The replenishment clock you are actually planning against
Ask three people at your company what the lead time on a magnet is and you will get three numbers, all of them wrong in the same direction. The quoted figure is almost always the production window, measured from the supplier's order acceptance to the goods leaving their dock. Your planning parameter needs to cover considerably more than that.
Totalled honestly, a custom sintered NdFeB part produced offshore lands somewhere between ten and sixteen weeks door to door, and a first article or a grade with heavy rare-earth content sits at the top of that band or beyond it. The lead time and MOQ guide breaks the production window down stage by stage; what matters for planning is that you carry the whole chain in your parameter, not the fraction of it your supplier is willing to commit to.
Internal cycle time and receiving inspection sit at the two ends of the chain and belong to nobody. They are the easiest weeks to remove and the ones most likely to be missing from the ERP lead-time field. Measure them from your own transaction history before you argue with the supplier about theirs.
Two structural features of magnets make this worse than a generic long-lead part. Tooling — press dies and sintering fixtures — is part-specific, so a second source cannot absorb a shortfall without its own qualification runs. And magnetized inventory is awkward to move quickly, because air freight of magnetized material runs into UN 2807 field limits. Expediting a magnet is genuinely harder than expediting a machined bracket, and your buffer should reflect that.
Measuring variability instead of assuming it
Safety stock exists to cover variance, not average demand. Sizing it therefore requires two numbers that most planners estimate by feel: how much demand moves, and how much lead time moves. The second is usually the one that hurts, and it is the one nobody measures.
Demand variability
Take at least twelve, preferably twenty-four, periods of actual consumption at the period length you plan in — weekly if you release weekly. Compute the standard deviation of that series and divide by the mean to get a coefficient of variation. Below about 0.25 the part is stable and forecastable. Between 0.25 and 0.75 it is ordinary. Above 0.75 you are not forecasting, you are guessing, and the answer for that part is a different supply model rather than a bigger buffer.
Strip out the noise that is not really demand: a one-off qualification build, a customer's inventory correction, the quarter you double-ordered because of a scare. Leaving those in inflates the deviation and buys buffer against events that will not repeat.
Supply variability
This is the honest work. For each of the last ten to twenty receipts, record the actual door-to-door elapsed time, not the promised one. The standard deviation of that series is your lead-time sigma. On offshore magnet supply it is common to find a mean around eleven weeks with a deviation of two to three weeks — and it is that deviation, squared and multiplied by the square of average demand, that dominates the safety stock formula.
In the combined formula, demand variance is multiplied by lead time, while lead-time variance is multiplied by the square of average demand. For a part with steady consumption and an erratic supply chain — which describes most magnet programs — the second term is frequently three or four times the first. Buffer sized on demand variability alone will be badly short.
If you have fewer than about eight receipts, you do not have a distribution. Use the supplier's own on-time record if they will share it, apply a deviation of roughly 20% of the mean lead time as a working placeholder, and flag the parameter for review once real history accumulates.
Sizing safety stock properly
The combined formula below covers variance in both demand and lead time. It is the only version worth using for magnets, because the single-variable simplifications assume a stable lead time and that is exactly the assumption that fails.
z = service factor for the target cycle service level
LT = average lead time, in the same periods as demand
σD = standard deviation of demand per period
D̄ = average demand per period
σLT = standard deviation of lead time, in periods
The reorder point is then average demand across the lead time plus that buffer:
A worked example
A sensor ring consumed at an average of 900 pieces per week, with a weekly deviation of 260 pieces. Lead time averages 11 weeks with a deviation of 2.5 weeks. At a 95% service level, z is 1.65.
Now cut the lead-time deviation from 2.5 weeks to 1.0 week — achievable through a scheduled release agreement or a domestic stocking position — and safety stock falls to roughly 1,850 pieces. Halving supply variability removed more than half the buffer. No forecasting improvement available to you will do that.
Choosing the service level honestly
| Service level | z | Buffer vs. 95% | Appropriate for |
|---|---|---|---|
| 90% | 1.28 | 0.78× | Low-value parts with a fast alternative or a tolerant build schedule |
| 95% | 1.65 | 1.00× | The sensible default for production magnets |
| 98% | 2.05 | 1.24× | Single-source parts, or where a stockout stops a line |
| 99% | 2.33 | 1.41× | Safety-critical or contractually penalised supply |
| 99.9% | 3.09 | 1.87× | Almost never justified; the cost curve goes vertical here |
Applying 99% across the board is a common and expensive default. Service level is a per-part decision driven by what a stockout actually costs, and on most portfolios the answer varies enormously between parts.
When MOQ overrides the order quantity you wanted
Classical economic order quantity balances ordering cost against holding cost and produces a tidy number. On magnets that number is frequently below the supplier's minimum, at which point the arithmetic stops being a decision and becomes an observation.
D = annual demand in pieces
S = fixed cost of placing and receiving one order
H = annual holding cost per piece, typically 18–28% of unit value
Magnet MOQs are driven by press setup, sintering furnace loading and coating batch size, not by arbitrary policy. A supplier running a 2,000-piece minimum on a custom arc segment is telling you the sinter load and the plating barrel do not economically run smaller. Pushing below it buys a setup charge rather than a lower commitment.
| Situation | What to order | The trap |
|---|---|---|
| EOQ well above MOQ | Order EOQ, rounded to a pack or pallet quantity | Rounding up a full pallet on a slow part quietly buys a year of cover |
| EOQ close to MOQ | Order MOQ; the difference is inside the noise | Optimising a number whose inputs are ±30% accurate |
| EOQ far below MOQ | Order MOQ, but reconsider the sourcing model entirely | Carrying two years of a part because the minimum said so |
| Tooling amortised in price | Model the tooling separately from the piece price | A small order carrying the whole die cost looks like a terrible unit price |
| Stocked standard part | Order to the buffer; MOQ is usually irrelevant | Applying custom-part logic to a catalogue item |
The third row is where the real decision lives. If the minimum forces two years of inventory onto a custom part, the honest options are to consolidate the specification with a neighbouring part so volume rises, to move to a stocked standard geometry, or to negotiate a blanket order with scheduled releases so the supplier runs the economic batch while you take delivery against consumption. That last structure is what blanket orders, consignment and VMI exist to provide, and it resolves the MOQ conflict better than any order-quantity formula.
