Why should-cost beats price benchmarking
Most magnet negotiations stall in the same place. The supplier says raw material went up. The buyer says the increase looks excessive. Neither side can prove anything, because neither is working from a decomposed cost. The conversation becomes a test of stubbornness rather than arithmetic.
A should-cost model breaks that deadlock. Instead of arguing about the total, you argue about the inputs — and inputs are checkable. If a supplier claims a 20% increase on a part that is 55% alloy by cost, the implied move in alloy price is roughly 36%. Either the index supports that or it does not.
Benchmarking still has a place, but it has two structural weaknesses for magnets. First, magnets are rarely comparable: two parts with identical dimensions can differ by 40% in cost because one is an EH grade with heavy rare-earth content and the other is a standard N-series. Second, benchmark data ages badly in a volatile market. A price paid nine months ago is a historical curiosity, not a target.
A model is good enough when it lands within roughly ±15% of quotes you already trust, using public indices and no supplier-confidential data. You are not trying to reconstruct the supplier's P&L. You are trying to know which questions to ask.
The cost stack, oxide to finished part
A sintered NdFeB magnet passes through a long chain, and cost accumulates at every step. Modeling it as a single blended dollars-per-kilogram number hides exactly the information you need.
Five of those steps scale with mass. Four scale with piece count. One — machining — scales with geometry complexity. That split is the backbone of the model, because it tells you which cost lines respond to volume and which respond only to design change.
Step one: finished mass and buy-weight
Start with what the part weighs, then with how much alloy had to be consumed to produce it. These are not the same number, and the gap between them is often the single largest recoverable cost in the part.
Sintered NdFeB has a density of approximately 7.5 g/cm³. Compute the finished volume from the drawing, multiply, and you have finished mass. Buy-weight is the mass of the sintered blank that had to be produced — and for cored or profiled parts it can be more than double the finished mass.
finished volume = π/4 × (25² − 15²) × 10 = 3,142 mm³
finished mass = 3.142 cm³ × 7.5 g/cm³ = 23.6 g
// blank is a solid disc, later cored and ground
blank 27 mm Ø × 12 mm → 6,870 mm³ → 51.5 g
material utilization = 23.6 / 51.5 = 46%
machining loss = 27.9 g per piece
Forty-six percent utilization is normal for a cored ring and is not, by itself, evidence of a bad supplier. But it does mean that every dollar of alloy price increase hits this part roughly twice as hard as it hits a simple pressed block. Two parts on the same index can have very different sensitivity.
Machining swarf and cored slugs have recoverable value. A supplier who recovers 70% of the loss at 40% of alloy value is returning real money to the cost base. If your model ignores that credit, you will consistently over-estimate cost and lose credibility in the negotiation. Ask whether scrap recovery sits in the quoted price or in the supplier's margin.
Step two: alloy cost and the rare-earth basket
Alloy is not one commodity. It is a weighted basket, and the weights change with grade. Model it as a composition, not as a price.
| Element | Typical share of alloy | Behavior | Grade sensitivity |
|---|---|---|---|
| Neodymium / praseodymium | ~29–32 wt% | Dominant cost driver; highly volatile | Roughly constant across N-series |
| Iron | ~64–68 wt% | Cheap, stable, effectively noise | None |
| Boron | ~1 wt% | Low cost, low volatility | None |
| Dysprosium / terbium | 0 to ~4 wt% | Very expensive; supply-constrained | Rises sharply through H → SH → UH → EH |
| Cobalt, aluminium, copper, niobium | ~1–3 wt% combined | Minor but not zero | Varies by producer recipe |
The heavy rare-earth line is where category strategy and engineering meet. Moving from a standard grade to an EH grade can add several percent dysprosium or terbium by weight, and those elements have historically traded at multiples of NdPr. An over-specified temperature rating is therefore not a small conservatism — it is a permanent structural cost, repeated on every piece for the life of the program.
Grain boundary diffusion has changed this calculus materially. It places heavy rare earth only where it is metallurgically useful rather than throughout the bulk, reaching a given coercivity with substantially less dysprosium or terbium. If your supplier offers it and your part is coercivity-driven, it belongs in the model as an alternative line, not as a footnote.
