The difference between a buyer who negotiates and one who accepts is usually not skill. It's that one of them knew roughly what the part should cost before the quote arrived.
You can get within about 15% with four terms and no costing software.
The model
should_cost = material + machining + setup_per_piece + margin
material = weight_kg × price_per_kg × (1 + scrap_rate)
machining = (cycle_minutes / 60) × shop_rate_per_hour
setup_per_piece = setup_hours × shop_rate_per_hour / batch_qty
margin = (material + machining + setup_per_piece) × markup
Four inputs you have to estimate. Three of them are easier than they look.
Worked: a 6061 aluminium bracket, 500 off
Material. Finished part 0.42 kg, starting billet 0.68 kg. 6061 at $4.60/kg, and machining scrap is typically 5–8%:
0.68 × 4.60 × 1.06 = $3.32
Machining. Roughly 4.5 minutes of cycle time at a $75/hour shop rate:
(4.5 / 60) × 75 = $5.63
Setup. 1.5 hours of programming and fixturing, amortised over the batch:
1.5 × 75 / 500 = $0.23
Margin. Job shops typically run 25–35% on top:
(3.32 + 5.63 + 0.23) × 0.30 = $2.75
Should-cost ≈ $11.93.
Against the real quotes from our spread example — $8.40, $11.10, $14.90 — the model lands almost exactly on the middle quote. That's the point: it doesn't tell you the price, it tells you which quotes need explaining.
The term that carries all the uncertainty
Cycle time. Material weight comes off the model, scrap rate and markup sit in narrow bands, but four minutes versus seven is a 50% swing in the largest controllable term.
The shortcut is not to estimate it in the abstract but to derive shop rate from parts you've already bought. Rearranged:
implied_shop_rate = (quoted_price − material − margin_share)
/ (cycle_hours + setup_hours/qty)
Run that across a dozen past awards and you'll find your suppliers cluster tightly — a real number for your suppliers, in your region, far better than any published average. After that, should-cost is arithmetic.
What it's actually for
Not beating suppliers down. A quote 15% under should-cost is as informative as one 40% over — it usually means the shop misread the spec, and that's a delivery risk you'd rather find now.
What it buys you is the ability to ask a specific question. "That's higher than I expected" invites a shrug. "I'm modelling about four and a half minutes of cycle time — where am I wrong?" gets you an answer, and often the answer is a design change that removes an operation.
How this gets solved
The model is only as good as its calibration, and calibration means quote history you can actually query: what did this class of part cost, from which supplier, at which quantity, over the last two years.
That history is the by-product RFQTrace keeps by default. Every quote stays attached to its RFQ with its line items, quantity and date — so the shop rate you back out is derived from your own awards rather than a benchmark from someone else's industry.