---
title: "Postponement for Seasonal Goods: When Late Dyeing Pays"
description: "Delaying dyeing or finishing seasonal goods pays when forecasts miss widely, variants sell evenly and the premium is low. Benetton, simulation, a worked sum."
url: "https://staging.khoi.co.uk/en/blog/tri-hoan-hang-theo-mua"
lang: en-GB
author: "Lê Quốc Khôi"
date: 2026-11-03
updated: 2026-11-03
topics: ["Supply chain & operations"]
tags: ["Postponement", "Benetton", "Seasonal goods", "Demand uncertainty", "Apparel supply chain"]
---

# Postponement for seasonal goods: when it pays and when it does not

## Key takeaways

- Postponement means keeping a product generic until demand is known, then finishing it. Benetton knitted garments from undyed yarn and dyed late; Dapiran (1992) says the firm's whole system could fill mid-season replenishment orders in two to four weeks, but he gives no cost for the delayed dyeing (Dapiran, 1992).
- A simulation on HP printer data (24-month life) finds the value of postponement rises with forecast error and with the number of variants. With eight variants and a 97 per cent fill-rate target, the break-even premium is 6 per cent of product cost at a 75 per cent error and 1.75 per cent at 30 per cent (Johnson and Anderson, 2000, p. 31).
- The value is lower when demand is lopsided across variants (Johnson and Anderson, 2000, pp. 29, 31). Lee (2002) places postponement in the responsive chain, which suits stable supply (Lee, 2002).
- Delaying colour pools only the split between colours; the total stays uncertain. Where a product's form cannot wait, Sport Obermeyer delayed the commitment of volume instead: early orders brought its forecast error down to about 10 per cent (Fisher, 1997, pp. 115-116).
- The five-colour example and every number in it are hypothetical, set and computed by me. None of the sources I read publishes Benetton's cost of late dyeing or its end-of-season markdowns.

Benetton knitted sweaters from undyed yarn and dyed the finished garments only when orders showed which colours were selling (Dapiran, 1992). Could a Vietnamese garment maker, or a buyer of seasonal ranges with a selling window of a few weeks, do the same? I read Dapiran's account of Benetton, a simulation of short-life products, Lee's framework article and research on Benetton after 2000, then ran a hypothetical sum of my own. The post on [efficient versus responsive chains](https://staging.khoi.co.uk/en/blog/chuoi-cung-ung-hieu-qua-hay-nhanh) covers choosing a chain product by product; this one takes a single tool from the responsive side.

## What is postponement, and what did Benetton do with knitwear?

Postponement moves the step that turns a generic product into a specific one (colour, language, sales channel) closer to the moment real demand is known. At Benetton that step was dyeing: the firm made garments from undyed yarn and held back the dye until news of the favoured colours arrived over electronic data interchange (EDI) (Dapiran, 1992). In the site library's terms, [postponement](https://staging.khoi.co.uk/en/glossary#postponement) pushes the [customer-order decoupling point](https://staging.khoi.co.uk/en/glossary#customer-order-decoupling-point) downstream: before it, work follows forecasts; after it, work follows orders.

Dapiran describes the old sequence, dye the yarn and then knit. Knitting is slow, so meeting service targets meant holding large stocks of finished garments, and the usual result was too little of the colour customers wanted, too much of the rest, and an end-of-season markdown. He lists four gains from late dyeing: expensive dye is not committed early, service improves, sales rise and fewer garments are written down (Dapiran, 1992). The article puts no figure on any of the four.

Late dyeing did not stand alone. The same article shows a package around it:

| Part of the package | According to Dapiran (1992) |
|:--|:--|
| Information | Agents in each country send orders by EDI; market data refreshed every 24 hours |
| Production | CAD design, cutting, knitting and dyeing kept in-house (high capital, scale economies); assembly and finishing subcontracted |
| Distribution | A $50 million automated distribution centre; all exports sent by air |
| Result Dapiran reports | Mid-season replenishment orders filled in two to four weeks, manufacture included |

Anyone copying this should ask whether they have daily order information, a dye house nearby and a fast way to ship. My reading of the table is that if one link is missing, the time saved by waiting has nowhere to go.

