Bottom-Up Should-Cost Model Construction and Governance

What is a bottom-up should-cost model?

Start with a defined purchase unit: a finished component to an agreed drawing, a service hour at a specified skill level, or a delivery route with known service conditions. Estimate the resources required for that unit, not the incumbent price. The model can be a spreadsheet, but each cell needs a unit, formula, source and owner. A price comparison alone asks whether an offer resembles other offers; the FAR definition of cost analysis asks whether separate cost elements and profit support a reasonable estimate.

That regulatory distinction is useful as an analytical model outside federal procurement, without importing its legal obligations.

Choose the boundary before calculating: include the supplier’s work up to the agreed delivery point, then keep buyer-side freight, duties, inspection or switching costs in a separate total-cost view. A blended number that quietly adds buyer costs to a supplier quote cannot explain the supplier’s own economics. For a service, replace a material bill with role hours, utilization and delivery infrastructure; keep the same discipline about units and assumptions.

For another way to frame the supplier conversation, see the procurement negotiation strategy. For category-level input choices, use the indirect spend category strategy alongside this unit model.

Which cost drivers should the template hold?

Make the first pass deliberately small: direct material quantity and unit input price; direct labor or machine time and a rate defined to exclude any overhead added separately; scrap or yield; a documented overhead allocation; and a separately labeled margin assumption. Capture a source date, currency, geography and specification for each driver. The DLA pricing guidance asks its own buyers to choose market-basket items reflecting scope and cost drivers. That agency example reinforces the need to match inputs to the purchase, but does not prescribe a private-sector model.

Here is a hypothetical calculation, not a market benchmark. Assume one kilogram of net material incorporated per accepted unit, an 80% good-output yield and an illustrative input price of $8 per kilogram: material is (1 ÷ 0.8) × $8 = $10 per good unit. Assume 0.4 conversion hours at a direct $30-per-hour rate excluding the separate overhead pool ($12), plus overhead at 25% of conversion cost ($3). Resource cost is $25; an illustrative 10% markup on that cost adds $2.50, producing a modeled price of $27.50.

Suppose the supplier quotes $31 for the same unit and terms: the difference from the $27.50 base case is $3.50, a question to investigate rather than a finding of overcharging. Holding other inputs fixed, a 70% yield makes material about $11.43 and modeled price about $29.07; instead, 0.5 conversion hours makes modeled price about $31.63. Those separate scenarios show which assumption could explain the quote; they do not establish the supplier’s actual yield, labor time, overhead or markup.

How should external indices enter the estimate?

For a US iron or steel input, the BLS iron and steel Producer Price Index, WPU101 is a monthly example of a material-price trend series. For a manufacturing labor trend, the BLS manufacturing wages and salaries Employment Cost Index, CIS1023000000000I is a quarterly example. Both are index numbers rather than the supplier’s invoice price or loaded hourly rate; the labor series covers wages and salaries, not every element of employer labor cost. Use either only when geography, material or role, specification and observation period fit the question being tested.

Record the exact series code and month or quarter alongside the model input it informs. Do not multiply a current index level by a material quantity or labor hour as though it were a dollar rate; use a separately evidenced base rate and, where defensible, the index’s change over time to test an update.

Keep the source selection review separate from the formula. A specialist should check whether the process route, scrap allowance and capacity assumption fit the specification. A category manager should check contractual delivery and volume terms; finance should check the allocation basis. The DLA directive describes technical support from product and engineering analysts for negotiation.

A private team can assign different reviewers; the transferable practice is to record who checked each technical and commercial assumption.

How can a team use the model in a supplier conversation?

Bring a short variance bridge: quote, modeled base case, and the effect of each uncertain driver. In the hypothetical example above, the $31 quote exceeds the $27.50 base by $3.50; changing only conversion time from 0.4 to 0.5 hours moves the estimate to about $31.63. Ask which assumption the supplier contests first. A different grade of material, batch size, quality requirement, warranty exposure or required capacity reservation can explain a gap without either side being wrong.

Avoid asking for a full cost book when a narrower technical question will decide the next step.

In the accounting-information study, the researchers collected interviews and questionnaires within an Egyptian textile supply chain and reported that mistrust and differing systems complicated formal sharing. Its context is narrow, but it illustrates why a model should invite explanation without presuming open-book access. The study does not show that a particular negotiation script saves money or that all suppliers will disclose their costs.

Agree what decision the updated model will support: request another quote, change a specification, test a different volume commitment, seek technical evidence, or accept the offered price with a documented reason. Freeze the model version used for that decision. In a competitive evaluation, safeguard each supplier’s confidential inputs and apply the decision criteria the buyer has set for that process.

What should happen after the negotiation?

Assign each input a review trigger rather than refreshing every cell on the same date. Revisit material inputs when a relevant series changes or a specification changes; revisit labor and overhead assumptions when process routing, capacity or volume changes. Retain prior versions so the team can distinguish a real driver movement from a new modeling choice. A change log should name the editor, evidence and effect on the estimate.

A model is most credible when it is easy to falsify. Ask engineering to identify the largest missing operation, ask finance whether a cost pool is double counted, and ask the supplier which assumption conflicts with its process. The version after those challenges is more useful than a polished first estimate with hidden cells. If the result remains too sensitive to an unobservable input, present a range and say what evidence would narrow it.

How should AI and human reviewers divide the work?

What are the limits of this method?