On a typical material, a perfect forecast takes about 8% out of the buffer. A lead time you can rely on takes about 60%.
Most transformation programmes are aimed at the first number. This page is about the second: how supply chains are actually steered, and why the answer sits between the operating model and the running system.
Christian Kroschl
Twenty years in pharma and manufacturing networks. The thinking here did not start with AI or APS. It started with variability, buffers, and who is allowed to decide.
Every board is now asking how agentic AI will lower cost, release working capital and raise productivity. This is the answer worth giving in the room.
<aside> →
The Board Wants Agentic AI. What Should the Supply Chain Leader Do? · about ten minutes, five figures
An agent has no business case. A changed decision has one. Agents can only act on decisions that are machine readable, and most are not. What transaction data can reveal, what it cannot, and how the AI question can be used to bring a running planning transformation back to the business case it started with.
Sixty seconds first, if you prefer: You are the agent puts you in the position and makes the same argument in one move.
</aside>
The Middle Nobody Owns · the argument in full, about eight minutes
A supply chain programme can succeed on both halves and still fail as a whole. Strategy stops at the concept, system integration starts at the scope, and the question neither one answers is how the supply chain is actually steered. With the numbers, and the simulators as evidence.
The Field Nobody Writes To · the mechanism underneath it, about nine minutes
Closed loop MRP closes around orders, and has done so correctly for fifty years. It has never closed around the fields that drive the plan: lead time, lot size, safety stock, capacity. Five separate traditions derived the return path between 1997 and 2007. Not one of them wired it. What has to be built, and the two preconditions you can check in an afternoon.
One operating model, four trade-offs and one experiment. The first tab is the frame the others sit inside.
The frame
Integrated Planning & Control.
Six principles and a twelve step planning process, from the supply model through S&OP, MPS, MRP and capacity planning to the shop floor, plus the operating model around it: systems, policies, roles, measurement and proof. Thirty two KPIs, four simulations, and a twenty step path in dependency order. Each element carries a definition, a worked example from the plant floor and a question to test yourself against.
Integrated Planning & Control
S&OP Suite
One decision, three bills.
A market asks for +30%, the product is six months out, and the same call reads as volume to Commercial, as capital to Finance, as service to Supply Chain. Six acts, a live value stream, the operating cards and a decision board, with a 2:17 voice-over if you would rather listen than read. Supply chain runs the process. It does not own the outcome alone.
The S&OP Simulation Suite
Pattern Wheel
Rhythm against reaction.
A fixed rotation sequence against reactive replanning, run day by day over a year. Run the plant to a repeating pattern and a customer order ships the same day on 93% of days. Run it reactively and that drops to 45. Move the time buffer, the stock buffer and the service target, and watch how much scatter your supplier is left to plan around.
The Pattern Wheel Simulator
Safety Stock
Which side of the formula drives your inventory.
Move the faders on forecast error, lead-time variability and service level, and see live which term is actually setting the buffer. The answer is rarely the one the forecast debate assumes.
The Safety Stock Synthesizer
Lead Time
One input, two consequences.
One set of lead-time components, two effects: how rigid the plan becomes, and how much safety stock it costs. Move the sliders and watch both shift together.
The Lead Time Cascade
You are the agent
Same data, three confident answers.
300 units, 520 of open demand, every transaction visible. Allocate — then watch two more runs on the same data argue their way, fluently, to different answers. What is missing is not data. It is one written sentence, and once it exists the same decision becomes instant, identical on every run, and explainable afterwards. The shortest version of the agentic AI argument: rules before tools.
You are the agent
APS and AI
Why APS and AI alone do not stabilise a supply chain, and why structural clarity matters more than optimisation power.
Operating model
Planning as an operating model: decision rights, governance, and the shift from isolated planning to orchestration.
Stabilisation
Why plans break in execution, and how buffers, decoupling and structure create operational stability.
Talent and data
Why technology initiatives fail without clean master data, the right capabilities, and clearly owned roles.
From the field
Foundations