Reduction of walking time

in the distribution center of De Bijenkorf

René de Koster 1, Kees Jan Roodbergen 1 and Ronald van Voorden 2
1
Erasmus University Rotterdam, Rotterdam School of Management, P.O. Box 1738,
3000 DR Rotterdam, The Netherlands
2
Districon B.V., Raadhuisstraat 32-34, 3603 AW Maarssen, The Netherlands

Published in: New trends in distribution logistics, M.G. Speranza and P. Stähly
(eds.), Springer, Berlin, 215-234.

Abstract. In many distribution centers, there is a constant pressure to reduce the

order throughput times. One such distribution center is the DC of De Bijenkorf, a
retail organization in The Netherlands with 7 subsidiaries and a product assortment
of about 300,000 SKUs (stock keeping units). The orders for the subsidiaries are
picked manually in this warehouse, which is very labor intensive. Furthermore
many shipments have to be finished at about the same time, which leads to peak
loads in the picking process. The picking process is therefore a costly operation.
In this study we have investigated the possibilities to pick the orders more
efficiently, without altering the storage or material handling equipment used or the
storage strategies. It appeared to be possible to obtain a reduction between 17 and
34% in walking time, by simply routing the pickers more efficiently. The amount
of walking time reduction depends on the routing algorithm used. The largest
saving is obtained by using an optimal routing algorithm that has been developed
for De Bijenkorf. The main reason for this substantial reduction in walking time, is
the change from one-sided picking to two-sided picking in the narrow aisles.
It is even possible to obtain a further reduction in walking time by clustering the
orders. Small orders can be combined on one pick cart and can be picked in a
single route. The combined picking of several orders (constrained by the size of
the orders and the cart capacity) leads to a total reduction of about 60% in walking
time, using a simple order clustering strategy in combination with a newly
developed routing strategy. The reduction in total order picking time and hence the
reduction in the number of pickers is about 19%.

Keywords. warehouse, order picking, routing, batching, case study.

1. Introduction

In many warehouses and distribution centers, short order throughput times are of
crucial importance. There are several causes for this.
• Suppliers of manufacturing companies are being forced to supply in a just-in-
time manner. Their customers have lowered their inventories and demand a
rapid and timely supply from their vendors.
• Many internationally operating companies have centralized their European
distribution in so-called EDCs, European distribution centers. These EDCs are
responsible for the warehousing function and distribution to multiple European
countries. The internal process organization often leads to wave picking, in
which the pick process is carried out in batch, governed by fixed truck departure
times for the different countries or regions. In order for these trucks to leave in
time, the orders must be ready before departure, regardless of their number.
However, in practice it can often be observed that there is a peak of departures
in the afternoon and of receipts in the morning. This leads to order picking
processes that are capable of handling peak loads in a timely manner.
• In contrast to the above way of working in EDCs, customer sales does not want
any concession in the customer service and simultaneously wants to guarantee
short delivery times (overnight for customers in a radius of 500 km of the
EDC). This leads to order cut off times that are as late as possible before the
truck departure times. Often special procedures are created to be able to handle
late emergency orders in time.
• Short delivery times are considered in many branches as a competitive weapon.
This puts pressure on the internal throughput times, especially order picking
throughput times.
• It becomes more and more difficult to realize short order throughput times
because of factors such as a gradual increase in assortment and smaller, yet
more frequent, orders. For the increasing assortment an increasing amount of
floor space is necessary. This in turn results in increased walking times per
order. Smaller orders (less items per line) and an increased frequency of
ordering lead to an increase in the work contents of order picking: less full
pallets can be picked and more single item picks are necessary.
• The increase of value added logistics (VAL) activities in many warehouses has
lead to additional activities that have to be carried out during or after the order
picking. These additional activities often lead to the necessity of picking and
handling such orders separately, within the short time frame available for
handling the orders.
• Especially in the retail business, the increased application of ECR (efficient
consumer response) concepts has lead to the direct transmission of order
information from scanning cash registers to the distribution centers. These
orders are then translated in replenishment instruction from the DC to the stores.
This often means more, but smaller orders that have to be supplied.

Most warehouses are gradually confronted with the above mentioned

developments. It is important to find an adequate solution to maintain short and
well-controlled throughput times. One such option is a radical new design
consisting of a new layout, further mechanization and automation of processes.
However, often also by less radical methods the efficiency of the order picking
process can be increased.
In this paper, a number of methods are discussed that can help improving the
efficiency of the order picking process, without layout change, or a change of
storage policies or of material handling equipment. Order picking is, in most
companies, a manual job. By a better organization of the process, a more efficient
order picking, it is often possible to obtain a substantial reduction of the order
throughput time. According to Tompkins et al. (1996) the operation costs in
warehouses are determined to a large extent by the order picking process
(approximately 55%).
The efficiency of the order picking process depends on factors that are difficult
to change, such as the chosen storage systems (racks), the layout, the order picking
system (order picking trucks, pick carts, pick-to-belt, pick-to-light, etc.), but also
by parts that are more easily changeable, such as the storage strategy (the storage
location determination), the sequence by which items are collected from storage
locations (routing strategy) and the possible clustering of customer orders in a
single order picking route (batching).
In section 2, these relatively simple strategies are considered in more detail. In
section 3 it is demonstrated by the case of the distribution center of ‘De Bijenkorf’
in Woerden, The Netherlands, that proper choices for these parts can lead to strong
improvements in the efficiency of the order picking process. In this case study the
focus is on routing and batching strategies. In section 4 some conclusions are
drawn.

