evaluating inventory segmentation strategies for ... · inventory optimization model a set fill...
TRANSCRIPT
Evaluating Inventory Segmentation Strategies for Aftermarket Service Parts in Heavy Industry using Linked Discrete-
Event and Monte Carlo Simulations
Author: Randolph L. Bradley
Advisor: Jarrod Goentzel
Sponsor: Heavy Industry
MIT SCM ResearchFest May 23-24, 2012
1 MIT SCM ResearchFest
Agenda
• Overview
• Hypothesis
• Methodology
• Data Analysis
• Simulating Baseline Strategy
• Simulating Alternate Strategy
• Stocking Policy
• Executive Summary
2 MIT SCM ResearchFest
Overview & Hypothesis
3 MIT SCM ResearchFest
Bernt A. Lilliestråle • Vice President, Shuttle Operation, Teekay Norway
• Teekay Petrojarl is a global leader in oil & gas shipping & offshore production
• Tech data, manuals, and standards often non-existent
• Focus on maintenance rather than maintaining data
The Persona of Heavy Industry
“My biggest fear is that I have a
ship down for maintenance, the
part is somewhere in Teekay’s
spares chain, and I don’t know it.”
– Bernt Lilliestråle
• Small fleet
‒ Five Floating Production, Storage and Offloading (FPSO) vessels
‒ Two shuttle tankers ‒ One storage tanker
• Knows his support challenges are systemic across heavy industry
Definitions
• Customers want all their parts when needed…
• Fill Rate
• Percent of parts shipped from stock on hand
• Objective function of inventory optimization
• Perfect Order Fulfillment (POF)
• Percent of orders shipped “on time and in full”
• Order Fulfillment Lead Time (OFLT)
• Average time from order placement to customer receipt
• Measures speed of service
• OFTL as contractual metric = POF within delivery window
• Modeled with simulation
MIT SCM ResearchFest 5
Hypothesis
Baseline AB Inventory Segmentation Strategy
• Group on price
• A=Parts ≤ $2,000
• B=Parts > $2,000
• Min 50% Fill Rate
MIT SCM ResearchFest 6
Alternate ABCDE Inventory Segmentation Strategy
• Group on annual use
• A=Parts 80% Annual Use
• B=Parts 15% Annual Use
• C=Parts 5% Annual Use
• D=$25 ≥ Parts > $1
• E=$1 ≤ Parts
Hypothesis: The alternate ABCDE inventory segmentation
strategy for consumable parts will outperform the baseline
AB strategy when measured on Order Fulfillment Lead
Time and inventory investment.
Baseline AB vs. Alternate ABCDE Inventory Segmentation Strategies
0%10%20%30%40%50%60%70%80%90%
100%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100%
% C
um
ula
tive
Par
t C
ou
nt
% Cumulative Annual Use (Demand * Unit Cost)
Segment % Annual
Use % Part Count
A: 80% Usage & >$25 79% 7%
B: 15% Usage & >$25 14% 14%
C: 5% Usage & >$25 4% 42%
D: $1<Unit Cost≤$25 3% 30%
E: Unit Cost≤$1 0% 7%
Total 100% 100%
MIT SCM ResearchFest 7
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
$0
$2
$5
$9
$1
4 $
19
$2
6 $
37
$5
2 $
73
$1
00
$1
43
$2
00
$2
81
$3
81
$5
45
$7
74
$1
,123
$1
,689
$2
,779
$5
,161
Cu
mu
lati
ve %
Par
t C
ou
nt
Unit Cost
Baseline AB Inventory Optimization Unit Cost vs. Cum. % Part Count
