Nearshoring replaces four weeks of ocean transit with a border crossing whose variance is larger than the variance it removed, and the variance lives almost entirely in one leg that nobody times separately. IMMEX puts a customs expiry on every temporary import, generally eighteen months for inputs, which makes pedimento age an inventory attribute alongside shelf life. The 2025 United States tariff orders exempted goods qualifying under USMCA, so origin certification went from worthless to worth the tariff rate on products that had never claimed preference because the duty was already zero. Model the crossing as its own leg with its own distribution before you tune anything else.
The plan that justified the plant has one number in it that turns out to be wrong. Transit from the new site in Guanajuato to the customer's distribution centre outside Chicago went in at four days, which is what the carrier quoted. Six months into production the measured door to door is running between five and eleven days, the spread comes almost entirely from a few hundred metres of bridge at Laredo, and the safety stock sized for a four day lead time with no variance is why the customer has started asking questions about the ramp.
This is the ordinary shape of a nearshoring miss. The business case was right about average transit. What moved was the variance, into a leg that no system in the company measures.
The crossing is a separate leg and it owns the variance
A truck leaving a plant in the Bajio for a US customer is usually three tractors and two yards. A Mexican long haul carrier runs to a yard on the Mexican side, a transfer carrier licensed to work the bridge takes the trailer across, and a US carrier picks it up in Laredo or El Paso for the linehaul. Each handoff has a queue, and the customs work sits underneath all of it: an export pedimento filed by the Mexican broker, a US entry filed on the other side, and the possibility that either side asks to look inside.
The Mexican selection mechanism is worth understanding because it is deliberately random. A shipment presented at the aduana draws either a green light and clears, or a red light and goes to physical inspection, and the draw does not respond to how well you prepared the file. It adds a long right tail to an otherwise tight process, which is the shape a safety stock formula built on a fixed lead time handles worst.
On top of that sit the closures. US Customs and Border Protection has more than once redeployed officers away from commercial processing and suspended operations at specific ports of entry, including the rail crossings at Eagle Pass and El Paso in December 2023. A network with one crossing in it has no answer to those weeks.
Trusted trader status helps in a way that is often described wrongly. C-TPAT on the US side and OEA on the Mexican side, with FAST lanes for qualifying moves, reduce average processing, and the larger benefit is to the tail: fewer inspections, dedicated lanes during congestion, faster recovery after a closure. If you size buffers off a distribution, a programme that cuts the ninety-fifth percentile is worth more than one that cuts the mean.
The practical change is to stop modelling plant to customer as one lead time. Time three legs: yard out in Mexico, yard in on the US side, delivery. Only the middle leg needs a distribution fitted to it, and it is the only one you can fix. The mechanics of treating lead time as a distribution are I3's, so the point here is narrower: you cannot fit a distribution to a leg you have never timestamped, and those timestamps usually sit in three carriers' systems and a broker's, joined by nobody.
Rail changes the arithmetic rather than the problem. Canadian Pacific Kansas City, formed when the Surface Transportation Board approved the combination in March 2023, runs a single line railway from Mexico into Canada and interchanges at Laredo without a drayage handoff. Transit is longer than truck and considerably less variable, which for a make to stock item is often the better trade, though a plant feeding a sequenced line will still want the truck.
IMMEX puts a customs expiry on your inventory
Most nearshored manufacturing runs under IMMEX, which lets you bring inputs in without paying duty or value added tax on condition they leave again as finished goods within a window. For raw materials, inputs and packaging the general window is eighteen months. Machinery sits under different terms.
Two things follow that planning systems rarely carry. The first is that each unit of temporary import has a discharge deadline attached to the pedimento it arrived on. Material sitting past that date has to be returned, converted to a definitive import with duty and tax paid, or explained. A slow moving component a planner would happily hold for two years on physical grounds has a customs life shorter than its shelf life, and no standard netting engine knows this.
The second is Anexo 24, the automated inventory control obligation, which requires the customs record and the physical record to reconcile, discharged in order. Your consumption logic and your customs discharge logic have to agree, and where they do not the difference shows up in an audit as unaccounted material, valued at the duty and tax never paid on it.
Carry pedimento age as an attribute on the lot, run first in first out against the customs record as well as the physical one, and treat the discharge date as a hard constraint in the same place you treat expiry. The disposition logic for material approaching a date is I5's subject and the same machinery works here, with the deadline coming from a customs file instead of a laboratory.
The value added tax layer deserves its own line in the working capital plan. Since the 2014 reform, temporary imports carry the sixteen per cent tax unless the importer holds IVA and IEPS certification, which provides a full credit; the alternative is a bond. Certification comes in modalities with different renewal cycles, and losing it converts sixteen per cent of the value of every temporary import into cash posted up front, in the week the certification lapses rather than at a convenient quarter end.
The scope of the regime also moves. A December 2024 amendment to the IMMEX decree took a set of finished apparel classifications out of temporary importation, and whether your codes sit on that list is a five minute check with your broker that changes the landed cost model entirely.