2. Reduction of the order pick time in distribution centers

Order picking and shipping customer orders within an agreed time is the core
function of a distribution center. In the picking process, customer orders are
converted into pick orders.
The time, needed for picking an order can be split in travel time (walking or
driving time), pick time and remaining time. The travel time is related to the
movement between locations that have to be visited (where the items are stored
that have to be picked). The pick time is related to grabbing the items. This
includes search for the article, grabbing the units, depositing them on the pick cart,
checking the pick, administration of the pick on the pick list (if applicable) and
reading the next location to be visited. The remaining time is related to activities
such as the acquisition of the pick order, dropping off the full pick cart, waiting
time for a next assignment, social contacts etc.
According to Tompkins et al. (1996), the travel time in a warehouse is, in
general, responsible for half the total order picking time (see table 1). Hence,
reduction of the travel distances and therefore of the travel times has a significant
impact on the total order picking time. In a stepwise approach for gradual
improvement of the order picking process, walking distance reduction is an
important candidate.

Table 1. Typical distribution of an order picker’s time (according to Tompkins et al.

(1996)).

Other

Setup

Pick % of orderpicker’s time

Search

Travel

0 10 20 30 40 50

There are several ways to reduce the travel distances in the order picking
process in a warehouse. It is possible to reduce walking distances by system
changes, using a higher degree of automation. Examples are the use of A-frames
(automatic product dispensers) for automated picking of small articles, AS/RS
warehouses, miniloads and carousel systems, in which the items that have to be
picked are automatically moved to the picker (‘goods-to-man’), or the use of pick-
to-belt order picking combined with automatic sortation.
On the other hand it is also possible to reduce walking times by less radical
changes. Instead of changing equipment, we can try to improve the operating
policies. In sections 2.1 - 2.3, we will restrict ourselves to conventional solutions
that are applicable in many warehouses. In specific we will discuss:
• Compact storage. The idea with compact storage is to make the area that has to
be traversed by the order pickers as small as possible.
• Routing. Given an order to be retrieved from storage, in which sequence should
the order picker visit the pick locations?
• Order batching. Combining two or more orders into one pick route can decrease
the total distance to be traveled by the order picker.

2.1 Compact storage

This can be reached by applying several principles, such as narrowing pick aisles
and densifying storage locations, separation of bulk and pick stock, dividing the
pick stock over pickfrequency based storage zones (ABC-storage).
2.1.1 Reduction of storage space needed

Aisles in warehouses may have various widths. This depends on the items and
product carriers that have to be stored and on the material handling systems used to
store and pick them. Since warehouse aisles occupy a considerable percentage of
the total storage floor area (costs) and since the average distance to be traveled for
order picking or storage increases with the aisle width, it is important to make the
aisles as narrow as feasible. In pallet storage areas this may lead to narrow-aisle
(semi) high-bay storage with aisle widths varying between 1.2 and 1.7 m. This is a
significant reduction when compared to traditional wide-aisle pallet storage with
aisles between 2.5 m (reachtrucks) and 3.5 m (forklift trucks). Of course, such a
choice has impact on the material handling systems that can be used and on the
order picking and storage policies that can be used.
A second way of reducing the storage area is a more efficient use of storage
locations. In many warehouses, the average storage location filling rate is low,
about 20%. By distinguishing between slot sizes, compacting stored loads as much
as possible (for example, storing multiple SKUs per location, or combining
multiple small loads of one SKU on one location), it is often possible to achieve
great savings in storage space. This can result in a smaller storage area and hence
in reduced throughput time and lower costs.
Another option, which may be the most important one, has to do with better
cooperation and coordination with suppliers. Storage space reduction can be
achieved in various manners. One can think of:
• ordering more frequent smaller quantities, which reduces inventory,
• using drop shipments whenever possible, which avoids handling and storing
items first,
• focusing on cross-docking rather than first bringing material to storage,
• reducing standard package quantities, which avoids separate locations for
broken packages.

