Mass customisation: individual output without individual cost

The buyer who wants one, and the data path that makes it viable

Demand comes from buyers who want something fitted to them and will accept a defined kind of choice rather than open-ended design: dental and orthopaedic appliances, spectacles, printed packaging, personalised consumer goods, made-to-measure furniture and apparel. The enabling condition is a clean data path from order to machine. Customer input has to arrive as structured data, convert into a build definition automatically, and reach production without transcription. Where that path is broken, the plant is doing bespoke work with a catalogue price attached, and the margin quietly disappears into office labour rather than onto the floor where anyone would see it.

A platform held common, with difference added late

The commitment is architectural: a shared platform, a defined set of variable attributes, and a deliberate decision about how late difference is introduced. Everything before that point is made against forecast and pooled; everything after is triggered by a real order. Finished goods therefore sit at almost nothing, since a completed unit belongs to one customer and to nobody else. The stock that does exist is common material and semi-finished product held at the postponement point, plus variety-specific consumables. Getting the split wrong hurts in both directions: too early and you hold variants nobody wanted, too late and customers wait for work that could have been completed in advance.

Equipment indifferent to what it makes, and proving each unit

Processes driven by digital instruction rather than physical tooling suit this best: cutting, printing, additive processes, programmable assembly and marking. Setup between units should approach nothing, because there is a setup between every unit. That is a different capital argument from a dedicated line, buying flexibility instead of throughput. Verification is the harder half. Each item is unique, so there is no reference sample to compare against and no meaningful sampling plan across units. Confidence has to come from a stable, monitored process combined with automated checks against the unit's own build definition, which is only possible where that definition is machine-readable in the first place.

Small, frequent supply, and where the model quietly breaks

Suppliers face uneven draw and small call-offs, uncomfortable for anyone organised around long runs. Common platform material is straightforward to plan; the awkward items are variety-specific ones needed in small quantities at short notice. Breakdowns are recognisable. Complexity leaks into operations as manual intervention, so touch time per unit creeps upward while the price stays flat. Order data proves ambiguous or wrong and the plant builds the wrong unique thing, which cannot be sold to anyone else. Returns are expensive by definition, since a personalised item has no second buyer and nothing worth recovering beyond raw material.

Buying against a distribution, and what genuinely scales

Purchasing plans from attribute distributions rather than part forecasts, buying to the expected spread of sizes, colours or specifications and revising as real orders reveal the pattern. Platform items can be committed confidently; the tail should be bought against firm demand even at a worse price. What scales here is order count, not option count. Putting more orders through the same process improves everything, because the fixed cost of the data path and the automation spreads further. Widening the attribute set usually does not, since each new dimension of choice multiplies the combinations that the process, the suppliers and the verification method must all handle.

Frequently asked questions

How is this different from simply offering more options?

Options are chosen from a list the plant already supports, and each one has a part behind it. Customisation generates a build definition from customer-supplied parameters, so the number of possible outputs is effectively unbounded while the number of physical items held stays small. That is only affordable where the process itself creates the difference, through a digitally driven operation rather than a stored variant. If every choice requires another component in the store, you are running an option catalogue and should plan it as one.

Why does the cost of a customised unit tend to drift upward?

Because exceptions accumulate. A parameter arrives that the automated path cannot interpret, so someone handles it manually and the fix becomes routine. Then a rarely used attribute needs a special material, then a customer requests something just outside the defined set and sales agrees. Each is small; together they reintroduce the labour the model was designed to avoid. Tracking the share of orders that pass through untouched is the most useful single measure of whether the economics still hold.

Where should customer-specific information enter the process?

As far downstream as the process physically permits, and once only. Every operation performed before the customisation point can be pooled across all orders, planned against aggregate demand, and run in efficient quantities. Everything after it belongs to one buyer and carries full risk if it is wrong. Capturing the information once, in a structured form that flows to each machine that needs it, also removes the most common defect in this model, which is a unit built correctly to incorrect instructions.