Mitigating demand uncertainty across a winery’s sales
channels through postponement
S. CHOLETTE, Assistant Professor of Decision Sciences
San Francisco State University College of Business
1600 Holloway Avenue
San Francisco, CA 94132 USA
Phone: 001 (415)405-2173
Fax: 001 (415) 405-0364
Email: [email protected]
Email is the preferred method of contact
1
Mitigating
demand uncertainty across a winery’s sales
channels through postponement
Abstract (word count: 190 words)
Wineries must allocate production across multiple sales channels before demand is
known. Misallocation may result in undesirable surpluses in some channels and lost sales
opportunities in others. We investigate this problem by constructing a mathematical
model for postponing channel differentiation. We provide a process overview for a
winery and present a two-stage stochastic linear program with fixed recourse that
maximises expected profit over a distribution of demand scenarios. In the first stage, the
winery allocates production to finished goods by channel and to intermediate inventory
points. Once demand is known, recourse variables include transformation of intermediate
inventories. Results from solving this model using a mix of data derived from interviews,
literature and placeholder parameters suggest that a considerable portion of production
should be held at both the labelling and packaging level. Furthermore, postponement
usage is advisable over a range of demand probabilities and can lead to significant
improvement in profitability. We compare our theoretical results with some actual
wineries’ practices and provide guidance for future research. Although other stochastic
programming applications evaluate postponement strategies within verticals such as high
technology, this research is their first application within the wine industry.
Keywords: Postponement, Stochastic Models, Linear Programming, Channel Allocation,
Wine Industry, Private Label
Word count of main text: 4300 words
2
1. Introduction: private label wines and the proliferation of sales channels
Wineries often sell wine through multiple channels, each with different packaging and
labelling requirements. Most US wineries sell their own brand, typically through retailers
via a wholesaler. Other sales channels include those of bulk wine and private label, wines
labelled with a brand owned by a supermarket or restaurant and sold exclusively through
that venue.
Private label wines had less than a 1% share of the total US wine market in the
late 1990s (Ward et al, 2002), but commanded a 20% share in 2003 (Popp, 2003). This
expansion is likely to continue, given trends in other countries. For instance, storebranded wines account for over half of sales in the three largest UK supermarkets
(Chaney, 2004).
Private label products are favoured by retailers for giving higher margins and
increasing customer loyalty (Ward et al, 2002) but provide wineries with mixed benefits.
Wineries receive lower prices and less brand recognition but benefit from a decreased
need to market wines and from assured store placement. Private label has become a
potential channel for many wineries, not just those producing mass-volume, inexpensive
wines, as private label wines exist across a wide range of prices and sales volumes.
With this additional market opportunity, wineries face the challenge of
determining product allocations across channels. Unlike many other industries where
manufacturing occurs continually or products can be sourced year round, wine production
follows a yearly cycle. Grapes are harvested or purchased once a year, and bottling
traditionally occurs a couple months later, before demands are fully known. Products
across different channels may contain the same wine, but they cannot be treated as
substitutes. Thus, a winery may consider hedging against demand uncertainty by
postponing the finishing processes that differentiate these products.
2. Postponement methodologies and their potential use within the wine industry
Although the concept of postponement was introduced into academic literature in 1950, it
received little attention for many years (Yang et al, 2004). Van Hoek (2001) points to a
recent renaissance of postponement research, but shows that approaches are far from
standard, with a dearth of studies that measure realised gains. Real world adaptation has
also been slower than expected (Yang et al, 2005). Case studies documenting actual
usage, such as by Lee et al (1993), Skipworth and Harrison (2004) and Van Hoek (1998),
show that the automotive and high technology industries have been more successful in
adopting postponement than many other verticals. Van Hoek (1999) emphasises that
agribusiness has lagged behind other industries in adopting postponement methodologies.