2.1.2 Separation of bulk and pick stock

The purpose of separating pick stock in a forward area (from which orders are
picked) and bulk stock in a reserve area (used for the replenishment of the pick
stock) is to reduce the space in the order picking zone, i.e. the forward area. This
can significantly reduce the travel time needed in the order picking process.
However, the gain in travel time has to be balanced against the additional time
needed to replenish the pick stock from the bulk stock. The main question to be
answered is: which quantity per article has to be stored in the forward area. Some
methods used in practice to solve this problem are:
• allocate the same space to every article in the forward area (‘equal space
method’);
• allocate to every article an amount of space sufficient to meet the demand over
a predefined period (‘equal time method’).
 Other procedures to allocate products to the forward area, with their
performance bounds, can be found in e.g. Hackman and Rosenblatt (1990) and
Hackman and Platzman (1990).
In most warehouses all articles have a pick position in the forward area. From
recent research by Van den Berg (1996) it appears that considerable savings are
possible by picking some articles directly from the reserve area. It has to be noted,
however, that order picking from the bulk area needs material handling systems
capable of both order picking and storage and retrieval of unit loads.

2.1.3 Pickfrequency class-based storage of articles in a warehouse

In general, travel times can be significantly reduced if zoning is used to store

the articles. Zoning can be applied in different ways. First, it is possible to create
separate storage areas and even separate storage systems, handling systems and
separate control for SKUs with different turnovers. For example, in the retail
business it is a custom to create a fixed zone for discount articles that move
extremely fast. The discount actions are often planned in advance and the storage
area is continuously adapted according to the sale action of the coming period.
A second way to apply zoning is subdividing a single physical storage system,
for example a pallet warehouse, in multiple logic zones, that contain SKUs
belonging to different pickfrequency classes. Fast moving items are stored close to
the front end of the racks at easily accessible locations and slow moving items are
stored further away. According to Hausman et al. (1976) pickfrequency class-
based storage can lead to a reduction of up to 60%, using only three pickfrequency
classes. The savings that can be reached, depend on the number of SKUs
responsible for a significant part of the picks to be carried out. Van den Berg
(1996) suggests that in some situations, five or six pickfrequency classes yield a
significant further reduction in travel time. More classes have virtually no
additional effect.
Although the above research was carried out in an automatic warehouse
environment with aisle-bound trucks or cranes, several authors have reported
similar results in other pick-storage layouts. See for example, Caron et al. (1998).

2.2 Routing

When the items that have to be picked in a single tour by the picker and their
corresponding storage locations have been established, the sequence has to be
determined by which these locations have to be visited. This sequence is
determined by a routing strategy. The chosen routing strategy has a direct impact
on the length and travel time of the tour. Good routing strategies can significantly
reduce travel time. De Koster and Van der Poort (1998) report reductions of 30%.
Ratliff and Rosenthal (1983) developed an algorithm to find shortest order
picking routes in warehouses with multiple parallel aisles, a central depot and
without cross aisles. Their method is based on dynamic programming, where they
start in the left most aisle and consider all possibilities to visit the next aisle. This
procedure is repeated until all aisles are added. Five different methods (transitions)
are distinguished to add an extra aisle and six transitions are distinguished to visit
the locations within the aisle. In routes resulting from their method it is possible
that an aisle is skipped first, but that it has to be finished later by returning to the
aisle. The algorithm of Ratliff and Rosenthal has been extended by De Koster and
Van der Poort (1998) for the situation that the picked items can be dropped off at
the head of every aisle (decentralized depositing).
In practice, usually simple heuristics are used. To the knowledge of the authors,
at the current moment no standard Warehouse Management System (WMS)
package exists that contains the Ratliff and Rosenthal algorithm. Methods that are
used include the ‘S-shape’ strategy, the ‘Midpoint’ return strategy and the ‘Largest
gap’ return strategy (see Hall (1993)). In the S-shape strategy, the aisles containing
items are fully traversed in a single direction. This is done in an S-shaped curve
fashion. In the application of this algorithm, a distinction has to be made between
single-sided picking and two-sided picking. See figure 1. Single-sided picking is
only to be preferred in case the aisles are either very wide or when many items
have to be visited within the aisles.

Figure 1. Warehouse (top view) with single-sided picking (left) and two-sided picking
(right) using the S-shape algorithm.
The midpoint return strategy and largest gap return strategy are both routing
strategies where the order picker travels in the aisle but always returns to the aisle
head where he started. The point of return in the aisle is either determined by the
aisle midpoint or by the largest gap between two sequential item locations. The
largest gap between such item locations is not traveled. The largest gap strategy is
always beneficial over the midpoint return strategy. See for example, Hall (1993).
Hall shows that if the number of items to be picked per aisle is small (less than
about 3) then the Largest gap strategy is beneficial over the S-shape strategy, with
respect to travel time.
 For the case that the warehouse has a middle aisle (cross aisle) as in figure 2, an
optimal routing algorithm has been developed by Roodbergen and De Koster
(1998). A middle aisle divides the warehouse in a front and a rear block. They also
developed a Combined heuristic for routing order pickers in a warehouse with any
number of cross aisles. The heuristic uses a dynamic programming technique to
combine S-shape and an aisle return strategy for traveling within aisles. In
traveling to a next aisle, significantly fewer possibilities are allowed than in the
optimal algorithm: only two (front or rear end). In the traveling within aisles the
best is selected of the S-shape strategy and the ‘enter-aisle-and-return’ heuristic. A
difference with the optimal algorithm is further that picking is done per warehouse
block and that no returning to previously skipped aisles can occur.
Examples of the different routing strategies can be found in figure 2.

