See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/317941763

Price differentiated channel switching in a fixed period fast fashion supply

chain

Article  in  International Journal of Production Economics · June 2017

DOI: 10.1016/j.ijpe.2017.06.030

CITATIONS READS
3 404

3 authors, including:

Sevilay Onal
New Jersey Institute of Technology
4 PUBLICATIONS   12 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Sevilay Onal on 13 September 2018.

The user has requested enhancement of the downloaded file.

 International Journal of Production Economics 193 (2017) 31–39

Contents lists available at ScienceDirect

International Journal of Production Economics

journal homepage: www.elsevier.com/locate/ijpe

Price differentiated channel switching in a fixed period fast fashion

supply chain
Jingran Zhang, Sevilay Onal, Sanchoy Das *
Department of Mechanical & Industrial Engineering, New Jersey Institute of Technology, Newark, NJ 07102, United States

A R T I C L E I N F O A B S T R A C T

Keywords: Fashion products are characterized by short product life cycles and high market success uncertainty. An unsuc-
Supply chain cessful product requires multiple price discounts to clear inventory. Fast-fashion retailers use a variety of stra-
Fast fashion tegies to counter these uncertainties including: frequently changing collections, quick response sourcing, and data
Channel switching
driven channel switching. This paper proposes a switching solution for the Fast Fashion Supply Chain (FFS) of
retailers who have preordered an initial or block inventory, and plan to use channel switching as opposed to
multiple discounting steps. We consider a retailer which ships product from a central warehouse to store/
clearance and outlet channels. The FFS Multi-Channel Switching (MCS) problem then is to monitor market de-
mand data, such that at the optimal period the store inventory is switched to clearance, and the warehouse in-
ventory is switched to the outlet channel. Using a linear projection of the moving average demand trend, we
estimate the remaining cycle revenue at any time in the cycle. Channel prices are fixed, and the remaining
revenue is a concave function of switching time. Using a set of conditions the objective is further simplified into
cases. The Linear Moving Average Trend (LMAT) heuristic prescribes whether a channel switch should be made in
the next period. The LMAT was compared with the optimal decision, the No-Switch rule and an intuitive heuristic
using a simulation experiment. The LMAT performed well and for most problems provided a solution within 0.4%
of the optimal. Confirming the LMAT can readily and effectively be applied to real time decision making in a FFS.

1. Introduction response sourcing of products (Iyer and Bergen, 1997), and/or demand
data driven placement of products in the appropriate retail channel. This
Rapid changes in consumer buying behavior coupled with supply paper considers the last strategy, whereby the retailer is able to use real-
chain redesigns, has motivated both existing and new retailers to adopt a time market demand data to replenish the stores and sequentially switch
variety of new inventory management strategies. These include omni- product inventory to alternate retail channels. In a more traditional
channel and multi-channel retailing which combines outlet and online environment such decisions would be made on sales forecasts, such that
stores with regular stores (Melis et al., 2015). Here we focus specifically store allocations are predetermined. When a product is unsuccessful then
on fashion goods, which are characterized by a short life cycle, high sequential markdown are the primary method to dispose the inventory.
customer demand uncertainty, long supply lead times, and high price When the product is held back in the warehouse, then a data driven
discounting after the regular selling period (Huang et al., 2014). strategy would switch the store channel off and the outlet channel on.
Generally supply chains are concerned with demand variance, or periods Şen (2008) characterizes the fashion industry as having short product
of low demand followed by periods of high demand. In fashion the un- life cycles, volatile and unpredictable demand, and long and inflexible
certainty is whether the product will or will not be a success. An un- supply processes. They note that a fashion product is typically defined by
successful product will then require multiple price discounts to clear the a 10 week life cycle, with five selling seasons per year. Choi (2007)
inventory. A new generation of retailers (e.g. Zara and H&M) have suc- observe that demand uncertainty is a crucial and prominent factor in the
cessfully developed and implemented fast-fashion supply chains (FFS) in stocking and pricing decisions for fashion retailers. In a traditional supply
an effort to reduce the use of discounting. These combine a variety of chain products are ordered several months in advance, and the success
strategies including, frequently changing low-cost clothing collections risk rests primarily with the retailer. In the more well-known FFS solu-
that mimic current luxury fashion trends (Joy et al., 2012), quick tion, as practiced by Zara, quick response sourcing is a key element of

* Corresponding author.
E-mail address: das@njit.edu (S. Das).

http://dx.doi.org/10.1016/j.ijpe.2017.06.030
Received 9 November 2016; Received in revised form 22 June 2017; Accepted 23 June 2017
Available online 27 June 2017
0925-5273/© 2017 Elsevier B.V. All rights reserved.
J. Zhang et al. International Journal of Production Economics 193 (2017) 31–39

