CBP Indefinitely Suspends $800 De Minimis Duty Exemption

The Analysis

From 24 June, a parcel worth eight hundred dollars or less stopped walking in the side door and started having to be filed. CBP published an interim final rule in the Federal Register on 24 June 2026 indefinitely suspending the de minimis administrative exemption for merchandise valued at $800 or less arriving through every mode other than the international postal network, and the rule is effective on the date of publication. There was no runway. Shipments tendered on 23 June under one set of expectations were in a different world on the 24th, and if you run a direct-to-consumer book into the United States by air, by ocean consolidation or by truck from Canada or Mexico, what changed overnight was not the freight rate. It was the paperwork, and the paperwork is where the money is.

The abstract is short and it says the thing plainly: entries of merchandise valued at $800 or less arriving through all modes other than the international postal network must use formal or informal entry procedures. De minimis was never a discount. It was a procedure, and the procedure is what has gone. The international postal network is temporarily spared and sits under a separate flat-rate structure, which means there is now a second track with its own arithmetic, its own paperwork, and an expiry date nobody has told you.

Comments are due on or before 24 July 2026. I write that date down not because you are going to file comments, although you may, but because an interim final rule with a comment window is a document that can still move. Whatever you do between now and then should be done in a way that leaves you able to go back over your own files. That is a records argument, and it is the part that gets skipped while everybody is busy chasing entries.

Start with one parcel, because that is the unit this rule actually bites. A consumer in Ohio orders a jacket from a brand that ships out of a consolidation point in Guangdong. Declared value forty-five dollars. Before 24 June that parcel moved on a manifest with a handful of fields and cleared with minimal intervention. From 24 June the same parcel has to travel as an entry. Somebody has to stand behind a tariff number, a country of origin, a transaction value and a consignee, and somebody has to do it in a form the entry system will accept.

Here is the thing I find myself repeating this week: the tariff number comes first and the relief question comes second. Look at the classification, then look at whether anything reduces or defers what that classification produces. Every file I have watched go wrong on a change like this went wrong because somebody asked the second question first. They asked whether the shipment qualified for something, got a provisional yes, built the flow around it, and only later found that the classification underneath was wrong. Then the relief did not apply, and by then the goods had moved and the amendment was the expensive kind.

That ordering is not pedantry, it is arithmetic. The classification sets the duty rate. Everything else you might claim sits on top of the rate rather than instead of it. If the number underneath is wrong, whatever sits on top of it is wrong in the same direction, and on a book of four thousand parcels a month you have not made one mistake, you have made four thousand.

So before anybody argues about anything else, establish who produces the classification and when. On a direct-to-consumer file the commercial data, meaning description, quantity, transaction value, country of origin and tariff number, originates with the seller or the exporter abroad. The transport data, meaning who is carrying it, under what reference and when it crosses, sits with the carrier or the consolidator. The entry data, meaning the importer of record identity, the bond, the entry type and the declaration itself, sits with you or with whoever files for you. Three organisations, three clocks, three document sets. A de minimis flow lets those three drift apart because nothing forces them into the same moment. An entry flow does not. An entry is one object assembled from three sources, and it fails as a whole if any one of them is late or wrong.

The practical consequence is that the cut-off moves earlier, and it moves to a place you do not control. Under the old flow you found out the data was bad when somebody downstream complained, and you fixed it at leisure. Under an entry flow you find out when the entry is rejected or the shipment is held, and both of those happen at the airport or the port, on the clock, with storage running. I would rather spend three minutes on a data set than three days on a hold, and most of what follows is that preference written out at length.

Now the time account on one parcel, because this is where the day is won or lost. Assume the order is placed at T zero. Assume your own system can populate value, description, quantity, origin and consignee within one hour, and assume the tariff number has to come from the supplier and takes twelve hours. Assume the parcel is tendered to the carrier four hours later, flies for two days, and lands. Assume the entry has to be filed at or before arrival. Assume a clean entry releases in six hours, and assume a held entry waits for a response from whoever owns the data. The difference between those two is not the six hours. It is whatever the supplier's response time turns out to be, and on most books that number is measured in days and has never been written down.

Write it down. That is the single most useful thing you can do this week: for every lane, the name of the person who supplies the tariff number, the clock on their response, and what you do when the clock runs out. If you cannot fill in those three fields you do not have a data flow, you have a hope, and hope is not a control.