Figure 2. Examples of four different routing strategies, for a warehouse with a middle aisle.
2.3 Order batching

Batch picking is defined in this paper as a method by which multiple orders are
picked in one pickroute. The items that are combined in the pickroute in this way
have to be sorted per order. Sorting per order can happen during picking (‘sort-
while-pick’) or afterwards (‘pick-and-sort’), the latter often via a sortation
machine.
The difficulty of batch picking is how to determine which orders can be
combined best in order to minimize the total travel time. A boundary condition is
that the capacity of the pick device (cart, pallet, roll cage) may not be exceeded.
The problem of finding an optimum is NP-hard (see Gademann et al. (1996)).
Gademann et al. (1996) describe a branch-and-bound algorithm for a related
problem where the objective is to minimize the maximum pick time of any of the
batches formed. Their algorithm is complex and has unpredictable (and often long)
calculation times.
However, simple heuristics can be used with very good results, as shown by De
Koster et al. (1999). In their paper, the batching heuristics were divided into three
groups:
• Simple, straightforward algorithms (like First Come, First Served)
• Seed algorithms
• Algorithms based on time savings.
Each of these groups will be described below in more detail. Other methods
could possibly be developed by adjusting solution methods for the Vehicle Routing
Problem (see e.g. Fisher (1995)) for the case of order batching.

2.3.1 ‘First Come First Served’

In the ’First Come First Served’ batching algorithm, the orders are assigned to
pickroutes in sequence of arrival, until the capacity of the pickdevice is reached.
Each time that this happens, a new route is started. Only complete orders are
assigned. This is a method widely used in practice, because of its simplicity.

2.3.2 Seed-algorithms

Seed algorithms consist of two different steps, which are repeated for each new
route. First, via a seed selection rule an order is selected, that has not yet been
included in a route. Next, with an order addition rule, orders are added one by one,
until the capacity of the pickdevice is reached or no order can be added. the
addition may, for instance, be based on the distance (measured with some distance
metric) between the existing orders in the route and the order that may be added.
The advantage of seed rules is, that they are in general simple to use and fast in
calculation.
2.3.3 Time savings algorithms

Time savings algorithms are based on the time savings that can be obtained if two
orders are combined instead of executed separately (inspired by the time savings
algorithm for vehicle routing by Clarke and Wright (1964)).
For each pair of orders the time saving is calculated. The order pair that yields
the maximum time saving and that fits on the pickdevice is combined in a route.
The order pair with the next greatest time saving is then taken. If one of the orders
has already been assigned to a route, the other order is added to that route, if
possible. If none of the orders is contained in a route, a new route is formed
containing the two orders. This process is repeated for the remaining pairs until all
orders have been assigned to a route.
In the calculation of the time savings, a routing algorithm may be used. This
may be one of above mentioned heuristics (S-shape, Largest gap, etc.) or an
optimal algorithm.
Several types of the above mentioned algorithms are discussed in De Koster et
al. (1999). In the next section the impact of efficient routing and batching on travel
time are discussed for the case of De Bijenkorf.

3. Case: The distribution center of ‘De Bijenkorf’

3.1 De Bijenkorf B.V.

De Bijenkorf B.V. is a chain of department stores in the Netherlands with 7 stores

in the major cities and over 3600 employees. The assortment is very broad, about
300,000 stock keeping units (SKU) and changes constantly. The assortment
contains fashion, consumer electronics, household appliances, books, furniture,
personal care products, etc.
All stores are supplied from one DC in Woerden. The arriving articles are
checked on quantity, quality and, in many cases, are labeled with the sales price.
Next, a part of the arrivals are directly cross-docked to the stores. Another part is
stored in one of the four storage areas. One of these storage areas is for hanging
fashion. The other three storage areas have conventional storage equipment, like a
bin storage area, a pallet area and an open shelf storage area.
In the bin storage area, the products are stored in plastic bins. They are also
picked in the same type of bins. In this area the picking is the most labor intensive.
3.2 Order picking in the bin storage area

3.2.1 The bin storage system

The plastic bins are stored on two different floors in shelf racks. Both floors
consist of 12 blocks each having 11 aisles of 18.6 m. length and an aisle width of
90 cm. Every rack has 42 sections offering space for 8 plastic bins, stored on top of
each other. Two adjacent blocks form a so-called preferred zone. Every article
stored in the bin storage area is assigned to one preferred zone, depending on the
cash register from which it is sold in the department stores. Therefore, products
sold at a singe cash register are grouped together in one preferred storage zone, a
form of family grouping of a similar kind as can be observed in food retail
warehouses. The most important activities in the bin storage are the storage of
plastic bins with incoming goods and the picking of single units from the plastic
bins.

