There is a contradiction that repeats itself in plants along the border. The company has a serious ERP, paid for, implemented by a large consultancy and governed from corporate. And the warehouse still runs on a spreadsheet that someone updates at the end of the shift.
The easy conclusion is that the ERP failed. It is almost always wrong, and believing it costs money: it leads to two-year replacement projects that end up in the same place, with a different spreadsheet.
The ERP is not failing at its job. Its job simply ends before your operation happens.
What an ERP does well, and why that is not a flaw
An ERP is a system of record. It exists so that accounting, purchasing, costing and taxes close and reconcile. It is built to close a month, not to govern a minute.
That is why it is deliberately slow. Every movement it accepts has accounting consequences, so it accepts it with validations, authorizations and a trail. That is correct. It is what you want from a general ledger.
But a picker holding a scanner at three in the morning, standing in front of a badly labeled rack, does not need a general ledger. They need to know which location to pull from, and they need an answer in half a second.
Those are two different problems. Asking one piece of software to solve both is what produces the spreadsheet.
Four signs it no longer reaches
There is a spreadsheet nobody dares delete. That is not sloppiness. It is the most honest symptom there is: someone needed data the system would not give and solved it themselves. That spreadsheet is a specification written by the operation. It is worth reading before proposing anything.
Data entry happens after the fact. During the shift people write on paper, and at the end someone types it in. The moment that happens, the system stopped being a source of truth and became a historical archive. You cannot react to something that already finished.
The data exists, but not at the level you need. The ERP knows the line produced four thousand pieces. It does not know that machine three was down forty minutes waiting for material, or who decided that, or whether it was the fourth time this week. Correct totals, missing detail.
The change you ask for takes a quarter and is decided in another country. That is not a technical problem, it is a governance one. And it delays more projects than anything else, because it does not show up until after you have signed.
Why “replace the ERP” is almost never the answer
First, because in a maquiladora it is rarely your decision. The ERP arrives from corporate, with the license, the chart of accounts and the policy. The plant receives it, it does not choose it.
Second, because replacement does not address the problem. If the gap is on the floor, at the dock and in the warehouse aisle, a new ERP will have the same gap in the same place, two years later, with a budget you no longer have to fix it.
What works is more modest: let the ERP do what it does well and build what it lacks alongside it, with a border drawn on purpose. That is what we mean by ERP work: it is rarely installing one, it is making the one you already have usable for operating.
The rule that prevents the second problem
One owner per piece of data.
Before a line of code is written, you decide in writing, by data type, who is in charge:
- The ERP owns the part master, the purchase order, the cost, and anything that ends up in a financial statement.
- The floor or warehouse system owns the location, the scan, the downtime reason, the operator and the exact timestamp.
Never both. The moment two systems can edit the same number, you have just hired someone to reconcile them, permanently, even if you do not know it yet.
The question that exposes this mistake is simple: if the ERP says 900 and the scanner says 880, which one is right? If the answer is “it depends”, the design is wrong.
How to connect without becoming a shadow system
There are three levels of integration. Start with the cheapest and move up only when something forces you to:
- Scheduled files. A nightly export from the ERP, an import back. It is inelegant and it works more often than people admit, especially when what travels is a master that rarely changes.
- A middle service. Your system does not talk to the ERP: it talks to a layer of your own that translates. When corporate upgrades the ERP — and they will — the change is absorbed in one place.
- Events. The movement is published to a queue and the ERP consumes it when it can. This is the right answer when volume is high or the floor cannot wait for the ERP to respond.
And one rule worth more than the three: the integration has to be able to fail out loud. Something has to say so when it did not run. The normal failure is not a bad integration, it is a silent one: for three weeks nobody noticed last night’s file came in empty.
The same goes for resends. If the same message arrives twice, it cannot double the inventory. That gets designed up front, or discovered during a physical count.
What has to be decided before anything is built
- Who owns each piece of data. The above, written down and agreed by both sides.
- What happens when they disagree. Not just who wins: also who looks at it, how often, and on what screen.
- What permission you have to talk to the ERP. A read-only user, a view, an API. Get it in writing before starting. This is the item that delays these projects most, and the one most underestimated, because it depends on an IT group that sits in another city with its own priorities.
- What happens at the next ERP upgrade. If your integration depends on an internal table, the next version breaks it. You want to know that beforehand, not afterwards.
What a satellite system does not fix
It does not fix an undefined process. If nobody has decided what counts as a good piece or who authorizes a movement, new software will only record the indecision faster. We wrote that about floor systems and it applies here just the same.
It does not exempt you from customs obligations. If your material came in under an IMMEX program, the duty to say where every unit is remains yours, and a corporate ERP often does not cover it because it was not designed for Mexico. That gets modeled separately and on purpose: we covered it in Annex 24.
It does not replace someone who knows the operation. A satellite system is designed by watching the shift, not by reading the ERP manual.
That border is rarely the only open front. When there are three or four at once, the order you touch them in ends up mattering more than which one you pick.
And if the satellite covering those fifteen meters already exists but was written by someone who no longer works here, find out what exactly you inherited before touching it: having the code is not the same as having the program installed.
That border has a consequence paid in money rather than convenience: while the floor operates outside the ERP, a part’s cost gets assembled by hand and arrives late. We covered it in what that part actually costs you.
If that is where you are — an ERP that closes the month properly and a plant that operates in Excel — the useful conversation does not start with software. It starts by drawing the border. That is what the twenty minutes are for.