A surprising share of custom magnet parts differ from a stocked standard by a fraction of a millimetre or by a coating choice that carries no functional requirement. Where the design can absorb it, moving to a stocked size collapses the MOQ problem and the lead-time problem in one step. That is the first question to ask in any cost reduction review.
Segmenting the portfolio so effort lands where it matters
Running the full parameter exercise on every part number is a good way to run it on none of them. Two axes are enough: annual spend, and predictability of demand. The intersection tells you which supply model each part deserves.
| Predictable (CV < 0.25) | Ordinary (0.25–0.75) | Erratic (CV > 0.75) | |
|---|---|---|---|
| High spend | Blanket order, scheduled releases, monthly parameter review | Blanket plus buffer at 95–98%; supplier holds finished goods | Consignment or make-to-order; do not buffer against noise |
| Medium spend | Standard reorder point, quarterly review | Reorder point at 95%, quarterly review | Move to a stocked standard part if the design allows |
| Low spend | Generous buffer; the carrying cost is trivial next to the transaction cost | Generous buffer, annual review | Buy to a fixed cover period and stop analysing it |
Low-spend parts deserve the least analysis and the most inventory. A retaining magnet costing forty cents that halts an assembly line has an absurd cost-of-stockout to cost-of-carry ratio; buy a year of it, put it on a shelf and take it off the review list. The planner hours saved are worth more than the working capital consumed.
The erratic, high-spend cell is the one that punishes the wrong answer. Buffer sized to genuinely random demand is enormous and mostly idle. These parts want a supply structure — consignment, a supplier-held finished goods position, or a qualified second source — rather than a bigger number in the safety stock field. The category strategy guide works through how those segments map to supplier roles.
Grouping parameters by material or geometry feels tidy and produces bad results, because two visually identical rings can sit in completely different demand regimes. Segment on the numbers.
What to actually send the supplier
Forecast sharing is the cheapest lead-time reduction available, and it is routinely done badly — either not at all, or as an unversioned spreadsheet that nobody on the receiving end trusts. A supplier who does not believe your forecast will not buy alloy against it, and the material stage is precisely where the weeks live.
The three-zone horizon
The frozen window should be at least as long as the supplier's material procurement stage. If alloy takes six weeks to secure and your firm window is four, the supplier is either carrying speculative inventory for you or quietly extending your lead time, and one of those is being priced into your part.
The bullwhip is mostly self-inflicted
Order variability amplifies at every step up the chain. On magnets there are typically four steps — you, the magnet producer, the alloy supplier, the separation plant — so a modest wobble in your releases arrives at the material stage as a violent one. The amplifiers are familiar and all of them are yours to remove: batching releases monthly when you consume weekly, ordering ahead of an announced price increase, inflating orders during allocation, and reacting to a single bad month with a parameter change.
The last is the most damaging and the least noticed. Safety stock parameters that get adjusted every time somebody is surprised produce an order signal that is pure noise. Change parameters on a schedule, against evidence, not in response to individual events.
Where the relationship supports it, share actual usage alongside purchase orders. A supplier who can see that your consumption is flat while your orders are lumpy will plan against the consumption and stop treating your order pattern as demand. This is the single highest-value piece of information you can give a magnet supplier, and it costs nothing.
Reviewing the parameters before they drift
Every planning parameter is a snapshot of conditions that have since changed. The failure mode is not setting them wrong initially; it is setting them correctly in a calm quarter and leaving them there through a tariff change, a supplier consolidation and a grade substitution.
| Trigger | Review | Because |
|---|---|---|
| Quarterly, scheduled | Lead-time mean and deviation from actual receipts | Drift is gradual and invisible between receipts |
| Two consecutive late receipts | Lead-time deviation, then the reorder point | Two is a pattern; one is weather |
| Buffer not recovering to full | Reorder point — the buffer is masking a low ROP | The classic symptom of an understated lead time |
| Any supplier or plant change | Everything; treat the history as reset | A new plant has a new distribution, not the old one |
| Grade, coating or drawing revision | Lead time and MOQ; possibly requalification | A revision can move the part to a different production route |
| Tariff, export-licence or origin change | Customs stage and total landed cost | Clearance risk sits inside the lead time, not beside it |
| Annual | Segmentation, service levels, holding cost rate | Spend and criticality move; the segments should move with them |
Two indicators are worth watching continuously rather than quarterly. Cover ratio — on-hand divided by average weekly consumption — should track the reorder point; a slow decline across several cycles means demand has grown past the parameter. And the ratio of stockouts to buffer consumption events tells you whether the buffer is genuinely absorbing variance or merely delaying failure.
Safety stock covers variance around a functioning supply chain. It does not cover an export licence being denied, a supplier exiting the grade, or tooling being lost. Those are discontinuities, and the response is a qualified second source, a last-time buy, or both — see supply risk monitoring, second-source qualification and obsolescence planning. Sizing a buffer against a discontinuity produces a number nobody will approve and that would not have helped anyway.
Finally, remember that the buffer does not have to sit on your balance sheet. A distributor holding domestic inventory converts your offshore lead-time distribution into a domestic one, which is precisely the variable that dominated the safety stock calculation above. That is usually a cheaper way to buy service level than working capital is.