3.2.2 Storage

Within the bin storage area, a semi-random storage strategy is applied. This means
that if a (group of) bin(s) containing a single SKU has to be stored, the information
system (named VIRGO) indicates the preferred storage zone where the bins have
to be stored. Within this zone, the warehouse employee may store the bin at any
free location. The employee has to keep the bins together (in a vertical stack) as
much as possible. Every employee tries to minimize the walking distance when
storing a group of bins and stores the bins at the free location closest to the
inbound conveyor. After storage, the storage location is confirmed to VIRGO.

3.2.3 Order picking

A pickorder consists of those SKUs that have been ordered by the department
corresponding to one cash register in a store and is picked by one order picker in
one preferred zone (although exceptions occur). Each order picker usually picks
one order at a time. The picked items are put in plastic bins on a small pick cart.
The sequence in which the orderlines (SKUs) have to be picked is indicated on the
pick list. This sequence is based on a simple sortation of the picklocations in
increasing order. When the pick order has been finished, the pickorder is
confirmed to VIRGO at a central terminal. The pick bins are placed on a conveyor
and transported to a sortation system where they are sorted per store and prepared
for further transportation.
3.3 Problem description

In 1992, a large business process reengineering project ‘Het Distributieproject’

was started at De Bijenkorf in order to make distribution processes more efficient.
This process resulted in a number of important changes. First, a new logistic
information system, VIRGO, was introduced for support of purchasing,
distribution and sales. This system also supports scanning cash registers, where
point-of-sales information is matched with the stock in the store and immediately
translated into replenishment orders at the warehouse, if necessary (an ECR
application). Second, the sales area in the stores was enlarged, so that a larger
number of SKUs is now displayed at the stores, rather than stored in the
warehouse, in order to increase sales. The stock per SKU has decreased at the
stores.
This has put more pressure on the warehouse. All stores are now supplied daily,
instead of once every two weeks. In the bin storage area, this has lead to many
small orders, instead of few and very large orders. The total walking distance
needed to collect the orders has increased significantly. In a study of Van Voorden
(1997) it appeared that an order picker in the bin storage area walks on average 7
km on a daily basis to collect the items.
The research project at De Bijenkorf focused on a reduction of these walking
distances. In the previous part of this paper several methods of travel time
reduction have been discussed, but here we focus only on two, relatively simply
implementable methods, namely sequencing the pick locations on the pick lists
(routing) and combining multiple orders per pickroute within a preferred zone.

3.4 Problem solution approach

The approach that was chosen consists of the following steps:

1. Order analysis. A fairly large number of orders in two representative preferred
zones of the bin storage area were each analyzed for:
- articles contained in it,
- storage location of each article,
- route taken by the order picker to pick the order (this actually required
walking the order together with the picker),
- time distribution of the picker: time for picking, walking and
administration,
- number of pick bins needed to pick the order (the number of pick bins
necessary to pick the order always fits on the pick cart),
- registration of incidents, such as: location empty registration, search for a
new empty pick bin, article not available at indicated location, etc.
2. Simulation of current situation. The current order structure and routing method
was implemented in a simulation program.
3. Simulation of alternative routing strategies. A number of routing strategies were
implemented in a simulation program and compared for a large number of
orders with the current method.
4. Simulation of batching strategies. A number of different batching strategies
were defined and investigated on practical feasibility.
5. Comparison of results. The batching strategies of the previous step were
implemented in a simulation program and compared for a large number of
orders with the current method.
6. Conclusion. Results were obtained and the best alternative was chosen.
7. Implementation. The best solution is currently being implemented.

The above steps are worked out in the sequel.

3.5 Order analysis

During two consecutive weeks all orders in the two preferred zones were recorded.
Some data of these orders is listed in table 2. Not only averages were obtained, but
also full frequency distributions of all quantities listed in table 2. In this table, units
per order indicates the number of product units (a single product or a set of
products in one wrapping) in an order. Lines per order stands for the number of
product types in an order (multiple units of one product account for one order line).

Table 2. Some data collected for two preferred storage zones.