Build the alloy cost as an explicit sum so each term can be challenged independently:
// conversion adder covers oxide→metal reduction and strip casting
// keep it separate — it moves with energy, not with the rare-earth index
Keeping the conversion adder separate matters. When an index spikes, only the first term should move. A supplier passing through the full percentage increase on the entire alloy price is over-recovering, and a decomposed model lets you say so precisely rather than merely suspecting it.
Step three: conversion, machining and finishing
Conversion is everything between alloy and inspected part. It is where the supplier's efficiency actually shows, and where volume leverage genuinely exists.
Two practical rules follow. First, volume discounts land almost entirely in conversion and setup, not in material — so a supplier offering a large break on a material-heavy part is either amortizing tooling or quoting optimistically. Second, tolerance and coating changes are usually easier savings than price concessions, because they change the work content rather than the supplier's margin.
Step four: assemble the model and land the cost
Put the pieces together on the worked example. Every figure below is an illustrative placeholder — the structure is the deliverable, not the numbers. Replace each input with your own index and your own supplier's disclosed rates.
| Line | Basis | Illustrative value | Per piece |
|---|---|---|---|
| Gross alloy | 51.5 g at $58/kg | $58/kg alloy | $2.99 |
| Scrap credit | 27.9 g, 70% recovered at 40% of value | — | −$0.45 |
| Net material | $2.54 | ||
| Press, sinter, heat treat | 51.5 g at $15/kg blank | $15/kg | $0.77 |
| Machining | Core, OD and ID grind | per piece | $0.35 |
| Coating | NiCuNi, barrel | per piece | $0.14 |
| Magnetize, test, pack | Saturation and sampling | per piece | $0.08 |
| Factory cost | $3.88 | ||
| Factory margin | 12% | $0.47 | |
| EXW price | $4.35 | ||
| Freight & insurance | ~4% | $0.17 | |
| CIF value | $4.52 | ||
| Duty | Rate × declared value | varies by origin | + d |
Duty deliberately sits outside the modeled total. Rates and origin treatment change faster than any other line in the stack, they depend on classification and on where substantial transformation occurred, and they are the one input where a wrong assumption produces a confidently wrong answer. Model it as a separate variable and keep it current alongside your country-of-origin position.
The three recurring errors: using finished mass instead of buy-weight, applying a rare-earth percentage move to the whole piece price instead of to the material share, and forgetting that heavy rare-earth content varies by grade so one index cannot cover a mixed portfolio. Each of these systematically distorts the answer in a predictable direction.
Step five: putting the model to work
A model that lives in a spreadsheet nobody opens is an artifact. A model that is attached to your quarterly price review is leverage. The difference is process.
Test every increase against the model first
When an increase arrives, compute the implied input move before responding. Divide the requested percentage by the material cost share to get the implied alloy move, then compare it to the published index. If the implied move exceeds the actual, you have a specific, unemotional question to ask.
Separate cost-driven from margin-driven changes
A supplier facing genuine input inflation deserves a mechanism, not a fight — that is what index-linked pricing is for. A supplier recovering margin under cover of an index deserves a different conversation. The model is how you tell them apart, and it protects good suppliers as much as it challenges opportunistic ones.
Rank your portfolio by sensitivity, not by spend
Sort every part by material cost share and by heavy rare-earth content. The high-share, high-Dy parts are your volatility exposure, and they may not be your largest line items. That ranking is the natural input to a category strategy and to part-level risk scoring.
Send the design savings back to engineering
The model routinely finds that the cheapest available action is not commercial. Relaxing a tolerance, accepting a lower grade validated against the real duty cycle, changing a ring to a segmented assembly to improve utilization — these often beat anything achievable at the negotiating table. Route them through value engineering with the numbers attached, because engineers respond to quantified trade-offs far better than to a request to cut cost.
Update indices monthly, conversion rates and margin assumptions annually or when a supplier's cost base visibly changes, and mass and yield figures whenever a drawing revision lands. A stale model is worse than none, because it will be wrong with authority.