## What does postponement buy when the selling season is short?

It buys pooling of uncertainty: you forecast the total once instead of forecasting each colour or type. Johnson and Anderson (2000) simulate this for short-life products sold through several channels, using historical HP printer data: a 24-month base life, a 75 per cent forecast error and a three-month lead time from the factory (Johnson and Anderson, 2000, pp. 25-27). Their postponement case plans on one forecast for the whole platform; the comparison case plans on a forecast for each variant (Johnson and Anderson, 2000, p. 23).

Postponement gave higher fill rates in every case, cut both holding cost and lost-sales cost, and its benefit grew with the number of variants (Johnson and Anderson, 2000, p. 27). Fill rates also varied less from run to run (Johnson and Anderson, 2000, p. 27). The authors did not run the model on garments, though they note that the demand pattern they model (a surge at launch, a few steady months, then decline) is typical of apparel and toys too (Johnson and Anderson, 2000, p. 21).

At Benetton, Dapiran records eight fashion collections on top of the two basic seasons, a complete change of lines ten times a year, and replenishment orders taken in mid-season (Dapiran, 1992). The money saved he does not give.

## When does postponement not pay?

When forecasts are already fairly good, when demand piles onto one variant, when the uncertainty is about something other than colour, and when the premium exceeds the saving. The table collects the evidence I found for each condition.

| Condition | Evidence | Source |
|:--|:--|:--|
| Forecast error is small | With eight variants and a 97% fill rate, the break-even premium falls from 6% of product cost (75% error) to 1.75% (30% error) | (Johnson and Anderson, 2000, p. 31) |
| Demand is lopsided across variants | Value stays positive but is smaller; strongest when variants sell about equally | (Johnson and Anderson, 2000, pp. 29, 31) |
| Supply is unstable | Postponement for innovative products fits best where supply bases are reliable; Lee notes fashion apparel has stable supply | (Lee, 2002) |
| Uncertainty is about style, not colour | Early Benetton had few classic shapes, plain colours, and colour as the main differentiator | (Crestanello and Tattara, 2008) |
| The premium is high | The new process may cost more; Lee writes that when demand is unpredictable a responsive chain suits better than a cost-efficient one | (Lee, 2002) |

The style row deserves a pause. Crestanello and Tattara (2008) write that Benetton began with plain-colour knitwear in few shapes, using cheap raw yarn dyed later; where the Carpi district sold yarn-dyed knitwear in many shapes, Benetton used colour as its main way of differentiating a product (Crestanello and Tattara, 2008). Late dyeing solves exactly that problem, uncertainty about colour. If the hard thing to guess is which style will sell, late dyeing does not help; that is my inference from the same description.

On the numbers, Johnson and Anderson built lopsided demand by letting one variant take 80 per cent of demand (with two or four variants) or 60 per cent (with eight) and found a smaller benefit than with even demand (Johnson and Anderson, 2000, pp. 29-30).

## What can you postpone if the product's form cannot wait?

You can postpone the commitment of volume. Fisher (1997) describes Sport Obermeyer, a ski-wear maker with 95 per cent new designs each year and forecasts that often missed by as much as 200 per cent (Fisher, 1997, p. 108). Each February it invited its 25 largest retailers to Aspen; their early orders let it forecast national demand to within about 10 per cent (Fisher, 1997, p. 115). It could commit half its production after those orders, though the other half still had to be committed early. Fisher writes that the firm's approach, which he calls accurate response, halved the cost of overproduction and underproduction and lifted profits by 60 per cent (Fisher, 1997, p. 116). He co-wrote the 1994 article on this case, so it is not independent evidence.

Benetton also moved from a rigid plan to a more flexible one from about 2003. According to Crestanello and Tattara (2008), it had earlier run two seasonal collections with 80 per cent of orders taken before the selling season; it then split its range by speed. The two main collections carry response times of four to nine months, the Trend collection one to four months, and the Progetti and Capi continuativi lines can reach the market in seven days (made in Italy) or fifteen days (imported) (Crestanello and Tattara, 2008). The two main collections still make up 75 per cent of annual output under the firm's main label, so most volume is still planned early (Crestanello and Tattara, 2008). The paper explains the change by competition from Zara, H&M and Mango and by cost, not by any failure of late dyeing, and I have no source saying late dyeing was dropped. The post on [Uniqlo and Zara](https://staging.khoi.co.uk/en/blog/uniqlo-zara-phu-hop-chien-luoc) shows how two large firms mix these approaches differently.