Preferred zone 1 Preferred zone 2
Ave. orders per day 47 29
Ave. lines per order 17.0 31.6
Ave. units per order 39.3 131.2
Ave. pick bins per order 3.6 3.4
Regression: bins/order = 0.994 + 0.156 lines/order 1.026 + 0.076 lines/order

From the collected data it was possible to find a pickfrequency distribution for
the different storage sections within the preferred storage zones. This was used
later on in a simulation program to generate random storage locations of pick
items. The storage location distribution is quite skewed, due to the fact that within
a preferred storage zone, a worker may freely find a location sufficiently large for
the bins to be stored.
The number of pick bins necessary for picking the order depends largely on the
size of the order, as can be seen from figure 3. The maximum number of pick bins
in one route is 12, in the current situation.
The regression analysis results are also listed in table 2. Due to the upward
rounding of the number of bins needed (even 0.1 bin means that 1 bin is needed) in
table 2, adding more variables in the regression equation (like the number of
units/order) does not yield a significantly better prediction of the number of pick
bins necessary. The regression results were used for the simulation.

Figure 3. Regression line for the average number of pick bins per group of orderlines. The
left figure is for preferred zone 1, the right one for preferred zone 2.

3.6 Analysis of the current pick routes

From the analysis of the pick routes it appeared that the large distances that are
traveled in the picking process can be attributed mainly to the single-sided
orderpicking. It is a well-known fact that two-sided orderpicking greatly
outperforms single-sided orderpicking unless the number of pick locations per aisle
is very large or the aisles are very wide, see Goetschalckx and Ratliff (1988). The
single-sided picking is again due to the location numbering. The location numbers
increase as indicated in figure 4. The pick locations are indicated on the pick list,
sorted on increasing location section numbers. The layout in figure 4 (and figure 5)
has been simplified slightly; the number of locations per rack is only 4 (42 in
reality).
In fact, the pickers use two main different traveling strategies. Some pickers
work from the middle aisle: they leave their pick cart in the middle aisle when
entering an aisle. Other pickers always take their pick cart with them and then
consequently travel the full aisle when they have to switch to a neighboring aisle.
However, all pickers pick strictly in the sequence indicated on the pick list. After
simulation (see next section), it appeared that the last method is slightly better on
average than the first one. For the comparison, only this best one was used.
In figure 5 an example is given of this last type of pick routes, in which 17
locations have to be visited. In the righthand storage block of the preferred zone,
the picker enters the aisle from the middle aisle and returns to the middle aisle
after the furthest pick in the aisle. In the lefthand storage block, the picker
traverses the full aisle, starting from the middle aisle, if the next pick location is
also in the lefthand storage block and the traveling direction matches with the
location numbering sequence (see figure 5).
 Two aisles have been marked with an arrow in figure 5, to show that in the top
aisle the picker returns to the middle aisle to make sure that in the next pick aisle
the travel direction matches the location numbering sequence. On the other hand,
in the lower aisle the picker travels the aisle completely also to achieve that the
traveling direction matches the location sortation.

Figure 4. Current location numbering.

Figure 5. Orderpicking route in the current method with full aisle traveling (lefthand block)
and return (righthand block).
3.7 Order simulation and results

For each of the two preferred storage zones, 10,000 orders have been randomly
generated. The travel times needed for these orders were calculated with different
routing and batching strategies, including the current one (see figure 5). The
number of orderlines of an order, the pick location of each of the articles and the
number of pick bins needed for the order, are all drawn from the corresponding
probability distributions based on the order analysis.

3.7.1 Routing results

In total, 3 different heuristic routing methods and an optimal algorithm have been
compared with the current routing method. The heuristic routing methods are S-
shape, Largest gap and the Combined heuristic (see Roodbergen and De Koster
(1998)). The optimal algorithm (see Roodbergen and De Koster (1998)) is based
on dynamic programming and similar to the algorithm of Ratliff and Rosenthal
(1983). The results can be found in table 3.

Table 3. Comparison of three heuristics and the optimal routing method with the current
routing method.

Method Average daily Difference with Total average Difference with

travel distance current method daily picktime current method
in bin storage (151.352 m.) (153 hours,
area 31 min.)
S-shape 125,738 m 16.9 % 145 hrs, 5,2 %
(25,614 m) 37 min. (7 hours, 54 min.)
Largest gap 116,979 m 22.7 % 142 hours, 6,9 %
(34.373 m) 55 min. (10 hours, 36 min.)
Combined 105,091 m 30.6 % 139 hours, 9,3 %
heuristic (46,261 m) 15 min. (14 hours, 16 min.)
Optimal 99,349 m 34.4 % 137 hours, 10,5 %
(52,003 m) 28 min. (16 hours, 3 min.)

