Should-Cost & Programme Estimate

What this tool does

CM8-314 builds the cost of a manufactured item from first principles and then takes it to programme scale. You enter the routing — the material it starts as, each operation it passes through, the processes you buy in and the parts you bolt on — with the rates, times and scrap rates that apply. The tool works out what one unit ought to cost, applies a learning curve across the programme quantity, escalates the result into the money of the year it will actually be spent, and converts a set of three-point ranges into a contingency at the confidence level you fund at.

That chain — unit build-up, learning, escalation, risk — is what separates a should-cost from a quote. A quote is a number somebody else chose. A should-cost is a structured argument, and its value is that every figure traces back to a line you can defend.

Everything runs inside this single file. There is no account, no upload and no network request of any kind. That matters here: a cost model contains your rates, your margins, your supplier prices and your view of a programme's risk, which are among the most commercially sensitive numbers you hold.

Two scales: unit and programme

The register describes one unit. Everything above it is programme scale. Keep the two apart when you read the numbers:

• Unit cost at reference is the build-up in base-year money at whichever unit number your rates describe. It is the number you take to a supplier.
• Programme, base-year applies the learning curve across the whole quantity, still in today's money. It is the number you compare against a competing bid on the same basis.
• Programme, then-year adds escalation. It is the number the finance function needs, because it is what will actually leave the bank.
• P-level funding adds risk-based contingency. It is the number you ask to have approved.

The four kinds of cost line

• Kind — What it is — Fields it uses
• Material — Casting, forging, bar, sheet, resin — Gross and net mass, price per kg, recovery
• Operation — An in-house step on your own plant — Cycle, cavities, setup, rates, efficiency, scrap
• Outsourced — A process you buy — plating, NDT, heat treat — Unit price, quantity, scrap
• Purchased — A bought-in component or sub-assembly — Unit price, quantity

Give every line a routing sequence. It orders the operation detail and, more importantly, determines which scrap rates apply to which lines.

Material and utilisation

Material is costed on what you buy, not on what ends up on the part.

On machined parts, utilisation is often the largest single lever. Turning a 420 kg casting into a 356 kg finished housing is 85% utilisation; on a billet part it is frequently below 40%, and the difference between buying a forging and buying bar can dwarf every cycle-time project on the line.

Effective time, not cycle time

Costing on raw cycle time understates the job, because the machine is not producing every minute you pay for.

Parts per cycle divides the cycle across everything it produces. Efficiency inflates run time to reflect the proportion of planned time the step actually produces — if you have an OEE figure, its availability and performance components are the right starting point. Setup is spread across the batch, and on a low-volume, long-setup operation it is frequently the largest element of conversion cost.

Scrap, yield and the multiplier

When a unit is scrapped at final test, you have lost the material, every operation before it, and the bought-in content already fitted. Treating scrap as a flat percentage uplift on the total understates the loss on late operations and overstates it on early ones.

The tool instead calculates, for every line, how many units must pass through it to ship one good unit:

where s₁…sₙ are the scrap rates of that line and every operation or outsourced process after it. Material carries the scrap of the whole routing; final test carries only its own.

Rolled yield is unforgiving, and that is the point. Ten operations at 2% each is not 2% — it is a rolled yield of 81.7%, and you must start 1.22 units for every one you ship.

Rates and manning

The machine rate is the cost of the machine standing there for an hour excluding the operator: depreciation or lease, floor space, power, maintenance. The labour rate is the fully loaded hourly cost of one operator — gross pay plus employment costs, not the wage.

Keeping them separate matters because they scale differently, and manning is where in-house and supplier cost structures most often diverge. The operators field takes fractions: 0.5 where one person tends two machines, 2 where assembly needs a pair.

Overhead, SG&A and margin

Factory overhead is absorbed on conversion cost only — machine, labour and consumables — and not on material or bought-in parts, because material passing through the factory does not consume supervision, quality and maintenance in proportion to its value. Absorbing overhead on material is the classic way to make a high-material part look artificially expensive and drive exactly the wrong sourcing decision.