This lack of homogeneous adaptation is in part attributable to some industries’
capability to employ processes that support postponement, such as adapting information
coordination technologies and modularising product design for use of component
commonalities (Van Hoek et al, 1998). Even more fundamentally, the ability of
organisations to change processes to enable postponement or even to adjust company
mindset to consider use of such techniques should not be ignored. Indeed, the extensive
survey by Yang et al (2005) finds that change management is the third most significant
barrier to implementing postponement. Although the wine industry is known for its
3
adherence to tradition, we believe it stands to benefit greatly from postponement, and
have undertaken research to better quantify this potential.
Classic postponement models often assume that the demands for varieties of a
product are independent, such as in Benetton’s knitting of sweaters prior to dyeing
different colours (Chopra and Meindl, 2004). This assumption would be problematic in
the wine market; a demand shift in one channel is likely to be reflected elsewhere. For
example, Americans’ increasing wine consumption is likely to raise US demand in all
channels. Thus, any plausible approach must consider demands jointly across channels.
Likewise, although only a few studies consider allocating wine production under
demand uncertainty, the most common approach involves utilising multi-stage, multifacility networks. This setup would theoretically allow wineries to effectively meet
varying regional demands (Yu and Li, 2000) and to delay final weight-added packaging
(bottles) until transport to the regional market has occurred (Van Hoek, 1997). However,
actual industry practice suggests that decentralising a winery’s bottling and distribution is
impractical. Both Orr (1999) and Twede et al (2000) emphasise that packaging is a highspeed automated process involving expensive equipment, favouring centralisation.
Despite the advantages that multiple localised warehouses may have for reaching
markets, wineries typically centralise warehousing and outbound logistics functions; both
E. & J. Gallo and the Mondavi division of Constellation rely primarily on a single
warehouse facility for their domestically produced wines.
A more promising technique for dealing with demand uncertainty for wine is
postponing the final form. Zinn and Bowersox (1988) define levels of form
postponement: manufacturing, assembly, packaging and labelling. No specific mention of
wine production and distribution is made, but the authors state the circumstances that
may make different levels of postponement appealing to winemakers. Labelling
postponement should be considered when the product is marketed under several brands,
and packaging postponement is appropriate when the product is sold in multiple sizes or
formats. As pointed out by Yang et al (2004), postponement is but one tool available to
manage uncertainty and may be of less value if these uncertain demands must be quickly
and frequently met. We assume a winery’s main concern is appropriately allocating
products over the course of the year, and quick response is of less importance.
The proliferation of sales channels available to the winery is analogous to the
increased product differentiation faced in other industries. Swaminathan and Tayur
(1998, 1999) employ stochastic programming techniques to delay final computer
configuration to effectively manage product differentiation. We take a similar approach,
and our model is the first documented analysis of such techniques applied to the wine
industry.
The rest of the paper is organised as follows. We present the post-production
process flow of a hypothetical winery with several sales channels. We then construct a
stochastic linear programming model that considers initial and recourse decisions. Using
parameters derived from interviews and literature as well as placeholder data whenever
real data was unavailable, we allocate production across finished goods channels and
intermediate inventories to optimise profits, given several demand scenarios. We
compare model results for different assumptions about these demand scenarios, noting
when postponement is likely to be utilised more. Lastly, we compare our theoretical
results with some actual industry practices and make suggestions for future research.
4
3. Modelling the potential for postponement at a winery
3.1. Process overview
Wineries rely on centralised bottling plants to smooth capacity needs, but an additional
benefit of this facility setup is the increased potential for postponement. Figure 1 depicts
a process model of a hypothetical winery that sells a single type of wine into four sales
channels. The model allows for both packaging postponement and labelling
postponement, as defined by Zinn and Bowersox (1988). The winery can chose to
initially allocate wine to the bulk sales channel, to bottle it immediately, or hold it in a
tank. Tanks serve as a point for accumulating intermediate inventory. While not a
saleable product for the end consumer, bulk wine can be a finished good from the
winery’s perspective, as it sold by tank directly to a negociant or another winery.