The results in table 3 have been obtained by extrapolation of the simulation

results of 2 preferred storage zones to all 12 preferred storage zones in the bin
storage area. It appears that even a simple heuristic such as S-shape or Largest gap
yields a significant reduction in traveling distance, namely 16.9% and 22.7%,
respectively. This magnitude of the reduction is mainly due to the change to two-
sided picking. If the smarter Combined heuristic or an optimal algorithm is used,
improvements of even 30.6% or 34.4%, respectively are obtainable.
Even though the reduction in walking distance is significant, the improvement
of the total pick time (which includes, besides traveling time, also picking time and
administration time) is far less. This is due to the fact that a large part of the non-
travel time is spent on removing bins from the racks, waiting for a non-occupied
computer terminal to confirm the picks and other administrative tasks. It is clear
that further improvements are possible here. The result is that the saving in total
pick time varies from 5.2% for S-shape up to 10.5% for the optimal algorithm.
Assuming that a productive manday is on average 7 manhours, this leads to a
reduction in personnel varying between 1.1 and 2.3 ftes.
It is clear that the optimal algorithm has an advantage over the 3 other
heuristics. In practice, there are however also some disadvantages. One such
disadvantage is that the algorithm is more complex than the other ones. It has to be
implemented in the core part of the Warehouse Management System, which is not
an easy task. Also, the optimal algorithm is not very easily adaptable. For example,
if the layout of a preferred storage zone would be changed from 2 to 3 adjacent
blocks, the algorithm is not usable anymore and not easily adaptable either.
Another disadvantage of the optimal algorithm is, that the sequence of the
locations on the picklist is not always straightforward to a picker: it does not work
block for block and also backtracking to previously skipped aisles is possible.
Also, the picker has freedom in deciding via which aisle head he moves to a
neighboring aisle. This could lead to longer walking times than expected. The
heuristics do not have these disadvantages, or to a less extent.
From the simulation results it appears that the Combined heuristic has very
good performance, but that the routing is much less complicated, than that of the
optimal algorithm. Therefore, the management of De Bijenkorf decided to
implement the Combined heuristic for routing the order pickers.

3.7.2 Batching results

In total, 4 different batching strategies have been compared with the current
method, in which no batching takes place. Since the batched orders are to be sorted
by order during the picking process (‘sort-while-pick’), it was also necessary to
investigate how the pick bins could be stacked on the pick cart. The results for
three different stack variants have been indicated in table 4.
All stack variants depend on the design of the pick cart. Two pick carts have
been sketched in figure 6. Pick cart type A has multiple levels. The pick bins are
individually accessible by sliding them off the cart, which offers the possibility of
picking a large number of orders at the same time. Pick cart B has only one level
on which two bins can be placed next to each other. Other bins can be stacked on
top of these two bins.
Stack variants I and II both use a pick cart of type A. The difference is, that the
pick cart for variant I has 6 layers with each 2 bins, which makes all bins
individually accessible. In variant II only 8 bins are individually accessible from 4
layers. The remaining 6 bins are stacked on top of each other on the top level (5th
layer). For stack variant III, pick cart B is used. With this pick cart only 2 bins are
directly accessible, since otherwise bins may have to be removed before a
particular bin is accessible. Therefore, in stack variant III only two orders can be
collected simultaneously. The picker starts with two empty bins, one for each of
the two orders. If a bin is full, another empty bin for the same order is stacked on
top of it. Empty bins are available everywhere in the bin storage area.
The advantage of batching variants using a pick cart of type A, is that a variable
number of orders can be collected, as long as the pick bins needed for these orders
fit on the cart. The disadvantage compared to pick cart B, where bins are stacked
on top of each other is, that less bins can be picked in one route (assuming a
maximum ergonomic stacking height).

Figure 6. Example of two different pick carts, type A (left) and type B (right).
It should be noted that the extra activities in sorting out the picked items in the
proper bin (for all variants) and retrieval of the proper bin from the cart (variants I
and II) cost extra pick time compared to the current situation. The extra time per
pick location is estimated at 4, 5 and 2 sec. for the three variants, respectively.
From the simulation experiments, it appeared that good results can be obtained
with a time-savings-based batching strategy. This strategy can be implemented
relatively easy. Also seed algorithms for batching have been implemented, but
these did not yield better performance (not even with an extended local 2-OPT
optimization procedure). Only the Combined routing heuristic was applied since
this was the routing method preferred by the management of the Bijenkorf. The
results can be found in table 4.
Although the difference in total average daily walking distance between the
three stacking variants is substantial (distance reductions varying between 57 %
and 68%), due to the difference in pick time, the difference in the total order pick
time is moderate. In practice, variant III, with sortation of only two orders, is the
simplest to the order pickers, with the least possibility (of the three) to make
mistakes: sorting the picked items for the wrong order. The savings in total pick
time mount up to about 19%, which is about 4.2 pickers, based on an effective 7
hours per picker per day.
 After the Combined routing algorithm has been implemented, De Bijenkorf will
implement the time savings batching heuristic (using the Combined routing
algorithm) with stack variant III.

Table 4. Average walking distance and total order pick time with order batching and three
different stacking methods in the bin storage area.

