Should-Cost Modeling with AI: Procurement Guide

What Is Should-Cost Modelling?

Should-cost modelling answers a fundamental procurement question: Given the raw materials, labour, and processes required to make this part, what should it actually cost?

A supplier quotes 45 per unit for a machined steel bracket. Is that fair? Should-cost modelling builds a bottom-up estimate:

• Raw steel cost: 6 kg at 450/tonne = 2.70
• Labour: 1.5 hours at 25/hour = 37.50
• Machine time: 0.5 hours at 80/hour = 40
• Overhead allocation: 15% of direct costs = 8.10
• Standard margin: 12% = 6.30
• Should-cost estimate: 40.60

The supplier quote of 45 is 11% above the should-cost estimate. This signals either the supplier is overcharging, or they have cost pressures we don't understand. Either way, it's a signal to negotiate or explore alternatives.

Should-cost modelling has been around for decades, but AI is transforming it by making models faster, more accurate, and able to handle complexity that manual spreadsheets struggle with.

How AI-Powered Should-Cost Works

Commodity Price Integration

Traditional should-cost models use static commodity prices. You update them quarterly, and they become stale. AI models update commodity prices daily from market data. When steel prices move 5%, your should-cost estimates automatically adjust.

Learning from Benchmarks

AI models learn cost structures from supplier benchmarks. If 100 similar suppliers produce this part at 38-42 per unit, and a new supplier quotes 50, that's a major red flag. The model learns typical cost distributions and flags outliers.

Handling Complexity

Manual should-cost spreadsheets struggle with complexity. How does changing part design affect cost? What if we switch from steel to aluminum? Traditional models require manual recalculation. AI models handle these scenarios instantly through parametric relationships.

Volume and Scale Effects

Should-cost is volume-dependent. A supplier might quote 50 per unit at 10K volume, but 38 per unit at 100K volume because of better machine utilization and purchasing power. AI models learn these relationships and price appropriately by volume.

Building a Should-Cost Model

Step 1: Define the part structure — Material type, weight, complexity, labour hours, equipment hours.

Step 2: Source commodity costs — Current prices for raw materials (steel, aluminum, copper, etc.). Most platforms pull this real-time from market data.

Step 3: Establish cost drivers — Labour rates, machine hour rates, overhead percentages, standard margins. These come from supplier interviews and industry benchmarks.

Step 4: Build the model — Combine materials + labour + overhead + margin into a total cost estimate. Test against actual supplier quotes to validate.

Step 5: Validate and refine — Compare should-cost estimates to actual supplier prices achieved. If they're off by 20%+, refine the cost driver assumptions.

Accuracy: What to Expect

AI should-cost models typically achieve:

• Standard components (90-94% accuracy): Bolts, fasteners, standard materials where processes are well-understood.
• Engineered components (85-90% accuracy): Custom brackets, specialized shapes, unique tooling. More unknowns about actual costs.
• Complex assemblies (80-85% accuracy): Multi-part assemblies with specialized processes. Harder to predict all cost elements.
• Highly specialized parts (75-82% accuracy): Aerospace parts, medical devices, components requiring specialized certification. Unique cost factors hard to model.

The key insight: should-cost is most useful as a negotiation support tool, not as absolute truth. If your should-cost says 38 and a supplier quotes 45, that's a conversation starter. It doesn't mean the supplier is lying; they may have constraints or cost factors you don't see.

Supplier Benchmarking

One of the most powerful uses of should-cost is benchmarking suppliers. When you're renewing a contract with Supplier A, what should their price be?

• Should-cost estimate: 38
• Supplier A current quote: 42
• Supplier B quote (new supplier): 39
• Supplier C quote: 41
• Industry benchmark median: 40

This data gives you powerful negotiation leverage. You can say: "Your quote is 42. Our analysis shows should-cost is 38, competitors are offering 39, and industry median is 40. We need you to come to 40 to stay competitive." Most suppliers will respond with a competitive quote if your data is credible.

Build vs Buy Decisions

Should-cost modelling supports make-vs-buy decisions. If you're considering insourcing a part currently purchased from suppliers:

• Supplier should-cost: 38
• Your internal should-cost (including overhead): 42
• Your internal cost + capital investment: 48

This tells you insourcing doesn't make financial sense. But if your internal cost is 35, it signals you should make the part internally.

Integration with Sourcing Platforms

Leading sourcing platforms (LevaData, Coupa, Jaggr) all include should-cost capabilities. These integrate with your ERP to pull BOM data and automatically build should-cost models for your components. The platform also maintains supplier benchmarks, so you get both should-cost and competitive quotes in one place.

Limitations and When to Be Careful

• Specialty materials: For proprietary or scarce materials, supplier costs may be much higher than commodity prices suggest.
• Custom processes: If a supplier has unique process capability, their costs may be legitimately higher.
• Quality and reliability: A supplier with better delivery performance or quality may deserve a price premium. Should-cost doesn't capture these factors.
• Relationship value: Sometimes paying a premium to a reliable supplier is worth the cost for supply security.
• Volume assumptions: If your volume assumption in the model is wrong, the should-cost estimate is wrong. Always validate volume expectations.

Implementation and ROI

Most companies see 2-4% cost savings from AI should-cost modelling, because suppliers realize you have better pricing data and are more willing to negotiate. Combined with better commodity price forecasting and strategic sourcing, the total savings can reach 5-8%.

Implementation requires clean BOM data and supplier cost structure knowledge. For most companies, this takes 3-4 months to get 80% of your direct materials covered.

Conclusion

AI-powered should-cost modelling gives procurement teams accurate, automated, real-time cost estimates that support better negotiations and strategic sourcing decisions. Combined with commodity price forecasting and supplier benchmarking, should-cost is a core capability for modern direct materials procurement.