inventory risk reduction. Supply lead times tend to be relatively short example of an FFS model and key aspects are reported by Ghemawat and
such that the store is able to manipulate and align supply quantitates in Nueno (2003). They observe that the FFS operations strategy combines
response to the observed product demand. Stratton and Warburton two critical features: (i) quick response production capabilities and (ii)
(2003) found that fashion retailers prefer to assign fast-response products enhanced product design capabilities (Cachon and Swinney, 2011). Caro
to local manufacturing in order to achiever shorter lead times. When this and Gallien (2010) found the Zara supply chain incorporates a fore-
is not possible, then an alternative solution is to preorder a large quantity casting model which would prescribe the initial block inventory, and an
to meet the expected demand for the selling cycle. This is termed the optimization model to control the warehouse to store replenishment once
wholesale purchase quantity and set by the store buyer or merchandiser. actual sales data is tracked. Iyer and Bergen (1997) discuss quick
Since market success for fashion products is difficult to predict, the or- response supply from manufacturing to retailer channels in general,
dered quantity is often not sold at regular prices and has to be discounted. while Cachon and Swinney (2009) give a detailed explanation about the
An FFS strategy then is to plan for a shorter product selling cycle, with strategic customer behavior under quick response. Huang et al. (2014)
more frequent style turnovers, so as to reduce but not eliminate the derive a dynamic pricing model with partial backlogging to investigate
discounting risk. Products are then switched between channels in a the important factors that influence the replenishment cycle and profit.
market responsive strategy. Caro and Gallien (2012) and Karakul (2008) show that regular demand
This paper proposes a multi-channel switching (MCS) solution for behavior is a function of price and age of the product while clearance or
fast-fashion retailers who have preordered an initial or block inventory. discounted price is more difficult to manipulate. From discussions with
These retailers plan to use channel switching as an alternative to multiple leading fashion retailers Choi (2007) found that many use a two-stage
discounting steps. We consider the case where the retailer operates a stocking policy, whereby an initial block inventory is supplemented
centralized warehouse from which product is supplied to multiple stores with a second stocking order using actual demand data. Pricing decisions
plus several retail outlets. In a multi-channel setting the outlet could be were also made similarly. In these papers a primary focus is on real time
an online store, where orders are fulfilled directly from the warehouse. inventory driven replenishment policy. Our model is similar in that we
The multi-channel model incorporates multiple retail channels which also consider the current store inventory and also generate a prediction of
may or may not operate simultaneously. Here we are restricted to three future demand. The proposed model though differs in that we consider
sequential channels. At the start of the selling cycle the wholesale pur- the case where we can switch to an alternate channel which is price
chase quantity is delivered to the warehouse, from which small quantity differentiated with stable demand.
replenishments are made to the stores. Product is sold in three sequential
pricing channels with no overlap: regular store price, clearance store 2.2. Multi-channel distribution and multi-period retailing
price, and outlet price. This is equivalent to a dynamic pricing model but
limited to only two predetermined price steps. For the successful product In today's retailing environment retailers are leveraging their supply
case demand remains strong through the selling cycle, and all the in- chains to expand sales volume and profit beyond their traditional store
ventory is sold in the regular channel. For the unsuccessful product case channels (Chiang et al., 2003; Ding et al., 2016). Several researchers have
demand weakens early on and a significant portion of the inventory is broadly studied customer behavior differences across channels and spe-
sold through clearance and outlet. The FFS Multi-Channel Switching cifically looked at channel adoption, channel choice and usage (Verhoef
(MCS) problem then is to monitor real-time demand and store inventory, et al., 2015). Innovations in retail promotions and expansion of outlet
such that at the optimal period the remaining store inventory is sold at malls are providing new retail channels that are readily integrated into a
clearance, and the warehouse inventory is switched to the outlet channel. multi-channel distribution strategy. Specifically dynamic pricing com-
The MCS objective is to maximize the total revenue. Using a linear bined with targeted promotions can be used to effectively and quickly sell
projection of the moving average demand trend, we estimate the excess inventory (Grewal et al., 2011). Coughlan and Soberman (2005)
remaining cycle revenue at any time in the cycle. This estimate is shown present an analysis of two possible structures of dual-distribution
to be a concave function of the switching time. Using a set of conditions through both regular retailer channel and outlet channel. One option is
the objective is further simplified. A heuristic solution is then developed to sell in multiple channels simultaneously. Alternatively, the manufac-
to determine whether a channel switch should be made in the next turer or retailer can make sequential decisions in two or three channels.
period. A simulation analysis on a set of experimental problems with The identify possible decisions as (i) how much to distribute to a primary
different demand behaviors is reported. These shows that the heuristic regular store channel, and (ii) whether or not to add an outlet into the
provides solutions within 2% of the optimal solution. The MCS problem distribution mix. The current omni-channel literature assumes parallel
has not been discussed previously in the research. We present a structural channels, our model differs in that we consider sequential channels with
definition of the problem and introduce the key parameters and decision each displaying a different demand behavior. Our models differ in that
variables. The model is shown to complement other FFS solution the channels operate sequentially with no overlap, such that customers
methods, which focus on pricing and inventory allocation to stores. The cannot exploit one channel against the other.
MCS demonstrates the utility of channel switching to retailers tran- Two-period pricing models are widely studied in the literature, most
sitioning to the FFS mode. Specifically it shows how they can leverage a of these consider the price to be the decision variable (Zhang et al.,
price differentiated outlet or online channel to maximize revenues from 2014). Zhou et al. (2015) consider a two period pricing model for
less successful products. launching fashion products. Three strategies are identified one of which
is labelled the S-Strategy: that is the firm launches a new style and stops
2. Background selling the previous one immediately. This operationally equivalent to
the model developed here, in that the old design is shifted to another
Several topics and problems related to FFS supply chains have been channel, so that the high value store channel is immediately focused on
addressed in the literature. In the following subsections, we review two the new product. Similar to this research they observe that luxury re-
of these in the context of this research. tailers will sell then their discontinued styles in their outlet stores. Here
we consider the price to fixed and decide on the switch time. Khouja et al.
2.1. Fast fashion and quick response supply (2010) analyze channel selection and price setting of a manufacturer or
retailer with several channel options. Most of the research is focused on
Şen (2008) provides an extensive review about the US fashion in- the consumer pricing behavior, and assume a known price demand
dustry and the supply chain driving it. They note that a quick response relationship. Here the demand is assumed to unknown, and channels
retailer will track sales at the store-level on a real-time basis, and decisions are made in real-time using tracked demand data. Others have
maintain minimal inventories at the store. Zara is the most prominent considered channel entry decision, most commonly an online or direct

32
J. Zhang et al. International Journal of Production Economics 193 (2017) 31–39

channel in addition to regular retail (Wang et al., 2016). These, though,

usually are not readily applicable to short life fashion products. A key
difference in our model is that pricing and consequently the channel
differentiation is fixed. The current literature focuses on manipulating
price within the channel to stimulate demand. We consider the case
where the retailer attempts to eliminate discount speculation from cus-
tomers, and uses the second channel to clear out the slow mov-
ing inventory.