Further product differentiation occurs during the bottling process as our sample
winery uses two different bottle types. One type is used both for its own brand and
Private Label A, and the other type is dedicated to another private label product, Private
Label B. This different packaging specification may result from size (such as 1.5l verses
750ml bottles), a different bottle shape or colour, or a different closure, such as a
synthetic cork instead of a natural cork. After the bottling has occurred, products can be
either be labelled immediately or left unlabelled. Blanks of the one bottle type can be
used to make either the winery’s own brand or Private Label A or can be held as
intermediate inventory for later differentiation. Since only one saleable product is made
with the other bottle type, no postponement advantage exists for storing blanks of Private
Label B.
3.2. Modelling methodology
If demand were known at the time of production, it would be straightforward to apply a
simple linear program to optimise allocation across channels. Even a greedy heuristic
could be used to allocate production to the highest margin channel until that demand was
satisfied before fulfilling demand for the next highest margin channel. But a more refined
approach is needed to reflect the inherent market uncertainties of the wine industry.
Demand in reality is revealed gradually over the course of the year, but we
assume a two-stage model, where channel demands are initially unknown, though the
scenarios and their probabilities are assumed to be identifiable. The winery must commit
to an initial allocation of wine to produce as finished goods by channel. The winery can
also hedge against uncertainty by allocating wine to intermediate inventories at the prebottling or pre-labelling stages. Processes where product finishing is postponed usually
costs more than the basic production process but allow for greater recourse.
The problem can be formulated as a two-stage stochastic linear program with
fixed recourse, similar to those constructed over 50 years ago by math programming
pioneers like Dantzig (1955). We consider the standard formulation of Birge and
Louveaux (1997):
max cTx + EQ(x,)
subject to Ax = b, x ≥ 0
(1)
T
where Q(x,) = max(q y|Wy = h-Tx, y ≥ 0 )
5
Assuming that the producer is risk-neutral, the optimal solution is to maximise expected
profits over the discrete probability distribution of the enumerated demand scenarios.
We utilise seven demand scenarios: a base scenario and six others that adjust the
demand by a percentage of base demand, as shown in figure 2. These scenarios were
defined to represent overall market growth or decline as well as individual brand effects
and reflect channel covariance. The standard situation we consider is that the base
scenario has a 25% likelihood and the other six scenarios are each 12.5% likely. Demand
swings for private label wines are assumed to be greater than for the winery’s own brand.
For example, if private label demand decreases, the retailer would be likely to
substantially reduce the order, perhaps even cancelling it, as seen in Wal-Mart’s decision
to discontinue Alcott Ridge, their private label sourced by E. & J. Gallo (Berger, 2004).
We also assume that the bulk wine channel cannot be oversupplied, which is a
simplification as this channel occasionally becomes saturated.
Defining several discrete scenarios would seem a more approachable way for
winemakers to provide their expertise on market uncertainties than to have them define
continuous joint probability distribution functions. Further scenario enumeration would
increase the size of the problem, but would not affect the model’s fundamental structure.
We present a small number of scenarios as well as a limited number of sales channels so
that the model and results can be clearly and succinctly presented to the reader.
3.3. Model structure
The model has two types of first-stage decision variables: xpc, the quantity to produce
initially as finished goods for each sales channel, and xii, the quantity to leave in each
intermediate inventory stage. The production of these finished and intermediate goods
cannot exceed the total available supply of wine, Q, as shown:
(2)
Q   xpc + xii
C =  brand, pl-a, pl-b, bulk  , I =  tanks,blanks 
cC
iI
Equation (2) is the only constraint that does not invoke demand scenarios as it is
concerned with first-stage decision variables. Initial decisions to produce finished goods
cannot be undone; the packaging and labelling processes are not reversible. While it is
physically possible to repackage wine, costs would be prohibitive, exceeding the value of
all but the most expensive wine. First-stage commitments to bulk wine sales are also
assumed to be irreversible.
The remaining five types of variables are recourse variables and include the
dimension of the demand scenario, S = (s1,s2…s7). The first of these, xti,c,s, lets the winery
decide how much intermediate inventory to turn into finished goods by sales channel.
