Head office should own the definitions and leave the data in the plant

Every multi-plant manufacturer I know has had the same argument. Head office wants comparable figures for the board and the lenders, and believes the way to get them is to take raw data from every plant into one central platform and calculate the figures there. The plants resist, because they know their data contains context a central analyst will miss and because comparisons tend to be used against whoever looks worst. My position is that head office should own the definitions, not the data. It should publish a short dictionary of shared terms, pay for the people who apply them in the plants and leave the data, with its local detail, where it is produced.

This is not a claim that central data platforms are mistaken. We run one at Tessandier Aerospace, and it holds our serial genealogy record and our financial data. It is a claim about where authority over operational meaning should sit, and about which arrangement produces figures that people stop arguing with.

Six ways to count scrap

In 2023 our board asked for a dashboard comparing our six plants on scrap, first-pass yield and equipment effectiveness. When my team assembled the figures, we found six definitions of scrap. Granby counted rework as scrap. Querétaro excluded scrap caused by customer drawing changes, on the reasonable ground that the plant had not caused it. Mirabel recorded scrap when a part was quarantined, not when it was condemned, so its figures included parts that were later released. Scrap reported by the plants differed from scrap cost in the general ledger by 11%.

None of these definitions was wrong. Each made sense to the plant that used it. The problem was that a comparison built from them was meaningless, and it had been used for two years to judge Querétaro as our weakest plant.

Our chief financial officer proposed the conventional remedy: a central team would take the raw data from every plant and calculate one set of figures. I understood the appeal. It would have been faster, and it would have given head office control. But I had watched that approach begin at Castelmont Précision in 2008, when its new owner moved quality systems to a distant head office, and former colleagues told me how it ended. The central figures were consistent and the plant did not believe them. Managers kept their own spreadsheets, reviews became arguments about whose numbers were right, and the people who understood the data stopped improving it because it was no longer theirs.

A dictionary short enough to defend

We did something different. A definitions committee, made up of plant managers, finance and my office, agreed 46 shared terms. Each has a named owner, a formula, a list of exceptions and, where money is involved, a monthly reconciliation to the general ledger. Scrap is recorded when a part is condemned, not when it is quarantined. Rework is counted separately from scrap. Scrap caused by a customer's drawing change is reported, but in its own category, so that a plant is not blamed for it and the cost is not hidden.

Forty-six is a deliberate number. A dictionary of four hundred terms would be more complete and nobody would read it. A dictionary short enough for a plant manager to defend in a review, term by term, is one that plant managers will actually apply. Plants may keep any additional measures they find useful for their own purposes. What they may not do is report a shared term in a different way.

The arrangement is written down in a short data agreement signed by each plant general manager and me. It says that the plant owns its operational data and its local detail, that my office owns the shared definitions and the reconciliation, and that disputes go to the definitions committee, whose decisions are published. It is not a legal document. It is a record of who decides what, and it has been more useful than most of our policies.

The approach had costs. The board dashboard arrived three quarters later than requested, and our chief financial officer had to explain the delay to the audit and risk committee. Head office gave up routine access to most raw plant data. Some analyses that would be easy with a central copy now require a request to a plant. I think those costs were worth paying, but a company with a pressing financing event or a covenant tied to plant figures might reasonably decide otherwise.

What a plant gets in return

A plant will not accept outside definitions unless it gets something for them. Ours received two things. First, my office pays half the cost of the eleven plant data stewards who apply the definitions and fix data problems at source. They report to their plant managers, not to me, but their work on shared definitions is reviewed by my team. Second, the comparisons became fair. Once rework was counted the same way everywhere, Querétaro ranked second of six on first-pass yield, and its investment request was approved the following spring. Nothing persuaded other plant managers of the value of the dictionary as quickly as that.

Plant-reported scrap now reconciles to the ledger within 0.6%. Group scrap cost fell from 3.9% of revenue in 2023 to 3.2% in 2025. The data did not do that; the operations teams did. But they could not have done it while the figures were in dispute, and the disputes ended when the definitions became shared and the data stayed local.

There is one lesson I learnt late. The first list of downtime reason codes was written by engineers and contained 140 codes. Operators chose Other 38% of the time, which made the list useless. We rewrote it with operators and their union representatives, and settled on 38 codes. Other fell to 7%. A definition is only shared if the people who record it helped write it, and in a plant those people are rarely in the meeting room.

My advice to another manufacturer is modest. Before centralising operational data, list the terms your board compares across plants, find out how many ways each is counted and ask who in each plant would defend the shared definition. If the answer is nobody, a central platform will not help.