Learning curves: Crawford and Wright

Unit cost falls as cumulative experience builds. The relationship is remarkably stable: each time cumulative quantity doubles, cost falls to a fixed percentage of what it was. That percentage is the learning rate — 85% means the 200th unit costs 85% of the 100th.

Two theories share that exponent and disagree about what it applies to. They are not interchangeable, and quoting a learning rate without saying which theory is in force is meaningless.

Crawford, or unit theory — the cost of the nth unit itself follows the curve:

Wright, or cumulative-average theory — the running average over the first n units follows the curve:

Aerospace airframe work has historically leaned on Wright; electronics, machining and much of defence procurement lean on Crawford. Use whichever your historic actuals were fitted with, and when you compare two estimates check first that they are on the same theory.

How much the choice matters depends entirely on where you calibrate, which is not obvious and is worth knowing before an argument about it. Calibrated at the first unit, the two diverge sharply — on a 22,000-unit run at 90%, Crawford returns a total about 18% higher than Wright from the identical percentage. Calibrated at a mature reference unit they converge to almost nothing: at unit 100 the same comparison differs by around 0.1%, because both curves are then anchored to the same point in the middle of the data and the bulk of the units sit far from unit one either way.

Set the rate per line. Touch labour typically lands between 80% and 90%; material around 95–98% as yield and utilisation improve; catalogue purchased parts 98–100%, because your supplier's learning accrues to them unless your contract says otherwise. A line left at 100% does not learn.

The reference unit

Cycle times almost never describe the first unit. They come from current production, at some unit number well into the run. The reference unit setting tells the tool which unit your rates describe, and it back-calculates the first-unit cost from there:

Getting this wrong is one of the largest single errors available in programme estimating. Entering mature rates and leaving the reference unit at 1 tells the tool your first unit is as cheap as your hundredth, and then discounts the whole programme below that — understating a long build substantially. The learning chart marks the reference unit; everything to the left of it is extrapolated backwards and is the least reliable part of the curve.

Escalation and then-year money

Rates in the register are base-year (constant) money. Money actually spent in year five is then-year (nominal) money, and on a long programme the difference is large.

The tool assumes units are delivered evenly across the production span, applies the learning curve to work out which units fall in which year, and escalates each year's work from the base year. Year one is taken as base-year money. At 2.8% over twelve years, the final year's work costs about 31% more in cash than the same work would today.

Never compare a base-year figure with a then-year one. It is the most frequent apples-to-oranges error in programme cost reporting, and it always favours whichever estimate happens to be in constant money.

Discounting

Present value discounts then-year cash back to today, so that spend in year ten is not weighted equally with spend in year one.

Discount nominal cash at a nominal rate, which is what this tool does. Discounting then-year money at a real rate double-counts inflation and will make a long programme look considerably cheaper than it is.

Three-point estimating and PERT

A single number is not an estimate; it is a guess with the uncertainty hidden. Every line carries an optimistic and a pessimistic percentage around its most-likely cost, and the tool converts the three points into a mean and a standard deviation using the PERT weighting:

Because the pessimistic tail is nearly always longer than the optimistic one, the PERT mean sits above the most-likely cost. That gap is not padding — it is the arithmetic consequence of asymmetric risk, and it is the honest starting point for contingency.

Ranges should reflect how well the line is known, not a house rule. A firm quote with an index clause might be −6%/+20%; a line priced by analogy from a previous platform might be −15%/+45%; something governed by a standard still in draft might be −20%/+60%. If every line in your model carries the same range, the ranges have not been thought about.

Correlation, and why independence lies

Adding standard deviations in quadrature assumes the lines are independent — that a bad month on material tells you nothing about labour. On a real programme that is false. Commodity prices, exchange rates, wage settlements, a schedule slip and a design change all push many lines the same way at once.

The correlation setting applies a single coefficient ρ across all pairs of lines. At ρ = 0 the lines are independent and the spread is at its narrowest; at ρ = 1 they move in lockstep and the standard deviations add directly. Independence is the assumption that quietly produces the confident, narrow, wrong ranges that make cost risk analysis disreputable. Something in the range of 20–40% is a common working assumption; if you have fitted correlations from historic programme data, use those instead.

