Micromanufacturing: The Conditions Under Which the Small Footprint Becomes the Stronger Design

The cost curve moved, and most strategies did not

The industrial logic of the last generation was straightforward and, for its conditions, correct. Concentrate volume in the largest feasible plant, place that plant where landed cost was lowest, and let scale absorb overhead. Long runs, low changeover frequency, long lead times, high inventory in transit. It worked because the variables it ignored — freight reliability, tariff exposure, geopolitical access, currency, single-source concentration — behaved themselves.

Those variables no longer behave themselves. Tariff regimes shift inside a fiscal year. Ocean freight prices and transit times swing by multiples on events nobody modelled. Export controls turn a qualified component into an unavailable one without warning. And concentration — one plant, one region, one supplier, one channel — has become the risk that boards ask about directly.

This does not make scale wrong. It makes the unit of comparison wrong. When the cost of a part was compared plant to plant, the biggest plant in the lowest-cost region usually won. When the comparison is total delivered cost including the cost of being wrong — inventory carried against uncertainty, expedites, obsolescence, lost orders, tariff exposure, the working capital tied up in six weeks of water — the answer changes for a real and growing share of products.

Scale did not stop working. The variables it was allowed to ignore stopped cooperating.

What micromanufacturing actually is

Micromanufacturing is the deliberate use of small, replicable production units placed close to demand, designed for fast changeover and a narrow, high-value product envelope. A unit might be a cell inside an existing plant, a leased bay in a regional market, or a containerized line placed at a customer's campus. The defining features are footprint, proximity and repeatability — not equipment size.

It is easiest to see by what it refuses. It refuses the assumption that unit cost is the only cost. It refuses the long run as the default lot size. It refuses the idea that each site needs its own bespoke way of working. And it refuses the notion that a small plant must be a technically unsophisticated one — the economics only hold when the small unit carries current process capability, automation where it earns its place, and real measurement.

The strategic point is optionality. A network of small units can add, move, retool or retire capacity in increments the business can actually fund, at a pace closer to the pace demand is now changing. A single large plant is a single large bet, held for a decade, on conditions that no longer stay still for a decade.

The supply chain consequence: fewer miles, more nodes, more governance

Moving production near demand does not simplify the supply chain. It changes which part is hard. Long-haul complexity falls — fewer ocean legs, less inventory in transit, shorter and more predictable replenishment. Network complexity rises — more nodes to qualify, more supplier relationships per unit of volume, more places where a standard can quietly drift.

That trade is only favorable when the governance is built. Multi-site manufacturing without a common operating standard produces the worst of both designs: the cost structure of a network and the discipline of a startup, repeated per site. Each plant develops its own process, its own definitions, its own numbers, and the enterprise loses the ability to compare its own sites to each other.

So the supply chain question stops being 'where is it cheapest to make' and becomes four architecture questions. Where must capacity sit to serve demand at the lead time we sell? What must be identical across every unit, and what may legitimately be local? Which suppliers are qualified to serve more than one node, and which are single-node by design? And what is the standard by which a new node is declared ready to produce?

Long-haul complexity falls and network complexity rises. Whether that is a good trade is decided by governance, not by geography.

Why earlier attempts disappointed

Small-footprint production is not a new idea, and the honest position is that the public record of it is thin: most attempts are private, the ones that are written up are written up by the people who sold them, and there is no body of independent evidence that a node network outperforms a concentrated plant in general. Anyone claiming otherwise is selling something. What can be read reliably is the pattern of how the attempts came apart.

Four failure conditions recur. The node was justified on unit cost, so it lost the comparison it was never designed to win. The standard was never authored, so the second node was a fresh invention and the third was a third one. The labor mix was assumed rather than decided, so the site ran on one indispensable person. And the node was given responsibility without authority — expected to hold a schedule while pricing, engineering changes and supplier decisions stayed elsewhere.

Each of those is an architecture failure, not a verdict on the footprint. That is the practical value of the pattern: it tells you what must be true before the capital is committed, rather than what to hope for after.

The evidence does not say small plants win. It says the ones that failed failed for reasons that were decided before the equipment arrived.

The capability problem is the real barrier

The equipment for a small unit is usually available. The capability to run it is usually not. A near-demand node needs people who can set up, change over, inspect, troubleshoot and improve — often in one shift, in a labor market that may never have supplied that skill, at a scale too small to carry a full functional staff.

This is where most micromanufacturing plans fail, and they fail quietly. The business case assumes a labor market that will supply qualified operators and technicians on demand. The market supplies neither at the rate required. The node then runs on heroics: one capable person holds the site together, and the model cannot be replicated because the thing being replicated was never written down.

Two commitments make the difference. First, the operating standard is authored before the second node opens — the process, the measures, the decision rights, the training path, the certification gates. Second, the sourcing mix is decided with finance as an explicit capital-allocation choice: what proportion of the requirement must arrive fully qualified, and what proportion the enterprise will deliberately develop, with a stated ratio, an owner, committed training capacity and a review date. An assumed ratio is how a development pathway becomes the fallback for failed recruiting.

Where this sits in the enterprise architecture

Micromanufacturing is not a plant project. It touches every layer of the operating architecture, which is precisely why it is so often executed as a facilities decision and then underperforms.

FOUNDATION carries the honest read: whether the leadership system, decision rights and measurement can hold a second and third node at all. SIMPLIFY makes the allocation choice — which markets, customers and products actually justify near-demand capacity, and what gets stopped or deferred to fund it, because a ranked wish list is not a decision. LEAD installs the operating cadence and the standard by which a node is declared ready. GROW places the capacity against where demand is going rather than where it has been. ARC runs the loop: analyze what the first node actually taught, refine the standard, commit the change before the next node is built.

The execution disciplines carry the rest. Front-to-Back 80/20 tells you which portion of the portfolio earns the network. Change and Transformation governs the shift in how work is done at each site. Program and Project Management sequences node openings, qualification and supplier onboarding as one program rather than a series of unrelated builds.

Executed as a facilities decision, a network of small plants becomes a network of small problems.

What the technology layer has to do

A network of small nodes cannot afford a headquarters at each one. The only way the model scales is if the standard travels digitally and the measurement is common. That is the practical work of the technology layer: one definition of a process, one set of measures, one place status is read rather than assembled, and the ability to see every node against the same standard on the same day.

CompassOS™ is where that lives in a Compass engagement — the standard published once, the value stream mapped, the measures instrumented, the readiness of each node visible without a site visit. The point is not dashboards. It is that a small unit can operate at a standard it did not have to invent, and the enterprise can tell the difference between a node that is behind and a node that is being run differently.

That is also the honest limit of the technology. It accelerates a decided architecture; it does not substitute for one. A network with no authored standard will use good software to distribute an unclear way of working faster.

How to test the case in one quarter

The case for a near-demand node does not need a year of study. It needs one product family, total delivered cost read honestly, and a written standard the second node could actually follow.

Take a single family where freight, lead time or tariff exposure is material. Build the total delivered cost both ways — including inventory carried against uncertainty, expedite history, tariff exposure and the working capital in transit — and state the assumptions that would have to be wrong for the answer to flip. Then map the value stream as it runs today and remove the time no customer would fund, because a node that replicates today's waste replicates it in a new geography at a new fixed cost.

Then decide the smallest real unit: what must be identical across nodes, who has authority at the node and who does not, the labor mix and how it will be developed, and the binary readiness gate. If the standard cannot be written, the network is not yet a strategy — and one plant is still the right answer for now.