agent boundary: The agent may run the scenario grid. Footprint decisions are capital decisions and stay with leadership.
Control towers do not remove trade-off owners
Visibility helps teams see conflicts earlier. It does not decide which customer, site, or product absorbs the impact.
Decision owner: Cross-functional supply chain leaders
Agent boundary: The agent may surface the conflict and the options. Allocation between customers stays a named human call.
Forecast agents need decision discipline
A forecast becomes useful only after a team agrees how bias, overrides, promotions, and uncertainty affect the plan.
Decision owner: Demand planners and commercial partners
Agent boundary: The agent may generate and challenge the baseline. Overrides need a named owner and a recorded reason.
Inventory agents make trade-offs visible
Stock is a buffer against uncertainty and a cost. Humans still choose which risk the business accepts.
Decision owner: Inventory and supply planners
Agent boundary: The agent may recommend targets. Accepting a service-level risk is a business decision.
Local inventory agents can create global mess
Each node can optimise its own target and still make the network worse. Shared objectives and escalation rules matter.
Decision owner: End-to-end network planners
Agent boundary: Node-level autonomy needs a network-level objective, or the agents will compete with each other.
VMI agents move work, not accountability
Handing replenishment to a supplier agent changes who acts first. It does not change who answers for a stockout on the line.
Decision owner: Procurement and supply planning with the supplier
Agent boundary: The agent may replenish inside an agreed band. Band changes are a joint commercial decision.
Supplier automation still runs on definitions, evidence, and trust.
Procurement and quality teams need clear data rights, scorecard rules, release authority, and escalation paths before an agent acts across company boundaries.
Supplier agents need trust boundaries
Data sharing and commitments do not become neutral because software moves them. Supplier-facing owners define what the agent may see and do.
Decision owner: Procurement and supplier-management owners
Agent boundary: Agent read and write rights across the company boundary are contractual, not technical, decisions.
Supplier scorecard agents need an agreed definition of late
Automated scorecards fail on definitions long before they fail on data. Confirmed date or requested date changes the whole picture.
Decision owner: Procurement and supplier quality
Agent boundary: The agent may score against an agreed rule set. Escalation and supplier status changes are human.
Capacity signals from suppliers are claims, not facts
An agent can track declared capacity, lead-time drift, and allocation behaviour. Whether to believe it, buffer against it, or dual-source is a judgement call.
Decision owner: Procurement and supply planning
Agent boundary: The agent may model exposure. Dual-sourcing and pre-buy commitments need procurement authority.
Incoming-quality agents work inside a validated system
Trend detection on incoming lots is useful. In a regulated environment the release decision follows a validated procedure, not a model score.
Decision owner: Quality, supplier quality, and procurement
Agent boundary: The agent may flag drift and prioritise inspection. Lot release stays with qualified quality personnel.
Compliance agents should chase documents, not hide exceptions
Agents can follow up missing evidence. Functional owners decide what happens when documents, approvals, and reality do not match.
Decision owner: Quality and compliance owners
Agent boundary: Chasing is delegable. Deciding that an exception is acceptable is not.
Factory physics gets the first and last word.
Models help only when maintenance, quality, planning, engineering, and production teams can challenge the signal and act inside the real operating constraints.
Predictive maintenance is a decision problem first
A model can flag risk. People decide when to stop equipment, what to inspect, and which production commitment can move.
Decision owner: Maintenance and engineering leaders
Agent boundary: The agent may predict and prioritise. Stopping a qualified line is a production and quality decision.
Computer vision in quality is not AI magic
The difficult work is defining defects, evidence, and escalation. The camera is only one part of the quality system.
Decision owner: Quality owners and frontline inspectors
Agent boundary: The agent may detect and classify. Defect definitions, validation, and release stay with quality.
Scheduling agents cannot remove factory physics
Agents can compare schedules quickly. Planners still own feasibility, line constraints, changeovers, and the cost of nervous replanning.
