Philip Kilby Principal Researcher NICTA and ANU [email protected] Logistics for Fast-Moving Consumer Goods: A Focus on Profits
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Philip Kilby
Principal Researcher
NICTA and ANU
Logistics for Fast-Moving Consumer
Goods: A Focus on Profits
2 / 38
NICTA
• National Information and Communications Technology
Australia (NICTA)
• Research in ICT since 2004
• Major Labs in Sydney, Melbourne, Canberra
• 700 researchers, including 300 PhD students
• Currently government funded
• Areas:
– Broadband and the Digital Economy
– Health
– Infrastructures, Transport and Logistics
– Safety and Security
Use-inspired Research
3 / 38
Outline
• Motivation
• Techniques and concepts
• Case study
• Conclusions
4 / 38
Motivation
Problem area:
• Delivery of goods and services to customers
• This talk will concentrate on
– Repeated routes: same route every week
– Delivery of goods from depots to customers
– e.g.: Delivery of bread from distribution centres to
customers
• Line-haul component (bakery to distribution centre)
considered separately.
5 / 38
Motivation
6 / 38
Motivation
7 / 38
Motivation
Traditional Vehicle Routing:
• Given
– a set of customers, and
– a fleet of vehicles
• Find the routes which cover all customers at minimum cost
8 / 38
Motivation
$ 27 $ 72
$ 190 $ 3
$ 258 $ 350
Revenue $550
Cost $350
Profit $200
9 / 38
Motivation
$ 27 $72
$ 190 $ 3
$ 258 $ 300
Revenue $550
Cost $300
Profit $250
10 / 38
Motivation
$ 27 $ 72
$ 190 $ 3
$ 258 $ 250
Revenue $520
Cost $250
Profit $270
11 / 38
Motivation
Profit-based vehicle routing (a.k.a. Profitable Tour Problem)
• Given
– customers with known revenues,
– vehicles with known fixed and variable operating costs
• Choose
– which customers to visit
– which vehicles and routes to use
•to maximize profit (= revenue – cost)
12 / 38
Motivation
Alternative channels
• Changed frequency (5 days Mon, Wed, Fri)
• Self-serve
• Buying groups
• New wholesale venues
Alternative strategies
• Impose a delivery charge
• Find more customers near underperforming customers
• Avoid the problem in the first place –
check cost of adding before signing customer
13 / 38
Techniques and concepts
14 / 38
Marginal cost
(cost with customer) minus (cost without customer)
15 / 38
Marginal cost
(cost with customer) minus (cost without customer)
16 / 38
Marginal cost
(cost with customer) minus (cost without customer)
works for consecutive customers
17 / 38
Marginal cost
(cost with customer) minus (cost without customer)
18 / 38
Marginal cost
(cost with customer) minus (cost without customer)
19 / 38
There is more…
(cost with customer)
20 / 38
There is more…
(cost without customer)
21 / 38
There is more…
(cost with customer)
22 / 38
Removing vehicles
Revenue $250,000 p/a
Fixed cost $100,000
Variable cost $100,000
Profit $ 50,000
Revenue $190,000 p/a
Fixed cost $ 50,000
Variable cost $ 70,000
Profit $ 70,000
1
2 1’
23 / 38
Removing vehicles
Revenue $250,000 p/a
Fixed cost $100,000
Variable cost $100,000
Profit $ 50,000
Revenue $250,000 p/a
Fixed cost $ 80,000
Variable cost $ 80,000
Profit $ 90,000
1
2
1’
Fleet size and mix is an
important part of identifying
profitable customers
24 / 38
Modified OrOpt
• OrOpt
– Given a starting tour, sets Sk of k consecutive customers
are moved from one position to another position (in
forward and reverse order) in the tour
• Modified OrOpt
– Just like OrOpt, but with one extra position in the tour,
which is the unassigned position
25 / 38
Large Neighborhood Search
For various fleet configurations:
• Solution = OrOptSweep(<input>, k)
• For i = 1 to n
Solution’ = Destroy(Solution)
Solution’’ = Repair(Solution’)
Solution’’’ = OrOptSweep(Solution’’, k)
if (Solution’’’ better than Solution)
Solution = Solution’’’
26 / 38
Case study
27 / 38
• Fast Moving Consumer Goods (FMCG) company
• 1,000,000 products
• 20,000 customers across Australia
• 100 distribution centres
• 600 vehicles
• 80% of transportation costs are “last mile logistics”
Case study
28 / 38
Case study
Classification
%Customers %Volume
Large Customers 13% 64%
Medium Customers 12% 17%
Small Customers 75% 19%
29 / 38
Case study
In some ways, a classic VRP
• Capacity constraints
• (Soft) time window constraints
• Compatibility constraints
(some customers can’t use some vehicles)
• Maximum duration constraint
Plus, a googly
• Same-driver constraint:
– Even though different routes are driven each day, the same
driver must visit a customer each day they are visited
30 / 38
Customers
31 / 38
Monday
32 / 38
Tuesday
33 / 38
Wednesday
34 / 38
“Grand tour”
35 / 38
Method
• Input
– Almost 20,000 customers
– More than 600 trucks
– Demand data for 91 days (= 13 weeks)
• Output
– Current customer base and current routes (Benchmark)
– Current customers base and optimized routes (VRP run)
– Optimised customer base and optimized routes (PTP run)
36 / 38
Results
• 15% of the customers (almost 3,000) were determined to be
unprofitable
• Significant cost savings identified by serving only 85% of the
customers
Classification %Contribution
Reduction
Large Customers 0%
Medium Customers 12%
Small Customers 88%
(13%→15%)
(12%→12%)
(75%→73%)
37 / 38
Implementation
Actual implementation achieved bulk of identified savings
• Change of channel for more than 1,000 customers
• Fleet reduced by 15%
• Distance reduced by 1,000,000 km (1 Gm?)
• Total transport costs reduced by 10%
38 / 38
Conclusions
Focus on total profitability
• More than just minimising costs
• Choosing customers, fleet and routes
• It works
Current research: Cost allocation methods
• Allocate total cost to all customers
• Correct answer: “Shapely values”
– exact, but very hard to calculate
• Need supportable approximations
From imagination to impact
40 / 38
Questions?
41 / 38
Sequence dependent
1
2
1
2
42 / 38
Sequence dependent
2
1
2
1
43 / 38
44 / 38
45 / 38
46 / 38
47 / 38
Related work
• Shapley value [Shapley, 1953]
– Represents the average marginal cost of a player in a co-
operative game
• Profitable Tour Problem (PTP) [Dell’Amico et al., 1995]
– Max viV piyi – (vi, vj)A cijxij
– Subject to:
– vjV\{vj} xji = yi (vi V)
– vjV\{vj} xij = yi (vi V)
– Subtour elimination
– y1 = 1
– xij {0, 1} ((vi, vj) A)
– yi {0, 1} (vi V)
− Fixed costs
− Vehicle capacities
− Service times
− Time windows
− Maximum durations
− Multiple routes
− Routes specifications
48 / 38
Future work
• Compare with other cost allocation and PTP methods
• Look into methods that are more “robust” to some change in
routing
• Explain why a customer is removed
– Helps to identify how to make a customer profitable again
Menkes van den Briel
Researcher
NICTA and UNSW
Phil Kilby
Principal Researcher
NICTA and ANU
Related Documents