Let me put money on it, with the assumptions somewhere you can check them. Assume a direct-to-consumer brand bringing four thousand parcels a month into the United States, average declared value forty-five dollars, so one hundred and eighty thousand dollars of goods value a month. Assume the average applicable duty rate across the assortment is twelve per cent. That is an assumption, not a finding, and you should replace it with your own lines. Twenty-one thousand six hundred dollars a month of duty that did not exist in May.

Then the part people forget. Assume a per-entry handling and filing charge of six dollars and fifty cents, a number you should get from your own filer and not from me. Four thousand entries a month at that rate is twenty-six thousand dollars. Assume your team spends three minutes per parcel assembling and checking the data set, at a loaded cost of sixty dollars an hour, so three dollars a parcel, twelve thousand dollars a month. Duty plus filing plus internal effort comes to fifty-nine thousand six hundred dollars a month. Against a goods value of one hundred and eighty thousand dollars that is roughly thirty-three per cent of what the goods are worth, landing on a business that was priced on the assumption that the border cost nothing.

Compare it with the old flow. Assume half a minute a parcel for data, fifty cents, two thousand dollars a month, no duty, minimal filing. The delta is about fifty-seven thousand six hundred dollars a month. If your contribution margin on that book is twenty-five per cent you were earning forty-five thousand dollars a month on the volume, and the rule has taken more than the entire margin. That is the sentence I would put in front of whoever owns the profit and loss, and I would put it there this week rather than at quarter end.

If the number is that large, the first question is not how to file. It is whether some of the volume belongs on the postal track at all. The international postal network is temporarily spared and runs under a separate flat-rate structure. Assume forty per cent of your parcels can be routed that way, one thousand six hundred a month. Assume the flat rate works out at eight dollars a parcel, again checked against your own quotes. That is twelve thousand eight hundred dollars. The same one thousand six hundred parcels on the standard track cost fourteen dollars ninety cents each on the assumptions above, or twenty-three thousand eight hundred and forty dollars. The difference is a little over eleven thousand dollars a month.

Assume the postal track is five days slower, and assume that slowness costs three percentage points of repeat purchase on that cohort. Assume each repeat order is worth forty-five dollars of revenue and eleven dollars of contribution. One thousand six hundred orders, three per cent, eleven dollars: about five hundred and twenty-eight dollars a month. On these assumptions the postal route is worth roughly eleven thousand a month and costs something in the region of five hundred. Even if I am wrong by a factor of three on the service cost the trade still stands, and it stands by a margin wide enough that you do not need to be precise to take it.

What it is not is permanent. Temporarily spared means spared until somebody decides otherwise, and a flat-rate structure is a policy choice that can be repriced without much notice. So treat the postal track as an option with an expiry date you cannot see, which changes how you build it. Keep the data set identical on both tracks. Keep the same records. Keep the standard-track capability warm on at least twenty per cent of volume, so that when the second track closes you are switching a percentage rather than relearning a procedure.

There is a second alternative, and it is the one that actually changes the arithmetic. Stop shipping parcels and start importing inventory. Assume the same four thousand units a month, but bought and shipped in bulk, arriving as four consolidated entries instead of four thousand. Assume the bulk purchase price is eight per cent below the unit price you pay today, so monthly goods value falls from one hundred and eighty thousand to about one hundred and sixty-five thousand six hundred. Assume duty at the same twelve per cent, nineteen thousand eight hundred and seventy-two dollars. Assume four entries cost a hundred and fifty dollars each in filing and handling, six hundred dollars, against twenty-six thousand on the parcel model. Assume data preparation is two hours per entry, eight hours a month at sixty dollars, four hundred and eighty dollars, against twelve thousand.

Assume domestic fulfilment, pick, pack and ship, at four dollars fifty a parcel, eighteen thousand dollars, and assume the international parcel postage you pay today is six dollars twenty a parcel, twenty-four thousand eight hundred dollars. Put it together: duty nineteen thousand eight hundred and seventy-two, filing six hundred, data four hundred and eighty, fulfilment eighteen thousand, so thirty-eight thousand nine hundred and fifty-two dollars a month. Against the parcel model under the new rule, which is fifty-nine thousand six hundred plus twenty-four thousand eight hundred of postage, or eighty-four thousand four hundred, the inventory model saves something like forty-five thousand a month.

It is not free, and the price of it is inventory. Assume you hold eight weeks of cover, so average inventory of about three hundred and thirty-one thousand dollars, and assume an annual carrying cost of twenty per cent covering capital, storage, insurance and shrinkage. That is sixty-six thousand two hundred and forty dollars a year, or five thousand five hundred and twenty a month. Assume one per cent a month of obsolescence and shrinkage on the stock, one thousand six hundred and fifty-six dollars. Net of both, the inventory model is still better than the parcel model under the new rule by roughly thirty-eight thousand dollars a month.