Segment % Annual
Use % Part Count
A: ≤$2,000 37% 92%
B: >$2,000 63% 8%
Total 100% 100%
Alternate ABCDE Inventory Segmentation Part Count vs. Annual Use
Alternate ABCDE Inventory Segmentation
E Consumables, 7% Parts,
$0<=Cost<=$1, 99% FR
C Consumables, 42% Parts, 5%
Annual Use, 99% FR
B Consumables, 14% Parts, 15%
Annual Use, 92% FR A Consumables, 7% Parts, 80%
Annual Use, 90% FR
D Consumables, 30% Parts,
$1<Cost<=$25, 99% FR
Inventory optimization uses a
“greedy algorithm” to select the
lowest cost quantity and mix of
parts achieving fill rate
constraint
MIT SCM ResearchFest 8
Pe
rce
nta
ge
Pe
rce
nta
ge
Pe
rce
nta
ge
Cost Cost Cost P
erc
en
tag
e
Cost
Pe
rce
nta
ge
Cost
90% Fill
Rate Goal
Average
Days Delay
“Availability”
Methodology
10 MIT SCM ResearchFest
Warehouse
Simulation Discrete
event
warehouse
simulation
New buys
arrive in lead
time
Student’s t-
test validates
sim
Fill Rate
Over Time
Statistical
Evaluation Ao, Fill Rate, and Avg Delay
Availability
Fill Rate
Avg. Delay ** Scaled to 10.00 = 100
Percentage
Cost
0
20
40
60
80
100
0 10000000 20000000 30000000 40000000 50000000 60000000
Inventory Optimization
Model B
Data Cleaning + Integration
Extract, Xform & Load (ETL)
Inventory Optimization
Model A
Set fill rate
goal
Extract
Model A
inputs and
outputs
Reuse data
for multiple
models
Model B vali-
dates Model
A fill rate
Strategic Inventory Optimization
Monte Carlo Demand Categorization & Metrics
Simulation
Discrete Event Warehouse Simulation
Metrics
Simulation
Categorization
& metrics
simulation
OFLT w/
days delay
Days Delay
OFLT Over Time OFLT w/
Confidence
Interval
MIT SCM ResearchFest 11
Simulation Methodology
70%
72%
74%
76%
78%
80%
82%
84%
86%
UCL21Sims
LCL21Sims
UCL4Sims
LCL4Sims
How much is enough? Optimal # warehouse discrete event simulations
Time
74%
76%
78%
80%
Tran
sact
ion
s in
OFL
T W
ind
ow
Sims vs. 80% Confidence Level Consumables Hi-Pri OFLT Jun-12
UCL Mean LCL
78%
79%
80%
81%
82%
Tra
nsa
ctio
ns
in O
FLT
W
ind
ow
Sims vs. 80% Confidence Level Consumables Hi-Pri OFLT Dec-12
UCL Mean LCL
80%
82%
84%
86%
Tran
sact
ion
s in
OFL
T W
ind
ow
Sims vs. 80% Confidence Level Consumables Hi-Pri OFLT Jun-13
UCL Mean LCL
Tra
ns
ac
tio
ns in
OF
LT
Win
do
w
80% confidence
interval similar from
4 to 21 iterations
Sweet spot for warehouse
discrete event simulation =
4 repetitions (+/- 1½%)
MIT SCM ResearchFest 14
+/- 1% +/- 1% +/- 1½%
Data Analysis
17 MIT SCM ResearchFest
Average Shipping Delay (Days) Consumable Parts High Priority 1.74 Consumable Parts Low Priority 1.94
Empirical Shipping Delay Distributions
23%
29% 27%
8% 7% 3% 1% 0% 0% 0% 0% 0% 0% 0%
0%
10%
20%
30%
40%
0 1 2 3 4 5 6 7 8-10 11-15 16 17-2728-38 SD39+
Empirical Shipping Delay CDF*
33%
14%
19%
13% 10% 9%
1% 0% 0% 0% 0% 0% 0% 0% 0%
5%
10%
15%
20%
25%
30%
35%
0 1 2 3 4 5 6 7 8-10 11-15 16 17-2728-38 SD39+
Empirical Shipping Delay CDF*
Hig
h P
rio
rity
L
ow
Pri
ori
ty
Co
nsu
mab
les
18 MIT SCM ResearchFest
Co
nsu
mab
les
Days
Days
*CDF = Cumulative Density Function