3. Problem formulation
Fig. 2. Three selling sequential channels.
We consider a retailer selling a single fast fashion product through
stores which are restocked from a central warehouse. Excess inventory is the store inventory when demand drops and should not occupy prime
sold through an outlet channel which is also supplied from the same retail space for long. An aggressive mark-down then allows for quick sale
warehouse. We assume a single retail store and a single outlet store of the residual inventory at the store. Since PO > PC, a retailer with an
without loss of generality. The FFS strategy of the retailer is described by established outlet channel can leverage it as an attractive option when
two attributes: compared to clearance sales. Outlet channels are known to attract price-
sensitive, non-service-sensitive consumers compared with regular
T The selling cycle, after which the product will no longer be sold retailer channels (Coughlan and Soberman, 2005). By making small and
Π The initial product inventory or block quantity available for sale in frequent replenishments to the store, the inventory risk and hence
period T clearance can be minimized. Setting the prices can be viewed as a tactical
decision and the retailer may decide these prices before product intro-
The inventory movements during the selling cycle are described in duction using demand and consumer buying behavior as a basis. Oper-
Fig. 1. We assume the store is restocked using a classical base stock ationally then, the management decision is when to switch between the
policy. The block inventory is sold through three sequential retails channels as actual demand data comes in.
channels with no overlap. Any residual inventory after T is assumed to be Fig. 2 illustrates the product flow and the associated switching points
unsold and have no revenue value. Though we assume the initial block in the FFS retail cycle. The objective of the MCS problem then is to
inventory is known and fixed the assumption is not restrictive. For a maximize the revenue by making the following two switching decisions:
successful product the retailer may use its fast supply chain to increase Π.
Once demand starts waning then it's unlikely that Π will be increased TC Time at which the store switches from regular to clearance price
further and becomes fixed. The switching decision becomes relevant TO Time at which store stops selling and all warehouse inventory is
once demand is declining and the final Π then become the model assigned to the outlet for immediate sale.
parameter. For an unsuccessful or low success product the initial and At TO any remaining store inventory that could not be sold at price PC
final Π will be the same. is destroyed, that is there no back shipment to the warehouse. Since
The first two channels are collocated at the store, while the third success for fashion products is unpredictable, switching decisions
channel could be either a physical or online outlet channel. Since Π is must leverage real-time market demand information.
fixed the product sourcing cost is fixed and not effected by any subse-
quent decisions. The internet has enabled price transparency and fast
fashion retailers are aware that customers are immediately alerted if the
product is available at lower prices at a simultaneous channel. This 3.1. The demand behavior
motivates the exclusive channel distribution policy at any time t. For each
item sold through the three channels the revenue price is assumed to be The primary uncertainty in the FFS problem is product demand, first
known and specified by the product merchandiser as follows: whether the product will be successful or not and then the rate at which
the demand will fade. Projecting demand for fashion products is in
PR Regular unit retail price for items sold at the store general a difficult task, and the behavior is best predicted from the actual
PC Clearance unit price for items sold at a store promotion sales data. Increasingly, customers are becoming forward looking, and
PO Outlet unit retail price for items sold through the outlet when products are continuously discounted they are able to predict a
future price from experience data. Customers arrive at the store at the
The regular channel has the highest price and in the best case scenario beginning of the selling cycle, observe the selling price PR and decide to
the entire block inventory is sold in this channel. We assume the pricing whether purchase it immediately or delay the purchase anticipating
relationship PR > PO > PC holds. Clearance sales are intended to clear out future discounts. Caro and Gallien (2010) observe that a FFS strategy can

Fig. 1. Life cycle of a fast fashion product.

33
J. Zhang et al. International Journal of Production Economics 193 (2017) 31–39

disrupt this behavior by limiting the discount steps and percentages. This demand is more stable and here it is assumed to be constant. The outlet
allows the retailer to limit the demand uncertainty caused by multiple demand is then constant and given by βA0. In general both α and β will be
pricing discounts. less than 1, though in a deep discounting situation they could be greater
Here the demand behavior is not restricted to any pre specified dis- than 1.
tribution, and the underlying demand behavior is unknown. Rather all
decisions are based on the actual trailing demand as recorded at the store. 3.2. The MCS objective function
It is assumed, though, that demand in the regular price channel starts
with a period of rising trend which is followed by a period of decreasing The block inventory is predetermined by marketing, and an input
trend. The model does not allow for a trend reversal once a declining parameter to the MCS problem. Then, since the initial supply costs are
trend is confirmed. Fig. 3 illustrates a sample demand behavior for fixed, the MCS objective is to maximize profit which is equivalent to
different rates of decline starting from the same initial demand. For a maximizing revenues. As described above the product is sold in three
successful product the demand rises steeply and then declines at a very channels, let NR be the total sales in the regular price channel, NC be the
slow rate. At the end of T demand is still strong, and likely the entire stock total sales in the clearance price channel, NO be the total sales in the
Π is depleted, indicating no need to make channel switches. For an un- outlet price channel, and NW be the unsold or salvage inventory at the
successful product the demand rises slowly and then starts to drop end of the selling cycle. Then for a given {Π,T} the MCS problem
quickly, such that demand is zero long before T. Clearly, at some point objective is:
sales should have shifted to clearance and then outlet sales.
The literature on the direct relationship between demand under Maximize Total Revenue: ϕ ¼ PR NR þ PC NC þ PO NO
(1)
regular and clearance or outlet pricing for fashion products is somewhat Such that: NR þ NC þ NO þ NW ¼ π
limited. Smith et al. (1994) and Caro and Gallien (2012) establish a
forecasting model for the fast fashion industry. They studied the case of where: 0 # TC # T, 0 # TO # T, and TC # TO
Zara, where the dependent variable is the demand rate of a specific The store is restocked using a (Q, R) base stock policy. Ideally Q is not
product over a finite period and the regression includes multiple pa- very large, so that the inventory risk at the store is minimized. We ignore
rameters such as product introduction time, current inventory levels, and the shipping cost of replenishments to the store and to the outlet. For the
competing products. They find customers are more sensitive to the regular retail channel the number of restocks will be determined by the
relative markdown than to the absolute price cut. By using the non- (Q, R) policy and independent of the switching decision. A large Q and/
changed coefficients of each term from the regression data of regular or R could increase the store inventory at the time of dropping sales,
price sales, we can estimate customer behavior during the clearance which would increase NC or sales in the lowest unit revenues channel.
period. Similarly, demand behavior in outlets, is influenced by many The holding cost is a function of the average inventory both in the
factors including relative price discounting, self-satisfaction of the warehouse and stores. Since the switching strategy in effect attempts to
shopping experience and brand image. It is well documented that outlet clear out the inventory with T the inventory cost would also be mini-
malls provide a shopping arena in which deals are available constantly mized. In an extended objective function the restocking and inventory
(Sierra and Hyman, 2011). Therefore, for a specific product, we expect costs could be modelled. The risk of lost sales is also disregarded here.
that in a finite selling cycle the outlet channel operates with constant
demand from a stabilized customer group, at a fixed discount level. 4. Solution method
Clearly this requires that the outlet price discount is set by management
at an attractive level to sustain the constant demand. The nominal solution to the MCS problem is to do nothing, that is
Fig. 4 shows the demand behavior when switching decisions are TC ¼ TO ¼ T, which implies we sell what we can at regular price and the
made. For modelling purposes we assume the clearance demand follows remaining inventory is wasted. In an optimal solution to the MCS prob-
the same pattern as that exhibited by the regular demand. Let At be the lem, though, TC # TO # T. A closed form solution to this problem is not
actual demand at time t, then the initial clearance demand is estimated as feasible since the demand behavior is uncertain at any point in the selling
(1þα)ATc, where α is the estimated increase in demand as the price is cycle. Note that any time during the retail cycle the future sales are
discounted from PR to PC. There are a wide range of pricing-demand projections, and therefore ϕ is also a projection. We propose a heuristic
models and our approach is that these will determine α. For example solution to the MCS problem, and make the following assumptions:
consider Choi (2007) model demand as a linear function of the con-
sumers' price sensitivity, and the regular or ‘‘normal’’ price. Outlet 1. At any time t, a linear trend model provides a reliable forecast of the
regular demand. The slope of the future trend is estimated by an N-

