Model landed cost as three scenarios per product: the current rate stack, a moderate increase of 10 to 25 percentage points on the most exposed categories, and a worst case combining rate spikes with the loss of any preferential claims. Build the model at the SKU or category level from your actual HTS classifications, and keep the duty component separate from freight and fees so rate changes flow through cleanly. The output is not a prediction; it is a decision tool showing which products stay profitable under stress and which need action now.
Why single-point estimates fail
Most landed cost models bake in today's duty rate as a fixed input, which makes the model precisely wrong the moment policy moves. The failure mode is silent: the model keeps producing confident numbers while reality diverges, and pricing decisions made on stale costs bleed margin for a quarter before anyone notices. Tariff volatility turned landed cost from an accounting exercise into a risk exercise, and risk needs ranges, not points.
The three scenarios are not forecasts of what will happen; they are boundaries of what you must survive. Current rates tell you where you stand. The moderate increase, calibrated to recent policy moves in your categories, tells you what normal volatility costs. The worst case tells you which products have no margin of safety. Decisions come from comparing the three, not from picking the middle one.
Building the model
Start with clean classifications, because every scenario multiplies the HTS rate. If your classifications are approximate, the scenarios amplify the error. List each product's current duty stack: MFN rate, any Section 301 or other additional duties, and preferential claims like free trade agreements with their qualification status. The worst-case scenario should assume the preferential claim fails, since claims are the first casualty of messy documentation.
Keep the model modular: product cost, freight, insurance, duties by component, MPF, HMF, and brokerage fees as separate line items. When a rate changes, you update one cell and the scenarios recompute. A model where duties are a blended percentage cannot do this and will be rebuilt from scratch every time policy moves, which is exactly when you have no time to rebuild it.
Reading the output
The key output is the margin-at-risk table: products ranked by how much margin they lose between the current and worst-case scenarios. The top of that list is your action list. For each exposed product, the options are pricing (can the market bear an increase), sourcing (alternative origins with better rates), engineering (design changes that shift classification), and acceptance (some products are worth running thin). The model does not make the choice; it sizes the stakes.
Watch the scenario spread, not just the levels. A product with stable costs across scenarios is strategically valuable even at lower margin, because it is plannable. A product with wild swings is a liability even at good current margins. Portfolio thinking beats product thinking here: the goal is a catalog whose blended margin survives the worst case, not individual heroes that collapse under stress.
Keeping it current
A scenario model decays fast. Assign an owner, set a monthly refresh cadence, and trigger ad-hoc updates on policy announcements, not after the quarter closes. The refresh is cheap if the model is modular: update the rate tables, re-run, review the margin-at-risk ranking. The expensive version is rebuilding, which is why modularity was the whole point.
Archive each month's run. The history shows whether your exposure is growing or shrinking, which is the trend leadership actually needs. A single snapshot says where you are; the series says whether your sourcing and pricing actions are working. And when the worst case arrives, the archive proves the planning was real, which matters more than anyone admits.
How do you pick the moderate increase level?
Look at the largest single-year moves in your categories over the past five years and use something in that range. The scenario should feel uncomfortably plausible, not apocalyptic.
Should the model include currency risk?
Ideally yes, as a separate scenario dimension. Currency and tariffs compound, but modeling them together obscures which lever is hurting you.
Who should own the model?
Whoever owns pricing or sourcing decisions, not finance alone. The model is a decision tool; its owner should be the person who acts on it.