Every company that imports anything has a landed-cost model somewhere. Usually a spreadsheet, occasionally something wired into the ERP, and almost always built by one person who left two reorganisations ago.
Open it and look at the duty column. If there is one column, and it holds one rate per item, the model is not describing what happens at the border any more. It is describing what used to happen at the border, in a period when a single rate really was the whole answer.
One rate per item was always a simplification
A duty rate is the output of several separate determinations, each of which can move independently. The classification of the goods. Their country of origin for duty purposes. The value on which duty is assessed. Whether any additional programme applies on top of the ordinary rate, and whether more than one does.
Collapse all of that into a single field and you get a number that is right on the day it was entered and slowly stops being right afterwards, without ever announcing that it has.
Most of the public argument about how tariffs feed into inflation and cost plays out at the level of the whole economy. The same mechanics run inside one company's cost sheet, just with less visibility, because nobody publishes a monthly index of your own understated duty.
The classification underneath the number is older than you think
Ask who assigned the HTS code for a given item and the answer is often that the broker did, at first import, from a product description someone wrote in a hurry.
That was reasonable. It also means the code reflects the product as it existed then. Products change. A housing goes from metal to composite, a device gains a battery, a kit starts shipping with an accessory it did not previously include, and any one of those can move the goods into a different subheading under the General Rules of Interpretation. The purchase order number stays the same, so nothing in the cost model flags it.
You see this most clearly in companies with broad catalogues, where a single code has been inherited by a whole family of items that were never individually classified. The family grows. The code does not.
The rules themselves also move. Chapter and section notes are amended, subheadings get created and retired, and the Harmonized System is revised on a cycle that most cost models are entirely unaware of. An item classified correctly under a subheading that no longer exists is not a small clerical issue, because the programme lists that determine additional exposure are written against specific subheadings. The code is what connects a product to a measure. Get the code wrong and every downstream number inherits the error, including the one finance is using to set price.
Dutiable value is not the invoice total
The second quiet gap is valuation. Transaction value starts from the price actually paid or payable, and then a set of statutory additions apply where they exist:
Assists, meaning tooling, dies, moulds or engineering that you provided to the supplier free or at reduced cost
Royalties and licence fees related to the imported goods that the buyer must pay as a condition of sale
Selling commissions incurred by the buyer
Packing costsProceeds of any subsequent resale that accrue to the seller
Almost none of those live in procurement's price file. Tooling sits in a capital project. Royalties sit with legal. The cost model pulls the unit price and stops there, which produces a dutiable value lower than the correct one and a duty estimate to match.
Freight and insurance treatment adds its own confusion, because the Incoterm determines what the invoice price already contains and therefore what has to be added back or taken out. A delivered price and an ex-works price for the same goods are not comparable inputs, and plenty of models mix the two across suppliers without adjusting.
The exposure runs the other way too. First sale for export, where duty is assessed on an earlier sale in a multi-tiered transaction, can lower the dutiable value legitimately when the chain is documented properly. CBP has been issuing first sale valuation questionnaires asking importers to show whether they used it and whether the records exist, within a short fixed response window. A Last Sale Valuation Act was introduced in the Senate in February 2026 that would assess duty on the last sale before importation instead. It is proposed legislation rather than law, and a model that assumes first sale will be available forever is making a bet it has not written down.
Origin assumptions that were true once
Origin is the field most likely to be wrong and least likely to be questioned, because it usually arrives pre-filled from the supplier.
For duty purposes the question is where the goods last underwent substantial transformation, which depends on the specific processing and on the rules that apply to that product. Final assembly somewhere does not automatically make that place the origin. When a supplier shifts a subassembly to a different plant, or brings in a component from a new source, the origin conclusion can change without the supplier mentioning it, and without anything visible on the invoice.
A cost model with origin hardcoded per item will carry that error silently for years.
Stacking is what actually breaks the spreadsheet
A single duty field fails for a structural reason. Programmes apply on top of the ordinary rate rather than replacing it, and more than one can apply to the same entry.
Section 301 exposure is keyed to product and origin. Section 232 exposure reaches metals and, increasingly, derivative products containing them, which means an item that reads as a finished good in your catalogue can still pick up an input-level measure. Chapter 99 provisions sit above the ordinary schedule. A model that stores one number cannot represent two measures landing together, so it either picks one and understates, or somebody manually adds a fudge factor that nobody revisits.
The fix is not a bigger number. It is a duty structure with one line per applicable programme, each traceable to the code and origin that triggered it.
Not everything charged at the border is a tariff
Entry summaries carry more than duty. Merchandise processing and harbour maintenance fees are assessed under their own authority. Excise applies to certain goods. Antidumping and countervailing duties are separate determinations with their own scope and their own cash deposit mechanics, and they are the single most common source of a landed cost that turns out to be wildly wrong after the fact.
A fair amount of internal confusion clears up once someone in the room can articulate the difference between import tax and tariff categories, because a charge collected on entry is not automatically a tariff and it does not behave like one when policy shifts.
February 2026 changed which lines are durable, not the total
The Supreme Court held in February 2026 that IEEPA does not authorise the President to impose tariffs of indefinite scope, striking down the IEEPA tariffs. The Court did not order refunds. The Court of International Trade later ordered refunds to importers of record, the government appealed, and CBP stood up a phased refund process while that appeal was still live.
Section 301 and Section 232 rest on different statutes and were not disturbed by the ruling. So an importer whose blended rate was mostly those programmes saw very little change, while one whose exposure ran through IEEPA saw a great deal. A model holding a blended rate cannot tell you which of those you are, which is the point at which the modelling shortcut stops being a shortcut and becomes a blind spot.
Understated duty is a retroactive problem
The awkward part is that a duty estimate is not just a planning artefact. The declaration behind it is a legal statement, and if the model was understating because the code or the value was wrong, the filing was wrong too.
There is an ordinary route for that. A post summary correction before liquidation, a protest after it. Describing the mechanism is easy; the hard part is that both depend on knowing the entry was wrong while the entry is still live, and a cost model that nobody reconciles against actual entry summaries will not tell you.
Reconciliation is the unglamorous fix. Pull what you actually paid from the entry data, compare it against what the model predicted, and look at the items where the two disagree by more than rounding. The disagreements are where the real classification and valuation questions are hiding, and they tend to be the same items every quarter, which is its own kind of useful signal.
Nobody enjoys that exercise. It surfaces things you would rather not have found, usually in the week before a close. But an importer who runs it knows the shape of their own exposure, and one who does not is quoting margin off a number that has been quietly drifting since whoever built the spreadsheet stopped maintaining it.