## What does the sum look like for a five-colour Tet T-shirt line?

This example is hypothetical. The label's name, the prices, the demand and every number are mine; the sum is a single-period ordering model (the newsvendor) with normal demand, computed by me and not taken from any source above.

The label "Thun Nha Minh" sells matching family T-shirts for Tet in five colours, ordered once about twelve weeks before the season. The shirt sells at VND 250,000, costs VND 110,000 when dyed before sewing, and leftovers clear at VND 80,000 after Tet. Demand for each colour averages 200 shirts with a standard deviation of 80 (40 per cent, the low end of the 40 to 100 per cent range Fisher calls typical for innovative goods (Fisher, 1997, pp. 107, 109)); I assume colours are independent. The postponed option orders undyed shirts together and dyes them once the colours are known; I assume the dye house can dye finished garments to an acceptable standard.

| Expected result per season | Each colour ordered separately | Ordered together, dyed later |
|:--|--:|--:|
| Shirts ordered | 1,372 (5 × 274) | 1,166 |
| Expected leftovers | 410 | 183 |
| Expected missed orders | 38 | 17 |
| Expected profit (VND million) | 122.4 | 132.1 if late dyeing costs nothing extra |

Both options face expected demand of 1,000 shirts; expected sales are about 962 when each colour is ordered separately and 983 when ordered together, because some orders are missed. Ordering together leaves one standard deviation to insure against, √5 × 80 or about 179, instead of five deviations of 80 adding up to 400. The result is 206 fewer shirts ordered, 227 fewer leftovers, 21 fewer missed orders and about VND 9.7 million more profit. Spread over the 1,166 shirts ordered together, that gain can absorb a late-dyeing premium of up to about VND 8,400 a shirt (3.3 per cent of the price). Change the assumptions and the numbers move the way the sources say: with a 20 per cent deviation per colour the break-even falls to about VND 4,500; if one colour takes 600 of the 1,000 shirts (600, 200, 100, 50, 50, each at 40 per cent) it falls to about VND 5,100.

The sum leaves out two things. Colours often rise and fall together with the weather or spending power, which shrinks the pooling gain. And total demand stays uncertain, which early orders in the Obermeyer style address and late dyeing does not. The approach suits a label with past sales by colour from which to estimate the deviation; with only monthly totals, the 80 is a guess.

## Going deeper: where does this come from, and what did I read?

Johnson and Anderson (2000) say the idea has been discussed for over fifty years, citing Alderson (1950) and Bucklin (1965), and that Zinn and Bowersox (1988) defined five kinds of postponement: labelling, packaging, assembly, manufacturing and time (Johnson and Anderson, 2000, p. 20). They also summarise Lee and Tang (1997): standardising components pays only when the investment is low; modular design pays when the postponed stage has a long lead time or the module is easy to handle; reversing operations pays when the deferred step adds much value (Johnson and Anderson, 2000, pp. 20-21). That is second-hand; I did not read the original.

The limits of the evidence:

- Johnson and Anderson's simulation rests on one HP product sold across several channels, with a life of 12 to 30 months, not a selling season of a few weeks (Johnson and Anderson, 2000, p. 26).
- Dapiran (1992) is descriptive; Benetton's figures come from the company and from analysts, with no control group. Camuffo, Romano and Vinelli (2001), based on fieldwork at Benetton's headquarters from March to September 2000, still list delayed dyeing among the firm's success factors (Camuffo et al., 2001); I could read only the opening. Lee (2002) is a conceptual article with examples, not a test.
- No source I read gives the premium Benetton paid for late dyeing or its end-of-season markdown rate, and none covers Vietnamese garment makers.