Fig. 3. Demand behavior scenarios for a fast fashion product.

34
J. Zhang et al. International Journal of Production Economics 193 (2017) 31–39

Fig. 4. Demand behavior in clearance and outlet channels.

Period moving average slope of the training demand. Let At be the Given the above assumptions, the problem reduces to a single deci-
actual observed demand in period t, then the N-Period slope at time t sion τ with τ ¼ TC, and in many cases τ ¼ TO ¼ TC implying no clearance
is: sales. The Linear Moving Average Trend (LMAT) Heuristic is proposed as
a solution to the MCS problem. The motivation for the LMAT heuristic,
At $ At$N centers on the first assumption. Similar to a classical moving average
δt ¼
N forecasting method, the expectation is that the past N-Period trend is a
Initially, δt will be positive, but the switching decision problem be- reliable indicator of future sales in the regular channel. Note that
comes relevant only after δt turns negative. By using the N-period moving switching is likely to occur in the latter part of the demand cycle when
average we dampen the effects of the demand change rate, similar to a the primary demand drop has already occurred. This N-Period trend line
classical moving average forecast. The forecasted regular demand for a then provides an estimate of the likely remaining revenues in the regular
future period τ then is: channel, allowing for comparison of revenue opportunities with the
alternate outlet channel.
Fτ ¼ At þ δt ðτ $ tÞ At any time t the system state is describe by {Is,t, Iw,t} where Is,t and Iw,t
are the store and warehouse inventory at time t. Assumption 4 above
proposes a fixed relationship between TC and TO. The LMAT heuristic first
determines the best switching time τ, and then decides whether To ¼ τ or
2. Clearance sales are used only to sellout the store inventory, no delayed to clear out some or all of the store inventory. As noted earlier,
additional shipments are made to the store once a switch is made. The when a switch is made at τ, then the first priority is to sell through the
forecast for clearance sales is assumed to start off with a α factor in- outlet channel. Only if T- τ is sufficiently long will the clearance channel
crease, such that, FTcþ1 ¼ ATc(1þα). Demand then follows a linear be activated.
trend similar to that observed during the regular sales period. The
clearance demand parameters then are:
4.1. The LMAT objective

b A0 $ At
δt ¼ The LMAT heuristic is time iterative and uses a forward looking
t
objective. At the current time t it estimates what would be the revenues, if
a switch was made at a future time t < τ < T. Equation (1) is then
Ft ¼ FTC $1 ð1 þ αÞ $ b
δ t ðt $ TC Þwhere t > TC
rewritten to project the sales in each channels, and therefore described
the expected revenues. Let ϕt,τ be the revenue expectation generated from
time t demand information, if a channel switch is made at τ. Then:

" # # $$%
At þ maxfAt þ δt ðτ $ tÞ; 0g At
Max : ϕt;τ ¼ PR ðΠ $ Iw;t $ Is;t Þ þ PR min ðIw;t þ Is;t Þ; *min ðτ $ tÞ; $
2 δt
0 8 n & ' o9 1
< # # $$ F þ max F þ b δ T $ Iw;τ
$ τ ;0 = " # $ % (2)
Fτ Iw;τ τ τ t βA0
A þ PO min Iw;τ ; T $ τ *βA0
þ PC @min Is;τ ; min $ ; max T $ $ τ; 0 *
: b
δt βA0 2 ; βA0

3. Outlet sales are uniform and constant at a discounted level such that
Ft ¼ βA0, when t > TO. The motivation being that outlet sales are more where,
stable due to attractive price discounting. n o
( At þmaxfAt þδt ðτ$tÞ;0g )
4. The solution strategy is to prefer outlet sales over clearance sales since 2
ðτ $ tÞ; $Aδtt þ R $ Is;t
*min
PO > PC. The clearance period TC to TO is therefore limited only to any Iw;τ ¼ Iw;t $ max Q*!! n o !; 0
!
A þmaxfAt þδt ðτ$tÞ;0g
balance of the selling period after accounting for projected outlet !t 2
*min ðτ $ tÞ; $Aδtt þ R $ Is;t !
sales. (3)