Equation (3) prevents invalid transformations, such as blanks becoming Private Label B
wines. The matrix |T| is defined in Appendix A, as are all data parameters:
(3)
xti ,c,s  Tc,i xic,i i  I, c  C, s  S
The next type of variable, xdi,s, records dumping. The winery has the option of
not processing all intermediate inventories. The decision to dump wine is not an enviable
one, but it may be the best option in low demand scenarios, as further finishing costs are
avoided and the winery may receive dumping subsidies. Equation (4) guarantees that all
intermediate inventories are either transformed into finished goods or dumped, since
storage between years is not allowed:
6
xii  xdi ,s   xti ,c ,s
i  I, s  S
(4)
cC
The model assumes that no inventory from the prior year remains and all of that
year’s production is sold with a year. While starting and ending inventories are important
considerations for other industries, they have less relevance in wine production. Most
wineries make vintage-dated products, such as 2004 Merlot, and do not blend in wine
from earlier years. Wineries have limited storage capacities, especially at intermediate
inventory points such as tanks, which must be emptied before the next year’s harvest can
be processed. Wine is a high value product and is expensive to keep long term. Therefore,
we can treat the yearly production and sales cycle as a single period model. Swaminathan
and Tayur’s study (1998) suggests that results would differ little if the additional
complexity of multiple periods were included.
The last three types of variables relate directly to market conditions: xfc,s, sales of
product at full price, xsc,s, sales at salvage value, and xvc,s, any demand shortfall
(violation) for each sales channel. Equation (5) ensures that no more finished goods are
sold at full price than were demanded in the channel:
(5)
xfc,s  Dc,s c  C, s  S
Conversely, minimum demand requirements are represented in equation (6) by
allowing the winery to under-deliver to demand, but recording the shortfall:
(6)
xvc,s  Dc,s - xfc,s c  C, s  S
The model currently assigns non-zero fees only for shortfalls of private label products, as
these are likely to be subject to stricter contracts. A winery experiencing an unexpected
demand surge across multiple channels may thus need to devote more wine to a private
label product, even though it likely has lower margins than the winery’s own brand.
Wine in each finished goods channel will either be sold at full price, Rc, or at
salvage value, Sc. Aside from tarnishing brand perception, salvaging wine is undesirable
because lower revenues are earned. Private label wines bear another company’s brand,
preventing a winery from receiving a non-zero salvage price. The final balance equation
(7) guarantees that all wine that is either initially finished or later transformed is then
sold. Excess channel supply must become salvage and cannot be repackaged or
repurposed to different channels:
(7)
xpc   xti ,c ,s  xfc ,s  xsc ,s c  C, s  S
iI
The objective is to maximise expected profits across all channels. While the
objective function (8) may appear complex, it can be broken down into component cost
and revenue terms. Deterministic costs include the production cost, G, and packaging
costs, Lc, of the initial allocation of finished goods and production cost and storage costs,
Wi, for intermediate goods. All other terms involve the dimension of demand scenario.
These include the costs associated with transforming intermediate inventories into the
desired finished goods, necessitating labelling/packaging costs, supplemented by the
additional costs, Ai,c, associated with postponement. Revenues are collected from multiple
sources: the sale of finished goods at full price and at salvage value. Product that is
dumped before being finished may receive a subsidy, Bi, although current research
focuses on US wineries, and we have set this term to zero. Violations are assessed fees,
Fc, in channels where demand shortfalls are considered a breach of contract.
7


max  Ps    Rc xf c , s  Sc xsc , s  Fc xvc , s    Bi xd i , s 
sS
iI
 cC



    Lc  G  xpc +  Wi  G xii 
iI
 cC

(8)
  Ps   Ai ,c  Lc xti ,c , s
sS
cC iI
The use of discrete probability distributions allows this problem to be represented
as a deterministic equivalent linear program. The model is implemented in the GAMS
modelling environment utilising MINOS as the solver. With 4 finished goods channels, 2
intermediate inventory points and 7 demand scenarios, the model has 161 variables in
total: 6 first-stage variables and 154 recourse variables, in addition to z, the total expected
profit. The enumeration of all 6 types of constraints results in 156 equations. As the
model is linear, solution times are sub-second and not of concern.