Origin certification went from worthless to load bearing
For a long time a large share of industrial goods entered the United States from Mexico at a zero most favoured nation rate, so claiming NAFTA or USMCA preference bought nothing and plenty of manufacturers never built the records to support a claim.
The tariff orders issued in early 2025 exempted goods that qualify under USMCA, and the value of a certification went overnight from zero to the full tariff rate. Companies that had never run an origin programme discovered they needed one, retroactively, for products whose bills of material had been sourced with no attention to where anything came from.
Qualifying is a product specific rule: a tariff shift, a regional value content threshold, or both, supported by supplier declarations that have to run down the bill of material to the level the rule reaches. The certification itself is light, with nine minimum data elements and no prescribed form, and the weight sits in the evidence behind it, kept for five years.
For vehicles and parts the thresholds are higher: seventy five per cent regional value content, a labour value content requirement tied to work performed at sixteen US dollars an hour, and a seventy per cent North American purchase requirement for steel and aluminium.
Origin therefore becomes a constraint inside the sourcing decision rather than a compliance task downstream of it. A cheaper component from Asia can drop a finished good below its regional value content threshold, and the tariff that follows is usually larger than the saving that caused it. Price the tariff delta per unit and put it in the objective function alongside the piece price. How that cost moves into customer prices is C2's subject, and the filing and record keeping workflow is FF4's.
The USMCA joint review scheduled for July 2026 is a scenario rather than a forecast. Build both branches in the network model while the exercise is cheap, so the answer exists before anybody needs it in a hurry.
Capacity the site selection model did not price
Three constraints show up after the site is chosen and none of them appear on a typical comparison spreadsheet.
Power. Interconnection timing with CFE gates any expansion that raises connected load, and industrial demand growth in Nuevo Leon and the Bajio has lengthened that queue. A second shift needing more load is a capacity increase with a utility lead time in front of it.
Water. Monterrey's 2022 shortage was the visible version of a structural condition. Concessions are administered nationally and traded locally, and some industrial parks sell water rights with the land while others assume you will find your own. For anything wet, the concession is part of the capacity number.
Labour, and its legal edge. Turnover at border and Bajio plants is high enough that a staffing plan needs its own attrition assumption rather than a national average. The USMCA Rapid Response Labor Mechanism operates at the level of a single facility, which means a labour dispute at one plant can affect that plant's access to preference while its neighbour is unaffected.
Add road security, because it changes achievable transit. The Secretariado Ejecutivo del Sistema Nacional de Seguridad Publica publishes monthly cargo theft counts by state, and the corridors running from Mexico City towards Puebla and towards Queretaro have been the persistent concentration. The controls that follow, daylight only movement on some routes, GPS monitoring, convoy or escort, add hours the carrier quote does not mention. Effective capacity is the line rate multiplied by the shifts you can staff and the hours you can legally and safely move product.
The domestic plan that arrives with the plant
Sites built to serve the United States usually end up selling into Mexico as well, and the domestic demand signal has a shape that a weekly model destroys.
Most formal employment pays on a quincena, the fifteenth and the last day of the month, and consumption follows it closely enough that a daily series shows two clear peaks a month. Aggregate that to weeks and the pattern aliases into noise, because the fifteenth wanders across weekdays. A day of month regressor on a daily model, or at minimum splitting the month into halves, recovers a signal that weekly aggregation deletes.
The calendar has fixed points worth carrying separately. Aguinaldo, the statutory bonus of at least fifteen days' pay, is due by 20 December and lifts the month independently of Christmas. El Buen Fin has run in November since 2011, which is enough history to model as a promotion rather than an anomaly. Mothers' day on 10 May falls on a date rather than a weekend and is one of the largest retail days of the year, and Semana Santa moves with Easter; the method for moving holidays is D6's, and the route to market questions behind a Mexican channel structure of chains, convenience networks and several hundred thousand small independent shops belong to U8 and Z10.
Where this stops
Fitting a distribution to the crossing works for the ordinary variance and fails for the thing most likely to hurt you. Port of entry closures, tariff orders and facility level labour actions are policy events, and you will have too few of them in your own history to estimate a frequency. They belong in the plan as scenarios with a named trigger and a prepared response, rather than as a tail you can claim to have quantified.
There is a data limit underneath that. Most companies cannot compute the crossing distribution at all, because the timestamps live in the Mexican carrier's system, the transfer carrier's system, the US carrier's system and the broker's file, with no shared shipment identifier. Until those are joined, every number here is an assumption.
The economics are conditional too. A network justified entirely by a tariff differential is a position on that differential holding, with a review date already on the calendar, and it should be written down as a bet rather than as a plan.
Pull the last six months of northbound shipments and compute the standard deviation of the crossing leg on its own, joining yard out in Mexico to yard in on the US side; if those two timestamps cannot be joined today, building that join is the project that comes before any buffer change.
More on this in regional on the DemandLabs platform.