Here is the part I want you to notice, though. Against the old world the inventory model is worse. The old parcel flow cost two thousand dollars of data plus twenty-four thousand eight hundred of postage, twenty-six thousand eight hundred a month. The inventory model costs thirty-eight thousand nine hundred and fifty-two. You can recover most of what the rule took and you cannot recover all of it. Anybody who tells you there is a structure that puts you back where you were in May is selling something.

One more line belongs in the calculation and it is not a customs line at all. If your checkout promises delivered duty paid and your margin model was built on zero duty, that promise has just become a cost, and it needs a number before your commercial team renews anything. I am not going to tell you who bears the duty under your terms of sale; that is a contract question and beyond what I should answer here. What I will say is that the duty is now a real line, it is measurable, and it belongs on the page that decides your prices rather than being discovered later in the variance.

Then there is the bond question, and it is the one that quietly stops filings. An entry needs a bond behind it. If your book was built on de minimis you may not have one in the size and form this volume needs, and getting one is not instant. Price both structures against your next twelve months of volume. Assume forty-eight thousand entries a year. Assume a continuous bond quoted at six hundred dollars a year, about one and a quarter cents an entry. Assume single transaction bonds quoted at eight dollars each, three hundred and eighty-four thousand dollars a year. Those are assumptions and you should replace them with real quotes, but the shape of the answer is not in doubt. The person who should own this is whoever signs for treasury, not whoever packs the boxes.

Set the trigger now rather than after the first bad week. Log every entry pulled for exam. Assume a baseline exam rate of five per cent on a clean data set. If your weekly exam rate on any lane exceeds five per cent for two consecutive weeks, stop tendering on that lane until the data supplier is fixed, because at that point you are paying storage and breaking delivery promises on every parcel in the queue, and the queue does not clear by itself. Five per cent sounds tight. It is meant to be tight, because the cost of a false stop is one lane for a week and the cost of a late stop is the whole book for a month.

The records are the part that protects you later. Four things on every file from now on: the entry number with the filing timestamp, the commercial data version with the supplier and the date it was supplied, every amendment with the reason and the time, and the duty actually paid per line. Sample fifty files a month and read them, do not just store them. The reason is the comment deadline. An interim final rule with comments due 24 July 2026 can come back different. If it does, the only way you will know which of your entries need another look is if your files say what you declared, when, and on whose data.

The pitfalls are the ordinary ones, which is what makes them dangerous. Amending late is one of them. Once an entry is filed, a correction is a procedure with its own timeline; assume your filer needs twenty-four to forty-eight hours to prepare and submit one, and assume the parcel does not wait politely while that happens. The lesson is unglamorous: get the data right before the entry goes in, because the amendment clock is slower than the parcel. Treating the tariff number as a formality is another. It is not a formality. It is the input that decides the duty and decides whether any relief applies, and a wrong number filed four thousand times a month is four thousand corrections. Treating the postal exemption as a permanent structure is a third. It is described as temporary. Build accordingly.

The one I would flag hardest is the fourth. Do not let anybody on your side answer the legal question on the fly. Whether your particular facts meet the conditions for informal rather than formal entry is not a shipping decision, and the mode-by-mode boundary in this rule is narrower than it looks from a warehouse. The applicability and the exceptions of this federal requirement are Dr. Ingrid Voss's ground, not mine. What I can tell you is what the file has to contain and who has to produce it, and that is where I stay.

There is also a version of this that is an opportunity, and it is smaller than people will tell you. If you are now filing entries on four thousand parcels a month, you have for the first time a clean data set on your own import book: real values, real origins, real classifications, real duty paid per line. That is the input you need to argue about sourcing, to consolidate lanes, and to decide which suppliers are worth their paperwork. It arrived attached to a cost you did not choose, but it is there, and nobody else in your business has it.

What I would do today is short. Freeze the bookings that do not have a complete data set. Get the entry procedure for each lane confirmed in writing. Get the bond priced and in place. Put the exam trigger on a dashboard somebody actually looks at. Put 24 July in the diary, because that is when the comment window closes and the rule stops being a draft in everything but name.

That is not a strategy. It is a week of work, and it is the difference between a book that keeps moving and a book that sits at the airport while somebody argues about a tariff number.

↑ Back to the key points

— By Derek Xu