What did I read? Dapiran (1992): the full text, from a Word copy on an Austrian university's server; it carries no printed page numbers, so I do not quote it. Johnson and Anderson (2000): the full text, with printed page numbers. Lee (2002): the full text, from a third-party copy without printed page numbers. Fisher (1997): the full Harvard Business Review reprint. Crestanello and Tattara (2008): the Italian-language version in the MPRA repository; I read most of sections 1 to 6 and the start of the conclusion. Camuffo and colleagues (2001): only the opening, the company box and the notes on the MIT Sloan Management Review page. Yang (2025): the full text. Not read: Lee and Tang (1997), Zinn and Bowersox (1988), Feitzinger and Lee (1997, the library's source for postponement, behind a paywall), Ferdows and colleagues (2004), and Caro and Martínez-de-Albéniz (2015).

## Frequently asked questions

### How does postponement differ from making to order?

Postponement moves the step that turns a generic product into a specific colour or type closer to the moment demand is known; everything before it still runs on forecasts. As I read it, make-to-order is the far end of that range. Dapiran (1992) describes Benetton filling a mid-season replenishment order within two to four weeks, manufacture included, by making only what was ordered, and credits the whole system of information, production and distribution, not delayed dyeing alone (Dapiran, 1992).

### Can late dyeing work for T-shirts, áo dài or printed fabric?

The sources I read do not test it. Benetton's case was plain-colour knitwear in a small number of classic shapes, with colour as the main way of telling products apart (Crestanello and Tattara, 2008). My inference: printed or woven cloth has its colour fixed before cutting, so there is no colour step to delay. Ask your dye house how far a finished garment can be dyed.

### What price premium can postponement carry before it stops paying?

There is no general figure. Johnson and Anderson's (2000) simulation, built on HP printer data, finds that with eight variants and a 97 per cent fill-rate target the break-even premium is 6 per cent of product cost at a 75 per cent forecast error and 1.75 per cent at 30 per cent (Johnson and Anderson, 2000, p. 31). Your goods may differ a great deal; the worked example below shows how to run your own sum.

### Does Zara delay dyeing the way Benetton does?

I could not verify it. A 2025 conference paper says Zara relies mainly on fabric suppliers, about half of them inside the Inditex group, that these mostly deliver undyed semi-finished fabric, and that Zara makes only 20 per cent of forecast sales volume in the first run; it cites no source for the two figures (Yang, 2025). The original study by Ferdows and colleagues (2004) is behind a paywall and I have not read it, so this post uses neither number.

## References

- Camuffo, A., Romano, P. and Vinelli, A. (2001) ‘Back to the future: Benetton transforms its global network’, *MIT Sloan Management Review*, 43(1), pp. 46–52. Available at: https://sloanreview.mit.edu/article/back-to-the-future-benetton-transforms-its-global-network/ (Accessed: 2026-11-03).
- Crestanello, P. and Tattara, G. (2008) ‘Da impresa radicata nel territorio a rete globale: la ristrutturazione del gruppo Benetton’, *Economia e Società Regionale*, no. 3, pp. 58–84. Available at: https://mpra.ub.uni-muenchen.de/12553/1/MPRA_paper_12553.pdf (Accessed: 2026-11-03).
- Dapiran, G.P. (1992) ‘Benetton: global logistics in action’, *International Journal of Physical Distribution & Logistics Management*, 22(6), pp. 7–11. https://doi.org/10.1108/eum0000000000416
- Fisher, M.L. (1997) ‘What is the right supply chain for your product?’, *Harvard Business Review*, 75(2), pp. 105–116. Available at: https://hbr.org/1997/03/what-is-the-right-supply-chain-for-your-product.
- Johnson, M.E. and Anderson, E. (2000) ‘Postponement strategies for channel derivatives’, *The International Journal of Logistics Management*, 11(1), pp. 19–35. https://doi.org/10.1108/09574090010806047
- Lee, H.L. (2002) ‘Aligning supply chain strategies with product uncertainties’, *California Management Review*, 44(3), pp. 105–119. https://doi.org/10.2307/41166135
- Yang, B. (2025) ‘Analysis of Zara's fast fashion supply chain rapid response strategy’, in *Advances in Social Science, Education and Humanities Research*. Atlantis Press, pp. 1339–1352. https://doi.org/10.2991/978-2-38476-440-2_151

---

Canonical HTML version: https://staging.khoi.co.uk/en/blog/tri-hoan-hang-theo-mua