35
J. Zhang et al. International Journal of Production Economics 193 (2017) 31–39

# still positive (Case #2), alternatively demand has dropped to zero

At þ maxfAt þ δt ðτ $ tÞ; 0g (Case #3).
Is;τ ¼ max Is; t þ Iw; t $ Iw;τ $
2 Case #4 – This represents the case where none of the first three
# $ $ (4)
At conditions holds, and indicates a situation where the demand has pro-
*min ðτ $ tÞ; $ ;0
δt gressively become weaker. The supply chain is therefore pressed to make
a switching decision in order to maximize the revenues.
All terms in equations (2)–(4) incorporate the assumptions listed
Case #5, #6 and #7 – In the previous cases only two of the channels
earlier. The first term in equation (2) is the revenue already generated
were active. When condition 1 holds but condition 2 does not hold, then
from regular channel sales, while the second term is the projected regular
the clearance channel will also be activated since the projected ware-
sales in the t to τ period. It considers the possibility of either selling out
house inventory at τ is not sufficient. When condition 4 holds, that is the
the block inventory before τ, or continuing sales through τ. The third
store inventory is large enough for clearance sales to continue through
term in (2) is the projected clearance sales, and considers only the time
the available time (Case #5).
remaining between regular and outlet channels. Given the strategy of
For each of the above cases the conditions allow equation (2) to be
preferring outlet sales to clearance, the remaining sales time allocated to
further simplified, and Table 2 describes the projected revenue as a
the outlet and clearance and channels if a switch is made at τ is:
function of τ.
# $
Iw;τ
Outlet Sales Time ¼ Min ;T $ τ 4.3. LMAT heuristic solution
βA0
# $ At the end of period t, the LMAT heuristic decides whether a switch
Iw;τ from the regular channel to either the clearance or outlet channel will be
Clearance Sales Time ¼ Max T $ τ $ ;0
βA0 made in the next period. The LMAT objective as described in section 4.3,
Note that the allocated time to the clearance channel may not be fully is to optimize the total revenue across all channels. The projected reve-
utilized if the store inventory is not sufficient. The fourth terms projects nue at time t is described by Table 3. These functions are concave and the
the outlets sales. optimal τ * is analytically derived and shown in Table 2. Then if τ # tþ1,
the LMAT heuristic prescribes a switch in the next period, else regular
4.2. Conditional optimization of the LMAT objective channel sales will continue. The heuristic steps are then:

Using a simulation analysis it can be shown that at any time t, 1. Starting from t ¼ 1 (end of period). Record the four state variables:
equation (2) is a concave function in the t # τ # T range. This indicates Iw,t, Is,t, At and δt.
there is a switch time τ* that optimizes ϕt,τ. Since a closed-form solution 2. If δt > 0 then there will no switch in the next period. Wait for tþ1
for τ* is not possible, we use a conditional approach to analytically demand data, and return to step 1.
breakdown equation (2) and derive an optimal solution. The following 3. Set τ ¼ tþ1 and estimate Iw,τ and Is,τ using equations (3) and (4)
five conditions allow equation (2) to be further analyzed and τ* derived. 4. Determine which conditions are satisfied and then use Table 1 to
determine which case is currently applicable to equation (2).
" # $% # $
δt At At 5. Using Table 3 determine τ* for the applicable case.
Condition 1 : At þ min ðT $ tÞ; $ min ðT $ tÞ; $ 6. If τ* # tþ1 then a switch is made in the next period. Else set t ¼ tþ1
2 δt δt
return to step 1 and wait for an update to the state variables.
' IW;t þ IS;t
7. Set TC ¼ tþ1 and TO ¼ Max{tþ1, T- Iw,t/βAO}

bI W;τ Table 3 is derived by taking the derivative of Table 2 revenue equa-

Condition 2 : ' ðT $ τÞ
βA0 tions for each of the listed cases. This decision policy is also summarized
in Table 3. We see that for four of the cases no switch is prescribed for the
At next period, while for one case a switch is definite in the next period. For
Condition 3 : $ ' ðτ $ tÞ two other cases the switch decision is predicated by a switch rule.
δt
# # $$ 5. Evaluation of the LMAT heuristic solution
Fτ Iw;τ
Condition 4 : Is;τ ' min $ ; max T $ $ τ; 0
b
δ βA0
n t & ' o A key analytical question then is how well the LMAT heuristic per-
I
Fτ þ max Fτ þ b
δ t T $ βAw;τ0 $ τ ; 0 forms in controlling the MSC fast fashion supply chain. We used a
( simulation model to compare the LMAT against the true optimal and two
2
other baseline heuristics. A data driven simulation model was built on the
Iw;τ Fτ MS-Excel/VBA platform. Key modelling parameters for the experimental
Condition 5 : ' þT$τ problem are shown in Table 4. The problem is representative of a typical
βA0 b δt
six month fashion retail cycle. The parameters were set such that at a
In combination the five conditions generate seven cases, as shown in
Table 1 and described below. Table 1
Case #1 – The simplest case where demand for the fast fashion Cases and the conditional relationships of the LMAT objective.
product is high and the forecasts indicate the current inventory can be Case Condition holds
sold out in the regular channel within T. If condition 1 holds then this is
1 2 3 4 5
the only likely case.
1 YES
Case #2 and #3 - The case where if a switch occurs at τ then the
2 NO YES YES
projected warehouse inventory is less than the forecasted maximum 3 NO YES NO
outlet sales. This implies there will be no time allocated for clearance 4 NO NO NO
sales since the outlet channel will be active for the entire remaining time. 5 NO NO YES YES
This happens when both conditions 1 and 2 hold. Further, there are two 6 NO NO YES NO YES
7 NO NO YES NO NO
possible scenarios, condition 3 holds implying at τ the regular demand is

36
J. Zhang et al. International Journal of Production Economics 193 (2017) 31–39

Table 2 Table 4
Projected Total Revenue at t for the Conditional Cases. Key parameters for the experimental MCS problem.