4. Results
4.1. Quantifying the profitability from postponement
The focus of our research to date has been in constructing and validating the conceptual
model. Although we interviewed several winemakers to understand their use and
perceptions of postponement, we do not have definitive values for all of the model
parameters. Depending on size, marketing efforts, wine quality and other factors, each
winery will have different parameter values. Appendix A shows the numerical values
used and indicates where validation has occurred or what the rational is for certain
values. For instance, channel revenues and costs have been structured so that the per-case
cost and profit margins of the branded channel are the highest, followed in decreasing
order by both Private Label products and bulk sales.
Even the use of placeholder data provides results worth studying. While most of
the decision variables are recourse variables, these are of less interest because
determining which products to finish and what revenues can be obtained is
straightforward once channel demands are known. The variables to examine are those
determined in the first stage; these control the degree and type of postponement used.
Figure 3 shows the optimal level for the first-stage variables for two different
situations: one where postponement is allowed and one where it is forbidden. When
postponement is allowed, it is optimal to delay the final finishing of 25% of the total
production, with two-thirds of this intermediate inventory allocated to the tanks. Without
postponement we would instead: 1) bottle more Private Label A, and risk oversupply
without any revenues from channel excess if demand were low and 2) sell more bulk
wine, sacrificing the revenues that would be earned in profitable channels were demand
higher. The potential value of postponement can be quantified by comparing optimised
expected profits from when postponement is allowed to when postponement is forbidden.
The use of postponement gives our hypothetical winery an expected increase in product
profitability of 8%, a considerable improvement.
8
4.2. Additional situational analysis: sensitivity of results to probability assumptions
The reader may question why we have selected the particular probability
distribution for the seven scenarios. Scenario probabilities are one of the least qualified
parameters in the model and are not easy to validate. To address this concern, we define
three additional situations, changing the probability distribution of the demand scenarios,
as shown in figure 4. For example, the “Expect high demand” situation assigns a greater
chance of larger demands on some or all of the channels, although still allowing a slight
possibility for channel demands to be low.
Optimisation results from these additional situations were compared to those from
the standard situation. Figure 5 shows that both packaging and labelling postponement
occur in all 4 situations. The differences are in the amount allocated to each of the
finished goods channels and intermediate inventory points. If the lower demand scenarios
have a greater likelihood, more postponement would be utilised; the fraction of
production allocated to intermediate inventory increases from 25% to 33% from the
standard situation to that of “Expect low demand.” Postponement usage did not decrease
in “Expect high demand,” but initial allocations changed from a guaranteed sale in the
bulk channel to a higher-margin but riskier channel, Private Label B.
Another result seen in figure 5 is that the usage of postponement should increase
when the producer has even less of an idea of what demand is likely to be. The uncertain
situation considers all scenarios equally likely and results in 42% of production being
held as intermediate inventory, the majority in tanks, which provide even more versatility
than blanks for transformation into finished goods.
We can extend these results to make additional conclusions. If channel demands
are less correlated than assumed, and scenarios such as simultaneous low demand for
Private Label A and high demand for Private Label B are considered to be possible or
even likely, benefits from postponement can only increase. Use of postponement to
manage product differentiation is analogous to using risk pooling to manage safety
inventory, and Chopra and Meindl (2004) show that products with lower demand
correlation will experience even greater savings from aggregation techniques.
4.3. Model results compared with practice
We interviewed operations managers at four US wineries of different sizes to compare
our theoretical results with actual industry practices. Interviews were difficult to procure
as most wineries are private firms concerned with maintaining their competitive
advantage, not to mention the consumers’ perceptions that wine is an artesian product.
They were reluctant to divulge operation details, and two asked to remain anonymous.
One of the unidentified wineries is a very large producer that supplies wines
within a wide price range from several regions, but primarily from California’s Central
Valley. Intermediate inventory is stored as blanks, as it can be used to source two
different product lines, both of which are brands owned by the winery. This process is
estimated to increase costs by approximately $1 per case and is only used for an
unspecified fraction of the two brands’ total production. No instances of postponement at
the tank level were identified.
