Late-stage customisation: holding product generic for as long as you can

Choosing where the product stops being generic

The postponement point is a design decision before it is a logistics one. It usually requires re-engineering so the differentiating feature is added rather than built in: a common chassis with a fitted power lead, a base fluid coloured late, a device configured by software at pack-out. Where the difference is embedded early — a moulded colour, a welded bracket, a printed body — postponement is unavailable without changing the design. Naming the point explicitly, then defending it against small engineering changes that creep upstream, is the continuing discipline. Each upstream creep converts pooled stock back into version-specific stock without anyone recording that it happened.

Generic stock absorbs the forecast error

Holding one item that can become many collapses the variance the plant has to cover: instead of predicting each version, you predict the family and let the mix reveal itself. That is worth most where versions are numerous, individually unpredictable and collectively stable — regional variants, colour ranges, badged equivalents of one product. Cover falls sharply for the same service level, obsolescence drops because generic units retain their options, and money tied up in slow versions is released. The residual stock problem moves to version-specific components: labels, manuals, adaptors, cartons.

The last operation has to be quick, cheap and dependable

Everything the approach promises depends on the finishing step fitting inside the promised lead time. Equipment there is usually light — labelling, packing, configuration, simple assembly — and can sit in a distribution centre rather than a factory, which is what makes local finishing viable at all. Capacity at that point decides responsiveness: it has to handle the peak day rather than the average, because the whole flow now waits on it. Adding a second finishing location is often cheaper than adding capacity further upstream. Staffing that step for the peak is usually done with flexible labour, since the work is quick to learn and the volume swings with the order pattern.

Defects at the final step reach the customer

Nothing follows the customising operation, so an error made there ships. A mislabelled language version, the wrong adaptor in the box, a firmware image intended for another market — these are the characteristic escapes, and they are clerical rather than technical. Controls belong at the point of application: scanned verification that the variant component matches the order, a checked clearance between versions, and a sample audit of completed packs. Where finishing happens away from the plant, the quality system has to reach into a site that thinks of itself as a warehouse.

Product structures and the systems that carry them

Order promising has to understand that a version does not exist until it is made, so availability comes from generic stock plus finishing capacity rather than from a variant balance. Bills of materials need a generic item plus a differentiation kit, and forecasting works at family level while replenishment of version-specific components works at item level — two planning behaviours inside one product. Upstream suppliers benefit, receiving stable orders for the common item instead of a jittery mix. The version-specific parts stay volatile, and they deserve short lead times even at a higher price.

Frequently asked questions

How do we find the right postponement point?

Look for the operation after which the number of distinct items multiplies, then ask what it would take to move that operation later. It is usually where colour, language, voltage, packaging or configuration enters. Test whether the resulting generic item is genuinely usable for every version — a common base still needing rework for some versions delivers none of the benefit. The right point often sits one step earlier than the obvious answer and needs a design change to reach.

Does postponement always reduce total inventory?

It cuts the stock needed for a given service level in most cases, though not universally. The saving comes from pooling variance, so it is largest where versions are many and individually erratic. Where a product has few versions with steady demand, the pooling benefit is small and may not repay the extra handling, the second-stage capacity and the additional transactions. The other trap is version-specific components, which still have to be forecast, stocked and expedited.

Can late customisation be done in a warehouse rather than a factory?

It often is, and that is where the responsiveness comes from, but the site has to be run as a production operation for those activities. That means controlled procedures, trained operators, calibrated equipment where measurement matters, batch or serial records linking back to the generic unit, and a documented release. Regulated products may require the finishing site to hold its own authorisation, a question to settle before the network is designed rather than afterwards.