Average daily Difference with Total average Difference with

Stack variant travel distance current daily picktime current method
in bin storage method (153 hours,
area (151.352 m.) 31 min.)
Variant I (12 bins) 54,196 m 64.2 % 123 hours, 19.2 %
(97,156 m) 58 min.
Variant II (14 bins, 48,571 m 67.9 % 124 hours, 18.9 %
max. 8 orders) (102,781 m) 31 min.
Variant III (14 bins, 64,825 m 57.2 % 124 hours, 18.9 %
max. 2 orders) (86,527 m) 32 min.

4. Conclusions

In this study, it has been shown that substantial savings in a warehouse can be
achieved by making the order pick process more efficient. In the bin storage area
of De Bijenkorf the travel distances could be reduced by 30% and the number of
pickers by 1.2, by application of a relatively simple routing heuristic. This
improvement is to a large extent due to the fact that order pickers currently follow
the location numbering, which results in single-sided picking. The introduction of a
simple heuristic and corresponding double-sided picking gives a significant
improvement. An optimal algorithm for routing was not considered necessary by
the management of De Bijenkorf, since travel time improvement was only 3,8%
higher than that of the best heuristic. Furthermore, confusion for the order pickers
might increase when introducing an optimal routing method. This is due less
intuitive routing.
If orders are batched as well, with a time-savings method and the combined
routing heuristic, even stronger savings can be achieved, about 68% reduction of
travel distance and a saving of 3 to 4 pickers. Better results may even be possible
by developing new batching methods based on their analog to the vehicle routing
problem.
Besides the above savings, more is probably achievable, by properly looking at
reduction of administration time per order. For example, by using barcodes and
scanners in the pick process. This would also eliminate the order confirmation
process.
References

Van den Berg, J.P. (1996): Planning and control of warehousing systems. PhD thesis,
Twente University of Technology
Caron, F. / Marchet G. / Perego A. (1998): Routing policies and COI-based storage
policies in picker-to-parts systems. International Journal of Production Research, 36(3),
713-732
Clarke G. / Wright W. (1964): Scheduling of vehicles from a central depot to a number of
delivery points. Operations Research, Vol. 12, 568-581
De Koster, R. / Van der Poort, E. (1998): Routing orderpickers in a warehouse: A
comparison between optimal and heuristic solutions. IIE Transactions 30, 469-480
De Koster, R. / Van der Poort, E. / Wolters, M. (1999): Efficient order batching methods
in warehouses. International Journal of Production Research, Vol. 37(7), 1479-1504
Elsayed, E.A. / Unal, I.O. (1989): Order batching algorithms and travel-time estimation for
automated storage/retrieval systems. International Journal of Production Research, Vol.
27(7), 1097-1114
Fisher, M. (1995): Vehicle Routing. in: Network Routing, M.O. Ball, T.L. Magnanti, C.L.
Monma, G.L. Nemhauser, eds. (Elsevier) Amsterdam
Gademann, N. / Van den Berg, J. / Van der Hoff, H. (1996): An order batching
algorithm for wave picking in a parallel-aisle warehouse. working paper, University of
Twente
Goetschalckx, M. / Ratliff, H.D. (1988): Order picking in an aisle. IIE Transactions 20(1),
53-62
Hackman, S.T. / Platzman, L.K. (1990): Near-optimal solution of generalized resource
allocation problems with large capacities. Operations Research, Vol. 38(5), 902-910
Hackman, S.T. / Rosenblatt (1990): Allocating items to an automated storage and retrieval
system. IIE Transactions 22(1), 7-14
Hall, R.W. (1993): Distance approximations for routing manual pickers in a warehouse. IIE
Transactions, Vol. 25(4), 76-87
Hausman, W.H. / Schwarz, L.B. / S.C. Graves (1976): Optimal storage assignment in
automatic warehousing systems. Management Science, Vol. 22, 629-638
Ratliff, H.D. / Rosenthal, A.S. (1983): Orderpicking in a rectangular warehouse: A
solvable case of the traveling salesman problem. Operations Research, Vol. 31, 507-521
Roodbergen, K.J. / De Koster, R. (1998): Determining order picking tours in a warehouse
with cross aisles. working paper, Rotterdam School of Management, Erasmus University
Rotterdam
Rosenwein, M.B. (1996): A comparison of heuristics for the problem of batching orders for
warehouse selection. International Journal of Production Research, Vol. 34(3), 657-664
Tompkins, J.A. / White J.A. / Bozer Y.A. / Frazelle E.H. / Tanchoco J.M.A. / Trevino,
J. (1996): Facilities planning. (Wiley) New York
Van Voorden, R.A. (1997): Distributiecentrum De Bijenkorf: Onderzoek naar de
loopafstanden in het bakkenmagazijn. Master’s thesis, Rotterdam School of
Management, Erasmus University Rotterdam