9
The JanKris winery in Paso Robles produces 25,000 cases annually of premium
wines priced between $7 to $14 per bottle. This winery has been growing for the past few
years and has actively expanded their product line, including blending a private label
wine for a large client. JanKris stores inventory in both the blank and tank phase to
allocate to different products as needed.
A small, unnamed boutique winery in Napa Valley produces 5000 cases a year of
wines priced above $14 per bottle. Before 1999, this winery would defer the labelling of
a significant part of their annual production because they were uncertain whether they
would be able to sell everything under their own label. However, they have since
discontinued this practice as demand has increased, effectively guaranteeing that they can
sell all their wine upon release.
The sole winery in this group of four that has never used postponement is Pine
Ridge. This Napa Valley winery produces 50,000 cases annually of wine priced above
$14 per bottle. Pine Ridge’s operations manager claims that deferring bottling has the
potential to decrease wine quality. It should be noted that Pine Ridge does not offer any
private label products and thus would benefit little from postponement.
Although far from an exhaustive survey, these interviews help to confirm results
from the theoretical model. Postponement is of less value to wineries that sell through a
single sales channel and that have consistent, high demand for their products. But for
wineries that do not have the luxury of being in this market position and for those that are
large or offer multiple product lines, postponement strategies can be useful. We found
more instances of postponement further downstream, at the labelling level, than at the
packaging level. Our model conversely favours more usage of tanks than blanks, but we
would be remiss to draw further conclusions without more data points.
4.4. Suggestions for future research
Foremost, additional research should include surveying more winemakers. We believe it
would also be illuminating to compare postponement practices in different countries.
European wineries may utilise postponement more as their greater emphasis on exporting
necessitates specific labels for many markets. Additionally, European markets sell more
store-branded wines, so local wineries are also more likely to source private label
products. With both the practice of exporting and the presence of private label wines
expected to increase in the US, it behoves American wineries to study the best practices
of their peers abroad.
Another valuable effort would be to perform in-depth case studies of different
wineries. Optimising the model using parameters, scenarios, and probabilities that have
been provided by a specific winery would provide a customised postponement strategy.
We could then compare results from different wineries to find more robust estimates for
standard model parameters, validate model assumptions, and ultimately better quantify
the potential of postponement across the industry as a whole.
For wineries that offer both blended and single varietal wines, a possible model
extension would consider optimising the components of the blends based on evolving
demands. Wineries often grow their own grapes or enter into long term purchasing
contracts, so varietal (type of grape) supply availability typically changes more slowly
than consumer trends. Recipes for blended wines could be adjusted to use more of the
10
less desired grapes varieties, saving the popular ones for sale as single varietal wines.
This extension would allow the use of a third type of form postponement: that of
assembly.
5. Conclusion
US wineries face an increasingly competitive, globalised marketplace. Many can no
longer expect to sell all the wine they produce under their own brand. As private label
wines gain popularity, more wineries will become sources and must determine how to
allocate production across sales channels. Producers who can achieve this balance
through techniques such as postponement will be at a competitive advantage.
Our stochastic programming model demonstrates that using form postponement at
the packaging and labelling stages can increase expected profits. Our hypothetical winery
had an 8% expected improvement in product profitability by postponing the finishing of
25% of production. What-if analysis shows that postponement usage increases if demand
is more likely to be lower or if wineries are unable to estimate channel demand
probabilities.
The competitive advantage of using postponement has still to be better qualified.
As every winery has different supply and demand characteristics, the optimal
postponement strategy will not be uniform across the industry. We recommend
performing case studies for a few wineries. We also suggest investigating postponement
practices of more wineries from around the globe and extending the model to optimise
the recipes of blended wines.
