The Overhead That Wasn't There
Why a costing house built in Changchun breaks in Nuevo León
The first time I reviewed the cost run of a Chinese automaker newly installed in Mexico's Bajío region, I thought a page was missing.
The unit cost of the vehicle was complete. Materials, broken out by supplier. Labor, with hours by station. And at the bottom, a line for manufacturing overhead at 6.8% of total cost. This in a brand-new plant, with robotic welding cells barely commissioned, forty process engineers on payroll, and a quality department that had not yet released a single saleable vehicle.
I asked the controller where the assembly line depreciation was.
He answered without hesitating.
"That sits upstairs, at corporate. Down here we book production."
It wasn't a data entry error. It was a costing house that worked well. Somewhere else.
The IMA put numbers to this in its field study on costing methodologies and cost management practices in the People's Republic of China, which surveyed close to two hundred companies and visited a dozen on site, including FAW, the state-owned automaker in Changchun that now builds a million vehicles a year. The finding that stopped me was not the most quoted one. At the median company surveyed, direct materials and direct labor together account for 92.3% of manufacturing cost.
Overhead is what's left over. And when something weighs 7%, it isn't worth arguing about how you spread it.
Three questions for anyone evaluating, competing with, or lending to a Chinese manufacturer setting up in North America.
- What happens to a costing system calibrated for 7% overhead when overhead reaches 30%?
- Does direct labor still work as an allocation base in a plant where direct labor barely exists?
- And what has to change before the first vehicle, rather than after the first year-end close?
One. The arithmetic that doesn't forgive
When overhead is 7% of cost, the allocation rule is nearly irrelevant. Spread it by machine hours, by labor dollars, by product weight, or by lottery. The maximum error you can introduce into unit cost is bounded by that 7%. A product can be undercosted by two or three points and the commercial decision doesn't move.
Swap the 7% for 30% and that same allocation error becomes a ten or fifteen point error on unit cost. Now the decision does move. List price moves. Which trim gets pushed to the channel moves. And the read on which plant is making money moves.
That inversion of proportions isn't theoretical. It's what happens when the operation crosses the border. Illustrative figures, not from any specific client: in a home plant, direct materials run around 78%, direct labor 8%, and manufacturing overhead 7%. In a North American plant of the same company, materials drop to roughly 61%, direct labor to 6%, and overhead climbs to 28%.
Three forces push that 28%. The first is automation. Every robot that replaces an operator converts direct cost into depreciation, and depreciation lives in overhead. The second is USMCA regional content. Annex 4-B requires regional value content reaching 75% for light vehicles, plus a labor value content component, and meeting it demands supplier engineering, homologation labs, and a trade compliance function that plant never had in Jilin. The third is the support structure required to operate in North America: EHS, environmental, labor relations, legal, warranty and service campaigns.
None of that is direct labor. All of it is overhead.
I got this wrong for a long time. I read a simple costing system as a poor costing system, and assumed the company running it misunderstood its own economics. Wrong. That system was well calibrated for its reality: marginal overhead, a narrow product mix, capacity almost always saturated. Building a sophisticated allocation model to spread 7% would have wasted analyst hours. The design judgment wasn't the failure. The assumption that economic reality travels with the company was.
Two. The disappearing denominator
The IMA study also documented how they spread it. Among companies that allocate overhead, 37% use direct labor cost as the base and another 24% use direct labor hours. Six in ten anchor the allocation to human work.
That's the right base for a plant in 1995. It's a time bomb for a plant in 2026.
Think about it as sensitivity. If direct labor is 6% of cost and overhead is 28%, every dollar of labor drags roughly five dollars of indirect cost behind it. A small variance in the standard time of one station, or a mix shift that moves hours between models, gets multiplied by five on its way to unit cost. The model isn't measuring resource consumption. It's amplifying noise.
There's a second-order effect that's more uncomfortable. When overhead follows labor, the products that consume the most automation look artificially cheap, because they consume little labor. In a plant running two platforms, the more robotized one subsidizes the less robotized one and nobody decides it. This is the classic mechanism behind the whale curve. A handful of variants generates 200% of profit and another handful gives it back, all invisible under full absorption costing with a single rate.