Case Total revenue T ¼ 180 Periods Π ¼ 23,000 Units A0 ¼ 200 Units

PR ¼ $100 PC ¼ $20 PO ¼ $35
1 PR *Π
α ¼ 0.4 β ¼ 0.5 N ¼ 20
2 PR *ðΠ$ Iw;t $ Is;t Þþ PR *2At þδ2t ðτ$tÞ*ðτ$ tÞþ PO *ðT$ τÞ*βA0 M ¼ 20
3 PR *ðΠ$ Iw;t $ Is;t Þ$ At
2δt *At *PR þ
PO *ðT$ τÞ*βA0
" %
4 At
PR *ðΠ$ Iw;t $ Is;t Þþ PR ( $ 2δt (At þ PO (Iw;τ

5 PR *ðΠ$ Iw;t $ Is;t Þþ PR *2At þδ2t ðτ$tÞ*ðts $ tÞþ PC *Is;τ þ PO *Iw;τ

& '
6 " % 2F þbδ T$Iw;τ $τ
τ t
Iw;τ
PR *ðΠ$ Iw;t $ Is;t Þþ PR ( 2At þδ2t ðτ$tÞ*ðτ$ tÞþ PC ( T$ βA
βA0

0
$ τ ( 2 þ

PO (Iw;τ
" %
7
PR *ðΠ$ Iw;t $ Is;t Þþ PR ( 2At þδ2t ðτ$tÞ (ðτ$ tÞþ PC ( $ Fτ
( F2τ þ PO (Iw;τ
bδ t

constant demand decline rate, the demand would be exactly zero at

1.22T, implying a selling cycle 22% longer than the planned cycle would
be needed to sell out the block inventory. Then if the fashion product had
average success or a mean demand of 0.5Ao over T, 80% of the starting
inventory would be sold if no other channels are accessed. Similarly if the
fashion product was not successful and mean demand is 0.33Ao, only 50%
of the inventory would be sold. The outlet and clearance prices discounts
are also realistic at 65% and 80%.
Fig. 5. Range of generated demand profiles.
5.1. Real time demand generator
5.2. Simulation results and analysis
Clearly the fast fashion revenue projections are going to be closely
related to the demand behavior. With this in mind we created a real time To benchmark the experimental results, two baseline switching rules
demand generator as an integral part of the simulation analysis. To were also evaluated in addition to the LMAT heuristic: (i) No Switch Rule
evaluate a wide range of product success behaviors we introduce d the – There is no switch and regular sales continue till T or when the demand
demand profile factor to characterize this behavior. drops to zero, whichever comes earlier, and (ii) Beta Switch Rule – If both
Consider a linear decline in demand as the nominal case, than as of the conditions Iw,t ' At(T-t)/2 and At(PR/PO) # βAO hold then a switch
shown in Fig. 5 a variety of demand profiles can be generated around the occurs. The Beta Switch is an intuitively smart logic rule, the first con-
nominal case. A positive d would be indicative of a successful product, dition checks whether it is likely the warehouse inventory can be sold in
with likely less need to use the clearance and outlet channels. In contrast the remain selling cycle. The second condition compares the price dis-
a negative d would be an unsuccessful product, and channel switching counted demand rates in the regular and outlet channels. In addition the
would be likely. Taking the nominal case where assuming demand drops optimal switching decision was determined by tracking the revenue ϕ if a
to zero at 0.9T, by changing d we were able to generate 15 problem sets. switch was made at each of the time periods, and the highest revenue
Within each set the generator uses a random variable to specify the actual switch was assigned as the optimal decision. It is a hindsight solution
demand for the current period. This allows a number of different runs to since it is implementable only after the fact.
be performed with each problem. Table 5 shows the d values for all For each problem M ¼ 30 simulations runs were conducted, the
problem sets. Over multiple product launches a fashion retailer can revenue and switch time were tracked for the optimal decision, LMAT,
expect only a few products will be successful and have a d>0.5. Typically No Switch and Beta Switch rules. This experimental set tests the LMAT
the majority of products would have average success or -0.5 < d < 0.5, robustness across the demand profiles and the randomness within each
while many are expected to be unsuccessful d < 0.5. The generator in- profile. We first examine the switching decision policy as a function of
cludes a random function which specifies the demand for each period t as the demand profile factor. Table 5 give the range of τ decisions and the
a function of d, t and At-1. Each simulation run will therefore generate a average ϕt,τ across the M runs for each problem set. For successful
unique demand sequence, with variance in the short term demand products (d>0.5) the store demand drops more slowly, and we observe
change rate. This variance will affect the four state variables Iw,t, Is,t, At that the switch occurs very late. The regular channel is active for more
and δt and for each simulation the LMAT heuristic run will therefore than 75% of the selling cycle and only a small portion of the block in-
generate a different switching decision. ventory is diverted to the other channels. For unsuccessful products

Table 3
τ* and the LMAT Decision Policy.

Case τ* Switch policy

1 Min{No switch (Tþ1), t when stock out} No Switch

2 βA0 PO
PR $At Switch If: βA0 PO
' At
δt þt PR

3 t$ Aδtt No Switch
4 t$ Aδtt No Switch
5 t$ Aδtt No Switch
#" " % $( & '
6 Iw;t w; tI
δt t$ At Þ(ðPR $ ð1þ αÞ*PC Þþ ½1þ ð1þ αÞ*δt $ b
δ t =2*( T$ (PC ðδt (PR $ 2ð1þ αÞ(δt (PC þ PC *b
δt Þ Rp *At þRδ *PC * T$ βA þt
0
βA0 Switch If: Rp δt $Rδ PC '0
7 t Switch Now

Where : Rp ¼ PR $ PC ð1þ αÞ Rδ ¼ δt ð1þ αÞ$ bδ t a.

37
J. Zhang et al. International Journal of Production Economics 193 (2017) 31–39

Table 5 Table 6
Problem sets and switching behavior. Relative performance of the LMAT and other rules.