Confidence, contingency and reserve

The tool converts the mean and standard deviation into a funding level using a normal approximation:

Contingency covers what you know can vary: the quantified spread of things already in the estimate. It belongs to the project manager and it is drawn down as uncertainty resolves.

Management reserve covers what is not in the estimate at all — scope that has not been identified yet. It sits above the confidence level, it belongs to the sponsor rather than the project, and releasing it should require a decision. Merging the two is how programmes quietly consume their risk cover before the risks arrive.

Estimate classes

An estimate's achievable accuracy is set by how well the scope is defined, not by how much effort went into the arithmetic. The class setting follows the five-level structure used in cost engineering practice — from Class 5 concept screening through to Class 1 firm bid — and reports the accuracy range that class is ordinarily expected to achieve.

• Class — Purpose — Scope defined — Typical accuracy
• Class 5 — Concept screening — 0–2% — −30% / +50%
• Class 4 — Study or feasibility — 1–15% — −20% / +30%
• Class 3 — Budget authorisation — 10–40% — −15% / +20%
• Class 2 — Control estimate or bid — 30–75% — −10% / +15%
• Class 1 — Check estimate or firm bid — 65–100% — −5% / +10%

These ranges are indicative. Published class systems give a range of ranges, and they differ by industry — process plant, building and manufacturing do not share the same expectations. Use your own organisation's bands where you have them.

The class tile compares the spread your own ranges produce against the band the class expects, and says so when your model is markedly narrower. That is worth attention: a Class 3 estimate that computes to ±6% is not a precise estimate, it is an over-confident one, and it usually means the ranges were filled in with a default rather than a judgement.

Evidence and basis

Every line records how its numbers were arrived at — quoted, measured, parametric, analogy or judgement. Judgement lines are marked in the register, and the tornado chart colours each line by its basis.

This is not bureaucracy. The first thing a competent counterparty does with your model is find the weakest line and argue about that one. Knowing which lines are soft, and which of those actually move the answer, tells you where an afternoon of work will buy the most credibility. The tornado chart ranks by σ rather than by cost for exactly that reason: a modest line with a wide range is often a better target than an expensive line you already have firm.

Using the model

Take the build-up, not the total. A single number invites a single counter-number; a breakdown invites a line-by-line discussion, which is where the information is. Expect to be wrong on individual lines — often the most valuable outcome is discovering that a supplier runs four cavities where you assumed one.

Comparing your model to a supplier's price needs care. Their overhead structure, manning, buying power, utilisation and learning position all differ from yours, and a gap is not automatically margin.

Printing and sharing

Print Report produces a report from whatever the current filter shows: header, headline figures, the charts, the full build-up and programme roll-up, the year-by-year profile, the operation detail, the complete routing and your closing notes. Print to PDF to circulate it.

Think before you send it. A model showing your overhead recovery, machine rates, learning assumptions and margin is a description of how your business makes money. Many organisations share the cost structure but strip the margin and risk lines before a model goes outside.

Saving your work

Cost lines, settings and the report header are written to this browser's local storage as you type, and the toolbar shows the time of the last save. That storage belongs to one browser on one computer.

Treat Export .json as the real save — one file containing everything, which Import .json restores anywhere. It is also how you version an estimate: export at each gate and you can show exactly what you believed and when, which is the first thing asked for when a programme is reviewed. Export CSV gives you the routing for spreadsheet work, and the template and CSV import buttons let you build a model from routing data you already hold. Reset asks twice, then erases everything this tool has stored. There is no undo.

Accuracy & disclaimer

This tool calculates from what you enter. It cannot tell whether a cycle time is achievable, whether a machine rate reflects your real cost of ownership, whether a learning rate will be realised, whether a scrap rate is representative, or whether the routing is the one a supplier would actually use.

The risk figures use a normal approximation with a single equicorrelation coefficient. They model cost uncertainty only — not schedule risk, discrete risk events, requirement growth or exchange-rate exposure. On a large programme those frequently dominate. Overhead absorption, escalation treatment, learning conventions and estimate-class bands differ between organisations and countries; reconcile to your own management accounts and estimating standards before quoting from this model.

This is an estimating and negotiation aid. It is not a quotation, not an audited cost, not accounting or investment advice, and not a substitute for your own commercial judgement.