Decision owner: Production planners and shop-floor leads
Agent boundary: The agent may propose a sequence. Freezing and releasing the schedule stays with production.
Batch-size agents cannot ignore changeover pain
Mathematical efficiency can collide with cleaning, setup, expiry, campaign, and service constraints.
Decision owner: Production and planning leaders
Agent boundary: The agent may optimise lot sizes. Campaign structure and cleaning validation set the hard limits.
OEE agents can only measure what the floor honestly records
Automated OEE exposes losses fast. It also exposes how loss reasons are coded, which is a people problem before it is an analytics problem.
Decision owner: Production leadership and manufacturing engineering
Agent boundary: The agent may compute and rank losses. Reason codes and improvement priorities need the line teams.
Cost agents need one version of the cost model
Product cost analysis is only as good as routings, BOMs, and overhead logic. Agents amplify a shared cost model or amplify the argument about it.
Decision owner: Manufacturing engineering, production, and controlling
Agent boundary: The agent may analyse variance and simulate. Standard-cost changes stay with controlling.
Digital twins need boring process truth
Simulation helps only when states, assumptions, and validation stay close to the real process.
Decision owner: Engineering and operations owners
Agent boundary: A twin may advise. It may not become the record of what actually happened.
Customer promises and reverse flows expose the real decision boundary.
Prediction and optimisation are useful. Spending money, changing a promise, approving a credit, or deciding disposition still needs explicit human authority.
Satisfaction agents read signals, people repair trust
Text, tickets, and delivery data can be mined for early dissatisfaction. The recovery gesture, and its cost, is a human choice.
Decision owner: Customer service and commercial operations leaders
Agent boundary: The agent may detect and summarise sentiment. It may not make goodwill commitments to a customer.
Fulfillment agents must protect customer trust
Shortages and delays become customer-facing decisions. People decide which promises can flex and which ones protect trust.
Decision owner: Fulfillment and customer-service leaders
Agent boundary: The agent may re-sequence and propose. Breaking a promised date is a commercial conversation.
A better ETA is only useful if someone acts on it
Predicting a late delivery early is cheap. Deciding to expedite, re-route, or call the customer is the part that costs something.
Decision owner: Transport, dispatch, and customer-service leads
Agent boundary: The agent may predict and alert. Expediting spend and customer notification follow agreed thresholds.
Routing agents need human exception logic
A neat route can ignore a driver, loading dock, safety issue, or customer promise. Dispatch defines the exceptions that matter.
Decision owner: Dispatch and logistics leads
Agent boundary: The agent may optimise the plan. Driver, safety, and site exceptions override the optimum.
Freight-cost agents expose the drivers, not the deal
Lane-level analysis shows where cost, mode, and service diverge. Carrier strategy and negotiation stay with people who own the relationship.
Decision owner: Transport management and procurement
Agent boundary: The agent may analyse and prepare the negotiation fact base. It does not commit volume to a carrier.
Return agents can triage, but disposition is a judgement
Reverse flows carry condition, entitlement, and compliance questions at once. An agent can classify and route the case; scrap, rework, or credit remains a decision with cost and quality consequences.
Decision owner: Customer service, quality, and warehouse leaders
Agent boundary: The agent may triage and prepare. Disposition, credit, and scrap approvals stay with quality and commercial owners.
Move from the argument to the artifact.
The operations map connects all twenty-seven decisions through six lenses: SCOR process, operating area, roles, systems, data, and AI role. Twenty-four decisions connect directly to full analytical briefs in the one-page case library; the remaining decisions link to a lecture or an honest map-only explanation rather than a fabricated destination.
The original cases are adapted from Frank Kienle's LinkedIn article series and public supply-chain analytics lectures. The expanded cases use the ASCM SCOR Digital Standard as a recognised process backbone. All examples are educational and contain no employer-, customer-, or supplier-confidential information.