Problem d Switching Range – φ optimal Problem LMAT Heuristic No Switch Rule Beta Switch Rule
τ*
Δτ* Δϕ* Δτ* Δϕ* Δτ* Δϕ*
1 $2.5 55 60 $830,884
1 $1 0.1% 92 35.1% 32 7.8%
2 $2.20 68 72 $876,216
2 $1 0.2% 81 24.3% 26 7.5%
3 $1.80 83 87 $958,618
3 $2 0.2% 63 14.4% 24 5.9%
4 $1.50 94 99 $1,018,876
4 $2 0.2% 53 10.6% 23 4.8%
5 $1.25 101 105 $1,075,123
5 $2 0.3% 45 8.3% 21 4.9%
6 $0.65 112 117 $1,203,552
6 $1 0.3% 34 5.2% 21 4.4%
7 $0.35 115 119 $1,249,240
7 $1 0.1% 32 4.9% 20 4.1%
8 0 119 122 $1,320,609
8 0 0.2% 28 4.0% 20 3.6%
9 0.15 124 127 $1,385,278
9 0 0.3% 26 3.5% 19 3.3%
10 0.40 125 128 $1,429,999
10 $1 0.3% 23 2.9% 16 2.8%
11 0.80 127 130 $1,494,657
11 0 0.2% 21 2.5% 15 2.2%
12 1.10 131 134 $1,568,814
12 $1 0.4% 19 2.2% 14 1.9%
13 1.50 134 135 $1,644,428
13 $1 0.5% 16 1.7% 11 1.7%
14 1.90 133 153 $1,692,815
14 $8 0.7% 8 0.9% $3 1.7%
15 2.50 147 163 $1,704,041
15 $13 0.9% $12 0.8% $8 3.4%

(d < $0.5) the switch is much earlier as the retailer activates other
channels to move the block inventory. In particular for market condition section 3.2 restock and holding costs are not considered in the MCS
where d < 1.5 the switching decision is quite aggressive and less than objective. When the holding cost is high and sales are slow, then the
50% of the selling cycle is in the regular channel. At the same time we see retailer will be motivated to clear the inventory quickly resulting in an
that on average the revenue doubles between problem 1 and 15. Clearly earlier switching decision. The MCS model does not explicitly consider
an early switch in problem 15, or a late switch in problem 1 would this case, and an extension could assign time inventory thresholds. Such
adversely affect ϕ. Table 5 also shows the range of switch time within the that if Is,t þ Iw,t is greater than a threshold an aggressive switch to the
30 simulations from each problem. Other than the last two problems, the outlet channel is activated. Another key assumption is that the channel
τ* range is within 5 periods. The results confirm that the problem rep- prices are known and fixed. Clearly, any changes in these prices will
resents a range of demand scenarios, providing a valid set of problems for affect the switching policy. But a reliable analysis would require addi-
testing the LMAT heuristic. tional assumptions about the price demand relationship, that is β ¼ f(PO).
Table 6 compares the performance of the LMAT, No Switch and Beta Given this relationship, multiple MCS models can be solved concurrently
Switch rules against the optimal solution. Δϕ* denotes the average rev- with different PO settings.
enue loss relative to ϕ* for each problem. The No-Switch rule is indica-
tive of the overall utility of channel switching in a FFS. For successful 6. Summary
products the benefits are less than 2.5%, since most of the inventory is
sold in the regular channel and Nw is relatively small. Depending on the Channel switching provides fast fashion retailers with an effective
gross margins for the product, even these small percentages could be strategy to reduce the dependence on multiple discounting steps.
significant. For products with average success the switching benefits are Implementing this strategy requires the retailer to monitor market de-
quite significant and found to be in the 3%–5% range. For unsuccessful mand data in real-time, and make immediate switching decisions. This
products the benefits of channel switching are substantial in the paper formulated the Multi-Channel (regular, clearance and outlet)
5%þ range. Problems 1 to 3 represent product that performed poorly in switch problem, with the objective of maximizing revenue from an initial
the market, and for these switching provides a 14%–35% revenue block inventory. Following a peak demand the demand rate is assumed to
opportunity. be monotonic decreasing. For an unsuccessful product the overall de-
For products with average and or high success the Beta Switch rule mand drops quickly and the need for discounting arises, while for a
matches the No Switch, so is not able to leverage the switching oppor- successful product the demand drops slowly and potentially the entire
tunity. But for unsuccessful products it does perform quite well and inventory can be sold in the regular channel. The objective is simplified
provides a solution within 4%–8% of the optimal solution. The LMAT into cases using a set of conditions, allowing for an analytical solution.
Heuristic performed very well and except for problems 13, 14 and 15, The Linear Moving Average Trend (LMAT) heuristic is proposed, it de-
Δϕ* was less than 0.4%. This confirms that LMAT can readily and cides whether a switch should be made from the regular channel in the
effectively be applied to real time decision making in a FFS situation. We next period.
also see that it performs best with unsuccessful products where Δϕ* was The internet has made pricing history transparent and the managerial
less than 0.2%. The performance strength relative to the Beta Switch was challenge for retailers is how to control pricing speculation. One solution
also greatest as d decreased. The LMAT heuristic was also found to be is to use price differentiated sequential channels, and the LMAT solution
quite robust and performance matched the optimal solution closely allows a retailer to make the switching decision, using real time demand
across the 15 problems. Table 6 shows Δτ* the average difference in data. Many retailers are operating an internet store along with brick-and-
switching times relative to the optimal solution in each run. The Beta mortar stores. Often the internet store is equivalent to an outlet store, and
Switch rule almost always prescribes a switch period late than the with the right pricing differentiation a retailer can use this model to
optimal. The LMAT heuristic though, almost always prescribe τ* to be optimize the revenue across the channels. FFSs are characterized by a
earlier than the optimal. For the majority of problems Δτ* was within a larger number of sequential product offerings, and a retail store can be
few periods of the optimal decision, and for three problems it matched choked by a slow moving product. In particular smaller retailers with a
the optimal solution. single or just a few stores can mitigate the risk by a quick switch to an
outlet channel as shown here. It is difficult for many retailers to match
the ultra-fast supply chain of Zara, an alternative strategy then would be
5.3. Effects of the modelling assumptions to launce multiple products with a fixed initial block inventory and
selling cycle that matches their customer profiles and supply capabilities.
Stores operate in a wide range of retailing formats, and a discussion of The model here shows that this could be quite effective in mitigating
the likely effects of some of the assumptions is relevant. As noted in fashion inventory risks.

