Nearshoring to Mexico: the systems bill nobody puts in the business case

The capital is real and it is still arriving.

Mexico climbed to 19th place in Kearney's 2026 FDI Confidence Index, up from 25th, one of the largest single-year gains of any country. Foreign direct investment reached roughly $41 billion in 2025, and the first quarter of 2026 set a record at $23.6 billion. Transport equipment alone accounts for close to half of manufacturing FDI, with aerospace, semiconductors and chemicals behind it, and the money continues to concentrate in Querétaro, Nuevo León, Tamaulipas, Guanajuato and Jalisco.

We work with these operations. The financial model that brought them here is usually careful and usually right.

It also has a consistent hole in it, and the hole is not tariffs, security or the USMCA review, all of which get argued about at length in the board pack. It is that nobody costed how the plant would be run, day to day, on software.

The business case ends at landed cost

Every nearshoring model we have seen compares the same variables: labour rates, freight, duty, inventory carrying cost, lead time to the customer, and some allowance for risk.

That comparison is sound. It is also a comparison of unit economics, and it implicitly assumes the new site will be operated with the same efficiency as the old one.

It will not be, for a reason that has nothing to do with the workforce. The systems that made the original plant efficient took fifteen years to accumulate, and none of them arrive with the equipment.

So the capital expenditure line covers the building, the machines, the fit-out and the hiring. The operating model quietly assumes a level of coordination that does not exist yet, and the gap gets filled by the only tools available on week one: Excel, WhatsApp and people who remember things.

What year one looks like

We have now walked into enough of these to describe the pattern without knowing the company.

Production scheduling runs on a workbook maintained by one person, and it is genuinely good, because that person is usually excellent, but it exists on one laptop and nobody else can operate it.

Quality records are captured on paper and photographed. The photographs live in a WhatsApp group. The group has 40 members and no retention policy.

Customs documentation is handled by a broker who is competent, and by an internal coordinator who reconciles what the broker filed against what the plant consumed, in a second workbook, monthly, from memory and email.

Maintenance is reactive because there is no equipment history, and there is no equipment history because the person who would have entered it is the same person maintaining the scheduling workbook.

Meanwhile the corporate ERP has been rolled out, on paper. In practice it receives a summary at month end, entered manually, and the plant runs on the workbooks. Everyone knows this. Nobody has time to fix it, because fixing it requires the same people who are currently absorbing 30% growth.

None of this shows up as a line item. It shows up as margin that never quite reaches the model, and as a plant that cannot answer basic questions about its own performance without three days of assembly.

The five things that need a real system on day one

Not eventually. Before the second year, when the workbooks have already become load-bearing.

Customs and inventory traceability. This is the non-negotiable one and it is routinely underestimated by foreign parents. An IMMEX programme carries a legal obligation, under Annex 24 of the Reglas Generales de Comercio Exterior, to operate an automated inventory control system that ties every movement of temporarily imported goods to its pedimento, invoice, production order and export documentation. This is not a reporting nicety. Failure can cost the IVA/IEPS certification and suspend the IMMEX programme itself, which is an existential event for an export operation. A spreadsheet does not satisfy it.

Multi-entity structure from the first table. Mexican operations are almost never one legal entity. There is a manufacturing entity, often a services entity employing the staff, sometimes a separate shelter or logistics arrangement. Costs, headcount and inventory belong to different entities, and that has to be in the data model at the beginning. It is the single hardest thing to add later, and we have said the same about claims operations across 28 legal entities, and the lesson transfers exactly.

Bilingual by construction. Not translated. The operators, supervisors and quality inspectors work in Spanish; the parent company reports in English. A system where the interface is English and the real work happens in a Spanish WhatsApp group has not been adopted, whatever the licence count says.

Shift-level operational data. Not monthly summaries. If the earliest a problem can be seen is the month-end pack, the plant is being managed with a four-week delay, and every improvement programme is arguing about numbers rather than acting on them.

Supplier onboarding and documentation. Local content requirements make supplier records a compliance artefact, not a convenience. With the USMCA review underway in 2026 and rules of origin under active discussion, the operations that can evidence their content position quickly are in a materially better place than the ones reconstructing it from invoices.

Why the parent's ERP does not land cleanly

The standard answer is "we will roll out the global system." It rarely works on the expected timeline, and the reasons are structural rather than technical.

The corporate template encodes the parent's regulatory environment, not Mexico's. Customs, payroll, entity structure and tax treatment are all different, and the configuration effort to represent them is usually larger than a new implementation would be.

The rollout is also sequenced last, because the new site is the smallest revenue line. So the plant spends its formative two years developing habits around workbooks, and by the time the ERP arrives it is competing with a working process that people trust.

And the parts of the operation that hurt most, the customs reconciliation, the quality evidence, the scheduling, are usually the parts the corporate template handles least well, because they are the most local.

This is not an argument against the ERP. It is an argument for not pretending the gap between now and the ERP is zero. That gap is typically three years long, and what gets built in it determines how the plant runs afterwards.

What we would do

Do not attempt the full system. Build the smallest thing that removes the load-bearing spreadsheet, and build it where the legal exposure is.

In practice that means starting with traceability across inventory, movements and customs documentation, because it is both the compliance obligation and the data foundation everything else needs. Then scheduling and quality capture, because those are where the institutional memory currently lives in one person's head. Then reporting, which becomes nearly free once the first two exist.

Sequenced that way, something real is in production in months rather than years, and each step pays for the next. Sequenced the other way, reporting first, because that is what the parent asks for, you build dashboards on top of numbers that are still being assembled by hand, and nothing underneath improves.

The bottom line

The nearshoring case for Mexico is strong and the investment figures support it. The risk is not that the thesis is wrong. It is that the model prices the plant and not the operation.

A site that cannot evidence its customs position, cannot see its own performance until month end, and depends on two people's workbooks is not cheaper. It is the same cost with the difficulty moved somewhere the finance model cannot see it.

The operations that get the full benefit are the ones that treated the operating system as part of the capital project rather than something to sort out once production stabilises, because production does not stabilise, it grows, and the window closes.

If you are standing up or scaling an operation in Mexico, our operations work starts by mapping what is holding the process together, and what replacing a spreadsheet-run operation costs sets out the real bill before anyone signs anything.