Perc
enta
ge
Perc
enta
ge
Simulating Baseline AB Inventory Segmentation Strategy
19 MIT SCM ResearchFest
Baseline AB Inventory Segmentation for High Priority Consumables
MIT SCM ResearchFest 20
70%
75%
80%
85%
90%
95%
100%
OF
LT
Goal
Conf High
Mean
Conf Low
Simulated ==> <== Historical
Open orders at end
of simulation cause
phantom spikes…
Rightsizing Inventory for Consumable Parts Baseline AB Inventory Segmentation Strategy
MIT SCM ResearchFest 22
Optimal inventory with
baseline AB inventory
segmentation stock levels
Optimal - Active 40%
Inactive 14%
Excess - Active 26%
Burned Down Years 1-5 - Active
20% Backorder Obligation
0%
Optimal - Active 40%
One-off Lay-in 4%
Manage to Optimal Inventory 44%
Current Customer Inventory 100%
Identify
Optimal
Inventory
Simulating Alternate ABCDE Inventory Segmentation Strategy
23 MIT SCM ResearchFest
65%
70%
75%
80%
85%
90%
OF
LT
Goal
Alternate ABCDE Inv Seg Conf High
Alternate ABCDE Inv Seg Mean
Alternate ABCDE Inv Seg Conf Low
Baseline AB Conf High
Baseline AB Mean
Baseline AB Conf Low
Alternate ABCDE Inventory Segmentation for High Priority Consumables
Ostensibly* 50% confident of
achieving metrics goal Aug 2013
Note that stock levels assume
Continuous Resupply!
*Must also address Stocking Policy: Continuous Resupply vs. Periodic Resupply
MIT SCM ResearchFest 24
Alternate ABCDE
Inventory Segmentation
Baseline AB Inventory
Segmentation
Open orders at end
of simulation cause
phantom spikes…
Optimal - Active 28%
Inactive 13% Excess - Active
40%
Burned Down Years 1-5 - Active
19% Backorder Obligation
0%
Optimal - Active 32%
One-off Lay-in 4%
Manage to Optimal Inventory 36%
Identify
Optimal
Inventory
Current Customer Inventory 100%
Rightsizing Inventory for Consumable Parts Alternate ABCDE Inventory Segmentation
MIT SCM ResearchFest 26
Optimal inventory with
alternate ABCDE inventory
segmentation stock levels
Stocking Policy for Baseline Stock Levels
MIT SCM ResearchFest 27
Stocking Policy for High Priority Consumable Parts
60%
65%
70%
75%
80%
85%
90%
95%
OF
LT
Goal
Continuous Replenishment
30 Day Periodic Resupply
90 Day Periodic Resupply
360 Day Periodic Resupply
MIT SCM ResearchFest 28
LEGEND • OFLT Goal
• Continuous Resupply o Inventory Optimization Strategy
Assumes Continuous Resupply
• 30-Day Periodic Resupply o Spares Order Review Board
Means 30 Day Periodic Resupply
• 90-Day Periodic Resupply o Quarterly Acquisition Funding
Means 90 Day Periodic Resupply
• 360-Day Periodic Resupply o Annual Acquisition Funding
Means 360 Day Periodic Resupply
Executive Summary
MIT SCM ResearchFest 32
Executive Summary
• Simulation reveals tomorrow’s supply chain today • Resolves disconnects between models and measures
• Inventory segmentation improves service parts mix
MIT SCM ResearchFest 33
Segmentation Strategy
Date OFLT Goal Achieved
Inventory Investment
Baseline AB Never 100%
Alternate ABCDE Mar-2013 80%
• Successful spares chain execution requires aligned stocking, review, and acquisition policies