Twodimensional Automated Planograms

1
Two-dimensional Automated Planograms
Ruibin Bai1, Tom van Woensel2, Graham Kendall1,
Edmund K. Burke1

• ASAP Research Group, School of Computer Science
  IT, University of Nottingham, Nottingham NG8 1BB,
  UK
• Technische Universiteit Eindhoven, Den Dolech 2,
  Pav. F05, Eindhoven NL 5600 MB, The Netherlands.

March 13-16th 2007 Dagstuhl
2
Motivation
Why Shelf Space Allocation?

• Retail industry is extremely competitive
• Very large product assortment (30,000).
• Shelves are expensive and limited resources.
• Research shows that attractive product layout can
  increase sales. However, designing it can be
  tedious and time consuming.
• Shelf space is related to inventory control and
  replenishment operations

3
Shelf space allocation Introduction
Traffic Flow Design
C
E
Category and brand location
Planograms
Promotions and special display
4
State-of-the-art planograms software

• Current software
• Retek SpaceMan GalaXXi
• Can check physical violations
• Drag and drop procedure (needs human interaction)
• Very few automation tools are available
• Experience based, no optimisation

A snap shot of GalaXXi 10.0 from Space IT
5
Basic Concepts
SKU (stock-keeping unit) unique identity of a
specific product or goods. SKU is the smallest
management unit in a retail store. Inventory refe
rs to the quantity of each SKU that is currently
held by a retailer displayed stock back room
stock.
Planogram A retail map or blue-print, defining
the amount of the shelf space allocated to each
SKU and its location.
6
Basic Concepts
Facing The quantity of an SKU that can be
directly seen on the shelves or fixtures by the
customers. Space elasticity Measure the
responsiveness of the sales with regards to the
change of allocated space (Curhan, 1972).
Location More attractive locations Entrance,
End of aisles, Shelves at similar eye-level.
7
Objectives

• Minimise cost (Economic Order Quantity (EOQ)
  model)
• Minimise number of replenishment
• Maximise total sales
• Maximise total profit

EOQ model
EOQ model
SSA model
8
Constraints

• Physical constraints
• 1D, 2D or even 3D
• Integrality constraints
• Constraints 1 and 2 are similar to
  constraints in multi-knapsack problem NP-Hard
  Problem
• Display requirements
• Lower and upper bounds,
• providers request, etc.
• Cluster Constraints
• Adjacency
• Weight constraints

9
A 2D SSA Model Problem Definition (1)
Given n SKUs (or items) and m shelves, with each
shelf and SKU having non-changeable sizes both in
height and in length, the problem is to allocate
appropriate facings to each SKU in order to
maximise the total sales.

• Notation
• xij length facing of shelf j allocated to SKU I
• pij Stacking coefficient
• xi total facing and

10
A 2D SSA Model Problem Definition (2)

• Notation
• yij
• Fi demand function
• A
• D
• c

otherwise
Location factor
11
A 2D SSA Model
st.
12
1D vs 2D Model
A numerical example m4, n4 (drawn from (Hwang
et al. 2004)).
Sales
H. Hwang, B. Choi, M.-J. Lee, A model for shelf
space allocation and inventory control
considering location and inventory level effects
on demand, International Journal of Production
Economics 97 (2) (2005) 185-195.
13
Optimisation Methodologies

• Gradient approach
• Meta-heuristic
• Multiple neighbourhood search approach hybridised
  with a simulated annealing hyper-heuristic
  learning mechanism.
• Neighbourhoods swap, shift, Interchange, add
  facing, delete facing.

14
Simulated Annealing Hyper-heuristic
Simulated Annealing Hyper-heuristic
Apply the selected heuristic
SA Criterion

• For example
• No. of heuristics
• The changes in evaluation function
• A new solution or not
• The distance between two solutions
• Whether it gets stuck or not
• Others

SA Criterion
Stochastic Heuristic Selection Mechanism
Feedback
Collecting domain-independent information
Domain Barrier

• Problem representation
• Evaluation Function
• Initial Solution
• Others

Heuristic Repository
Problem Domain
15
Empirical input data

• Collected from a European supermarket chain,
  experiment data contained SKUs from 44 stores
• Data are separated into two groups based on the
  store sizes large/ small.
• Parameters estimation (a, ß )
• ---Linear regression
• Two problem instances were created
• Pn6 m3, n6
• Pn29 m5, n29

16
Computational results (1)
17
Computational results (2)
Computational results for Pn29
18
Sensitivity Analysis
Shelf Space
19
Sensitivity Analysis
Sensitivity of parameter estimation error
20
Conclusions

• Shelf space allocation and its relationship with
  multi- knapsack problem
• A practical model that be used to automate and
  optimise the design of planograms and product
  layout.
• Heuristic/meta-heuristic approaches for
  optimising retail shelf space allocation
• Future work uncertainty of market and demand
• --stochastic programming models?
• --integrated with inventory control models
• --integrate with RFID systems

21
Optimising Retail Shelf Space Allocation
Thank you!!! Comments / Questions?