38
J. Zhang et al. International Journal of Production Economics 193 (2017) 31–39

Using a series of test problems, representing different levels of Grewal, D., Ailawadi, K.L., Gauri, D., Hall, K., Kopalle, P., Robertson, J.R., 2011.
Innovations in retail pricing and promotions. J. Retail. 87, S43–S52. http://
product success, the LMAT heuristic was compared with the optimal
dx.doi.org/10.1016/j.jretai.2011.04.008.
decisions and the No-Switch and Beta-Switch rules. The No-Switch rule is Huang, Y.-S., Hsu, C.-S., Ho, J.-W., 2014. Dynamic pricing for fashion goods with partial
indicative of the overall utility of channel switching in a FFS. For prod- backlogging. Int. J. Prod. Res. 52 (14), 4299–4314. http://dx.doi.org/10.1080/
ucts that performed poorly in the market, channel switching provides a 00207543.2014.881576.
Iyer, A.V., Bergen, M.E., 1997. Quick response in manufacturer-retailer channels. Manag.
14%–35% revenue opportunity. For products with average and or high Sci. 43 (4), 559–570.
success the Beta Switch rule matches the No Switch, and was unable to Joy, A., Sherry, J.F., Venkatesh, A., Wang, J., Chan, R., 2012. Fast fashion, sustainability,
leverage the switching opportunity. But for unsuccessful products it does and the ethical appeal of luxury brands. Fash. Theory J. Dress, Body Cult. 16 (3),
273–296. http://dx.doi.org/10.2752/175174112x13340749707123.
perform quite well and provides a solution within 4%–8% of the optimal Karakul, M., 2008. Joint pricing and procurement of fashion products in the existence of
solution. The LMAT Heuristic performed very well consistently and for clearance markets. Int. J. Prod. Econ. 114 (2), 487–506. http://dx.doi.org/10.1016/
the majority of test problems provided a solution within 0.4% of the j.ijpe.2007.03.026.
Khouja, M., Park, S., Cai, G., 2010. Channel selection and pricing in the presence of retail-
optimal. This confirms that LMAT can readily and effectively be applied captive consumers. Int. J. Prod. Econ. 125 (1), 84–95. http://dx.doi.org/10.1016/
to real time decision making in a FFS situation. j.ijpe.2010.01.005.
Melis, K., Campo, K., Breugelmans, E., Lamey, L., 2015. The impact of the multi-Channel
retail mix on online store choice: does online experience matter? J. Retail. 91 (2),
References 272–288. http://dx.doi.org/10.1016/j.jretai.2014.12.004.
Şen, A., 2008. The US fashion industry: a supply chain review. Int. J. Prod. Econ. 114 (2),
Cachon, G.P., Swinney, R., 2009. Purchasing, pricing, and quick response in the presence 571–593. http://dx.doi.org/10.1016/j.ijpe.2007.05.022.
of strategic consumers. Manag. Sci. 55 (3), 497–511. http://dx.doi.org/10.1287/ Sierra, J.J., Hyman, M.R., 2011. Outlet mall shoppers' intentions to purchase apparel: a
mnsc.1080.0948. dual-process perspective. J. Retail. Consumer Serv. 18 (4), 341–347. http://
Cachon, G.P., Swinney, R., 2011. The value of fast fashion: quick response, enhanced dx.doi.org/10.1016/j.jretconser.2011.03.002.
design, and strategic consumer behavior. Manag. Sci. 57 (4), 778–795. http:// Smith, S.A., McIntyre, S.H., Achabal, D.D., 1994. A two-stage sales forecasting procedure
dx.doi.org/10.1287/mnsc.1100.1303. using discounted least squares. J. Mark. Res. 31 (1), 44–56. http://dx.doi.org/
Caro, F., Gallien, J., 2010. Inventory management of a fast-fashion retail network. Oper. 10.2307/3151945.
Res. 58 (2), 257–273. http://dx.doi.org/10.1287/opre.1090.0698. Stratton, R., Warburton, R.D.H., 2003. The strategic integration of agile and lean supply.
Caro, F., Gallien, J., 2012. Clearance pricing optimization for a fast-fashion retailer. Oper. Int. J. Prod. Econ. 85 (2), 183–198. http://dx.doi.org/10.1016/S0925-5273(03)
Res. 60 (6), 1404–1422. http://dx.doi.org/10.1287/opre.1120.1102. 00109-9.
Chiang, W.-y.K., Chhajed, Dilp, Hess, James D., 2003. Direct marketing, indirect profits a Verhoef, P.C., Kannan, P.K., Inman, J.J., 2015. From multi-channel retailing to omni-
strategic analysis of dual-channel supply-chain design. Manag. Sci. 49 (1), 1–20. channel retailing. J. Retail. 91 (2), 174–181. http://dx.doi.org/10.1016/
Choi, T.-M., 2007. Pre-season stocking and pricing decisions for fashion retailers with j.jretai.2015.02.005.
multiple information updating. Int. J. Prod. Econ. 106 (1), 146–170. http:// Wang, W., Li, G., Cheng, T.C.E., 2016. Channel selection in a supply chain with a multi-
dx.doi.org/10.1016/j.ijpe.2006.05.009. channel retailer: the role of channel operating costs. Int. J. Prod. Econ. 173, 54–65.
Coughlan, A.T., Soberman, D.A., 2005. Strategic segmentation using outlet malls. Int. J. http://dx.doi.org/10.1016/j.ijpe.2015.12.004.
Res. Mark. 22 (1), 61–86. http://dx.doi.org/10.1016/j.ijresmar.2004.03.003. Zhou, E., Zhang, J., Gou, Q., Liang, L., 2015. A two period pricing model for new fashion
Ding, Q., Dong, C., Pan, Z., 2016. A hierarchical pricing decision process on a dual- style launching strategy. Int. J. Prod. Econ. 160, 144–156. http://dx.doi.org/
channel problem with one manufacturer and one retailer. Int. J. Prod. Econ. 175, 10.1016/j.ijpe.2014.10.008.
197–212. http://dx.doi.org/10.1016/j.ijpe.2016.02.014. Zhang, J., Gou, Q., Zhang, J., Liang, L., 2014. Supply chain pricing decisions with price
Ghemawat, P., Nueno, J.L., 2003. Zara Fast Fashion Case Study. Harvard Business School reduction during the selling season. Int. J. Prod. Res. 52 (1), 165–187.
Multimedia Case, 9-703-16.

39

View publication stats