11
Appendix A: Model Parameters
Cost and Production Parameters
C
G
Lc
Represents a small winery
Production costs are highly variable, but Folwell and Castaldi (1987)
Pre-bottling cost of producing wine
$
30 suggest an inflation adjusted $4-$5/bottle for medium wineries, net
per case
of packaging and labelling
Marketing, bottling and labeling cost per case equivalent, by channel
Total annual case production
12000
brand $
pl-a $
pl-b $
bulk $
Wi
18
12 Costs are structured so that they are higher for brand than private
9 label. These costs are by definition zero for bulk wine
-
Per case packaging and storage costs for intermediate inventory
tanks $
1 Tank costs are for storage. Blank costs are higher as they include
4 both the material and processing costs of bottling
blanks $
A i,c
Additional per case costs that come with the transformation of intermediate inventory i to sales channel c
tanks
brand $
8
$
pl-a $
8
$
pl-b $
7
$
-
$
-
bulk $
T i,c
blanks
-
(1) The negative for "blanks" reflects the fact that bottling
(1) (and its associated costs) will have already occurred.
Postponement will still be more expensive than initial
allocation, by $3 case more from blanks.
The [0,1] matrix mapping the transformation of intermediate inventory i into finished goods for sales channel c
tanks
blanks
brand
1
1
pl-a
1
pl-b
1
1 1 = a legal transformation
0 0 = a forbidden transformation
bulk
1
0
Demand and Pricing Parameters
Rc
Per case full-price revenue by sales channel
brand $
Sc
pl-a $
78 Revenues will vary with winery size and reputation, but are
54 structured so that they are higher for brand than private labels, and
pl-b $
45 that when costs are included, profit by case is highest for brand and
bulk $
33
Per case salvage value by sales channel
brand $
Fc
36
pl-a $
-
pl-b $
-
bulk $
-
Only branded wine has salvage value. These salvage values are
constructed so that products sell at a loss.
Per case fee for shorting demand by sales channel
brand $
pl-a $
pl-b $
bulk $
Bi
lowest for bulk.
Fees are structured so that only private label wines have fees, and
10 Private Label A has higher fees than Private Label B. These are per
8 unit fees, whether the shortage is 10 cases or 10000 cases
-
Subsidy (buy-back) for dumping wine at intermediate inventory points
blanks $
-
tanks $
-
No subsidies are used at present, as U.S. wineries do not receive
payments for dumping.
Ps
Probability of scenario s
D c,s
Per case demand across channels for scenarios: = P s *Dem_base, where Dem_base is below
Dem_base
These values are shown in Figure 2
brand
pl-a
pl-b
bulk
5000 Demand is structured so that Brand still repesents the primary
2000 channel, but bulk wine is an unsaturated market. These parameters
2000 were used to so that under many (but not all) demand scenarios, the
12000 winery would have to sell to all 4 channels.
12
References
Berger, D., Sam’s Club: A new direction in wine, April 2nd, 2004, Available online at:
http://www.adamsbevgroup.com/bd/2004/0402/0402wnbs.asp (accessed 5 January, 2006)
Birge, J. R. and Louveaux, F., Introduction to Stochastic Programming, 1997 (SpringerVerlag: New York).
Chaney, I. M., Own-label in the UK grocery market, International Journal of Wine
Marketing, 2004, 16(3), 5-14.
Chopra, S. and Meindl, P. Supply Chain Management: Strategy, Planning and Operation,
2004 (Prentice Hall: Upper Saddle River, New Jersey).
Dantzig, G.B., Linear programming under uncertainty, Management Science, 1955,
1(3&4), 197-206.
Folwell, R. J. and Castaldi, M.A., Economics of size in wineries and impacts of pricing
and product mix decisions, Agribusiness, 1987, 3(3), 281-294.
Lee, H.L., Billington, C. and Carter, B., Hewlett-Packard gains control of inventory and
services through design for localization, Interfaces, 23, 11.
Popp, J., Private label makes its mark, Beverage Industry, 2003, 94(6), 19.
Orr, S., The role of capacity management in manufacturing strategy: experiences from
the Australian wine industry, Technology Analysis & Strategic Management, 1999, 11(1),
45-53.
Skipworth, H. and Harrison, A., Implications of form postponement to manufacturing: a
case study, International Journal of Production Research, 2004, 42(10), 2063-2081.
Swaminathan, J. and Tayur, S. Managing product lines through delayed differentiation
using vanilla boxes, Management Science, 1998, 44(12), 0161-0172.