I had this conversation with the general manager of a Chinese auto parts operation in Coahuila while we reviewed margin by part number.
"Our highest volume product is the one with the thinnest margin. We've spent two years cutting its price to defend the volume."
I showed him the allocation. That product ran on the new line, the automated one, the line consuming nearly all the specialized maintenance and most of the depreciation. And almost no direct labor.
He went quiet for a while.
"So the margin we're defending isn't the margin we have."
It wasn't. The Profit Contribution Model we built afterward moved that product from an apparent 14% margin to a real 3.5%, and moved a low-volume variant from second place on the discontinuation list to first place on the expansion list.
Three. Actual cost versus normal capacity
From here on this stops being a management problem and becomes an audit problem.
Nine out of ten companies in the study allocate overhead on actual cost incurred. Only one percent use a standard or budgeted rate. They close the month, add up what they spent, divide by what they produced, and cost the unit that way.
With saturated capacity, the method works. It's simple, it ties to the general ledger, and it generates no variances nobody knows how to interpret.
But a new automotive plant in North America doesn't start saturated. It starts in ramp-up, eighteen to twenty-four months running at 30%, 50%, 65% of installed capacity. Divide full actual overhead by a volume that's half of design and you get an inflated unit cost that doesn't describe product economics. It describes the utilization gap.
That cost goes into inventory. And there it collides with the standard.
Mexico's NIF C-4, IAS 2, and ASC 330 converge on one criterion. Fixed production overhead is allocated based on normal operating capacity, and the portion left unallocated because of underutilization is expensed in the period rather than capitalized into inventory. Abnormal idle capacity is not a product cost. It's a period charge.
A plant spreading actual overhead over actual volume during ramp-up is, in substance, capitalizing idle capacity. In an audit that becomes an adjustment. And if the company exports from Mexico under the IMMEX regime, with the inventory traceability required by Annex 24 of the tax authority's rules, the conversation gets more specific still.
None of the executives I've worked with in these operations were hiding anything. The IMA study reached a comparable conclusion in its own terrain: it found no evidence that costing differences between China and the West explain dumping behavior. The issue is calibration, not intent. An instrument tuned to measure inside one range stops reading well outside it.
A costing system doesn't age. It relocates. The day the plant changes country, mix, or level of automation, the model that described the business yesterday starts describing a different one.
What to do with this
If you run finance in a Chinese operation newly installed in North America, or compete against one, or lend to one, there's a review you can start this week without launching a project.
Take the plant P&L and calculate what share of manufacturing cost is overhead. If it clears 15%, the inherited allocation base no longer serves you and needs revisiting before the next pricing cycle. Then ask for the rate detail. If the denominator is direct labor and direct labor sits below 10% of cost, you already know which product is subsidizing which. And check whether the allocation uses normal capacity or actual monthly volume. If it's actual volume and the plant is in ramp-up, you have an inventory valuation issue your auditor will find anyway.
Three numbers. Half a morning. You don't need a new model to find out whether the one you have is lying to you.
Looking forward, the pressure runs in one direction. USMCA regional content thresholds push these manufacturers to localize more manufacturing, and every point of localization adds fixed structure. Automation keeps replacing labor hours. And the treaty review keeps rules of origin on the table, which means unit cost traceability moves from a finance topic to a legal one. A company arriving at that conversation with absorption costing and a single labor-based rate will have more explaining to do than it would like.
The advantage these companies built was never an accounting advantage. It was scale, execution speed, and operating discipline, and that discipline is real. FAW keeps more than seventeen hundred people in the finance function and once declared an entire corporate year under a cost reduction banner. They know how to control. What they carry is an instrument tuned for a temperature range that is no longer theirs.
Recalibrating it costs eight weeks. Discovering it in the audit costs considerably more.
I'd be interested in what this group has seen. For those of you supporting inbound manufacturing investment, how are you handling the normal capacity question during extended ramp-up periods, and are you seeing allocation bases get revisited before or after the first audit cycle?