Static safety stock is either over-stock or a stockout. Almost never the right one.

The problem

The fixed buffer per SKU is decided once a year and lives off intuition.

Classic safety stock is calculated like this in most mid-sized plants: once a year, in a spreadsheet, with a simple formula (lead time × average demand × safety coefficient) and a manual adjustment from whoever runs the warehouse. What happens next is predictable:

1. SKUs with stable demand and reliable supply — the buffer stays in place out of inertia. It weighs on working capital without contributing anything. Nobody lowers it because there is no incident either.
2. Critical SKUs with high variability — the buffer falls short when an unexpected campaign, a supplier delay or a change in customer behaviour arrives. The stockout hits and everyone fights the fire.
3. Seasonal SKUs — the buffer is inflated all year just in case and deflates exactly when the campaign starts, because nobody had time to touch it.
4. New SKUs — they are launched by eye with indefinite over-stock until a year goes by and the manager remembers to adjust.

The problem is not the formula — it is that nobody can keep that formula alive per SKU, in a plant with hundreds or thousands of references, when demand and supply move every week. The stock manager would have to spend the entire working day on the calculation, and does not, because there is another job to do.

How it fits the IRIS system

iLEAN does not replace the stock manager — it gives them an agent that recalculates on their behalf.

IRIS (Industrial Reality Intelligence Systems) is the category; iLEAN is the system. Dynamic safety stock calculation is one of the cleanest examples of the iLEAN principle: the data exists (ERP movements, real measured lead time, historical demand, seasonality), but nobody has time to operate it. The stock manager's agent is the librarian that was missing from a library full of data with no reader.

Static stock is thought through once a year. Dynamic stock is thought through every time demand or supply changes. The person signs off; the agent sustains it.

The iLEAN pieces applied to dynamic safety stock:

• iLEAN Brain (Central) — the multi-agent brain where the coverage calculation logic per SKU lives. Here historical demand, commercial leading indicators, real lead time per supplier, SKU criticality (margin, substitutability, impact of a stockout) and retailer constraints all come together.
• iLEAN Agent (for the stock manager) — an assistant that keeps coverage alive per SKU, proposes parameter readjustments when reality moves, flags critical SKUs with growing variability, and prepares the dashboard the manager reviews each morning. It proposes; the person signs off.
• iLEAN Connect — captures the signals the ERP cannot see: the supplier email warning about a delay, the sales rep's message about a campaign starting earlier, the shift in a key account's behaviour. Those signals are exactly the ones that break the static buffer — and the ones dynamic calculation has to see at second zero in order to anticipate.

See the full IRIS architecture →

Before and after

Static safety stock vs. dynamic stock with iLEAN

Aspect	Static safety stock	With iLEAN Brain + Agent
Recalculation frequency	Annual (sometimes half-yearly)	Continuous, on any relevant change
Granularity	By family or ABC category	Per individual SKU
Variables considered	Average demand + promised lead time	+ real lead time + variability + seasonality + leading indicators
New SKUs	By eye, indefinite over-stock	Proxy from analogous SKUs, progressive adjustment
Seasonal SKUs	Fixed buffer all year round	Curve anticipated per campaign, orderly release afterwards
Total stock	Inflated by psychological safety	Optimised per SKU, same coverage or better

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

This block is an estimate to be validated with the specific data of your plant. It works as an order of magnitude for the steering committee; we refine it during the diagnostic.

• Multi-SKU plant (from 100 to several thousand active references) with safety stock calculated in a spreadsheet or in a basic ERP module, reviewed annually or half-yearly.
• Immersion of 3-5 days, Pareto identified in the inventory (the SKUs that concentrate 80% of the cost and of the stockouts).
• First dynamic calculation agent operating within a few weeks; first recalculation cycle per SKU validated by the stock manager.
• Total stock reduction between 15% and 35% in the first measurable cycle, with the same service level or better (conservative goal: hold or improve the service level, never worsen it).
• Reduction of stockouts on critical SKUs ≥ 30% through better anticipation of real variability.
• Indicative payback between 4 and 9 months. The hard lever is releasing the working capital tied up in inventory.

The underlying fact — volatility is here to stay

Ocean freight went from ~1,500 USD/container to >20,000 at the 2021 peak, rebounded in 2024 (Red Sea) and then fell sharply on overcapacity. The honest argument is not that transport is expensive today — it is that long chains became fragile and volatile, and safety stock calculated when everything was stable no longer protects. Dynamic calculation is not an optional improvement; it is how inventory reacts to volatility without inflating the buffer blindly.

And the CAIO's doubt — does the AI invent the coverage?

What if the agent lowers a critical buffer because of a hallucination and we run out during the campaign? — hallucination is a problem of free generation, not of anchored tasks. Dynamic stock calculation is pure statistics over ERP data — there is no free generation, there is optimisation anchored to time series. On anchored tasks the best models pushed the error below 1.5% ^[1]. And even so, the agent proposes; the stock manager signs off before anything enters the ERP. The three safety rings are there precisely for this.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.