Swaminathan, J. and Tayur, S., Stochastic programming models for managing product
variety, In Quantitative Models for Supply Chain Management (International Series in
Operations Research & Management Science) edited by S. Tayur, 585-624, 1999
(Kluwer Academic Publishers: Norwell, Massachusetts).
Twede, D., Clarke R.H., and Tait, J.A., Packaging postponement: a global packaging
strategy, Packaging Technology and Science, 2000, 13, 105-115.
Van Hoek, R.I., Postponed manufacturing: a case study in the food supply chain, Supply
Chain Management, 1997, 2(2), 63-75.
Van Hoek, R.I., Logistics and virtual integration: postponement, outsourcing and the flow
13
of information, International Journal of Physical Distribution & Logistics Management,
1998, 28(7), 508-523.
Van Hoek, R.I., Commandeur, H.R., and Vos, B., Reconfiguring logistics systems
through postponement strategies, Journal of Business Logistics, 1998, 19(1), 33-54.
Van Hoek, R.I., Postponement and the reconfiguration challenge for food supply chains,
Supply Chain Management, 1999, 4(1), 18-34.
Van Hoek R.I., The rediscovery of postponement; a literature review and directions for
research, Journal of Operations Management, 2001, 19, 161-184.
Ward, M. B., Shimshack, J.P., Perloff, J.M. and Harris, J.M., Effects of the private-label
invasion in food industries, American Journal of Agricultural Economics, 2002, 84, 961973.
Yang, B., Burns, N.D. and Backhouse, C.J., Management of uncertainty through
postponement, International Journal of Production Research, 2004, 42(6), 1049-1064.
Yang, B., Burns, N.D. and Backhouse, C.J., An empirical investigation into the barriers
to postponement, International Journal of Production Research, 2005, 43(5), 991-1005.
Yu, C.S. and Li, H.L., A robust optimization model for stochastic logistic problems,
International Journal of Production Economics, 64(2000), 385-397.
Zinn, W. and Bowersox, D.J., Planning physical distribution with the principle of
postponement, Journal of Business Logistics, 1988, 9(2), 117-136.
14
Titles and Captions for Figures
Figure 1
Title: Overview of post-blending production
Caption: Arrows denote allowed transitions between stages in the process of
creating finished goods for all channels. Intermediate inventory points are
underlined.
Figure 2
Title: Demand scenarios: percentage of base demand by sales channel
Caption: Demands for the other six scenarios are referenced to the base
scenario.
Figure 3
Title: Allocation of production when postponement is allowed or forbidden
Caption: The first four categories represent the initial allocations to finished
products, and the last two show allocations to intermediate inventory points.
Figure 4
Title: Probability distributions for the different situations
Caption: While the situations have different probabilities for the scenarios, all
scenarios still have a possibility of occurring.
Figure 5
Title: Summary of first-stage decision variables by situation
Caption: Postponement utilisation is shown in the last two columns.
15
Figure 1
Branded
channel
Tanked wine
Bottling
process
Blank
wine
Labeling
process
Private
Label A
channel
Private
Label B
channel
Bulk wine channel
Unlabeled
Pl-B brand
16
Figure 2
160%
140%
120%
100%
80%
60%
40%
20%
0%
brand
pl-a
pl-b
s1: base
s2: high all
s3: high
brand
s4: high PL s5: low all
s6: low
brand
s7: low PL
17
Figure 3
45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
postponement
forbidden
postponement
allowed
xp.brand
xp.pl-a
xp.pl-b
xp.bulk
xi.tanks
xi.blanks
18
Figure 4
30%
Standard
situation
Expect low
demand
Expect high
demand
Uncertain
situation
25%
20%
15%
10%
5%
0%
s1: s2: high s3: high s4: high s5: low s6: low s7: low
base
all
brand
PL
all
brand
PL
19
Figure 5
45%
40%
35%
30%
25%
20%
15%
10%
5%
0%
xp.brand
xp.pl-a
xp.pl-b
xp.bulk
xi.tanks
xi.blanks
Standard situation
Expect low
demand
Expect high
demand
Uncertain
situation
20