How do you decide what to produce customers will buy? Marshall Fisher

ISC/SCLC SPRING 2006 MEETING
Hilton Marco Island
APRIL 29, 2006
How do you decide what to produce
when you don’t know what your
customers will buy?
Marshall Fisher
UPS Professor, The Wharton School
Cofounder and Chairman, 4R Systems
© 2006 Marshall L. Fisher
Products differ
Cost of lost sale
Low
High
Risk of obsolescence
Low
High
Forecast accuracy
High
Low
Low
High
Long
Short
Product variety
Product life cycle
Functional
© 2005 Marshall L. Fisher
Innovative
And supply strategies differ
Factory focus
Inventory Strategy
Lead-time focus
Supplier selection
Product-design strategy
High utilization
Maintain buffer
capacity
High turns
Significant buffer stocks of
components and FGs
Low cost trumps
short lead-time
Aggressively
shorten lead-time
Low cost
Integral for max
performance at min cost
Physically
efficient
© 2005 Marshall L. Fisher
Speed & flexibility
Modular to enable
postponed differentiation
Market
responsive
Respond quickly
to unpredictable
demand to
minimize
stockouts,
markdowns, and
obsolete inventory
© 2005 Marshall L. Fisher
Life cycle < 1 year
Gross Margin > 35%
High Product Variety
Functional Products
Innovative Products
Efficient
Supply Chain
Supply predictable
demand efficiently
at lowest cost
Life cycle > 2 years
Gross Margin < 35%
Low Product Variety
match
mismatch
Responsive
Supply Chain
Need to match supply strategy with product type
mismatch
match
So as to minimize total of two types of costs
Raw
Materials
•
Component
Manufacturer
Retailers
Suppliers
Physical Production/Distribution Costs
– Production Costs
– Transportation Costs
– Facility Utilization rates
– Inventory carrying cost on pipeline and cycle stocks
•
Supply/Demand Mismatch Costs
– Lost revenue and profit margin when supply is less than demand
– Product and parts scrapped or sold at a loss when supply exceeds demand
– Inventory carrying cost on safety stocks
© 2005 Marshall L. Fisher
Consumers
Dell reaps benefits from supply chain innovation
Dell
S&P 500
“Supply chain management is what it’s all about“ Tom
Meredith, Chief Financial Officer of Dell
Source: Open manufacturing Online, July 28, 1998
© 2005 Marshall L. Fisher
P & G has grown earnings faster than sales by cutting
supply chain costs
P&G Net Sales and Net Earnings 1990-98
40000
10000
35000
9000
6000
20000
5000
15000
4000
3000
10000
2000
trendline
1000
0
1998
1997
1996
1995
1994
1993
1992
1991
1990
0
Source: Company 10K reports
Note: 1993-94 accounting change makes series discontinuous.
© 2005 Marshall L. Fisher
Net Earnings ($M)
Net Sales ($M)
7000
25000
5000
During 94-98
8000
30000
Net Sales
grew 22.3%
Net Earnings
grew 71%
A page from Sport Obermeyer’s
product catalog
© 2005 Marshall L. Fisher
Next year’s
catalog
© 2005 Marshall L. Fisher
Obermeyer’s styles are fashion-forward and
change every year
© 2005 Marshall L. Fisher
The Obermeyer supply chain stretches from
Asia to Aspen
Factories in
China
DC in Denver
800 Ski
Retailers
Obermeyer’s planning calendar is driven by when it
snows
© 2005 Marshall L. Fisher
Which parka family sold best?
Black Voodoo sold 4000
© 2005 Marshall L. Fisher
Sold 4
Initial forecasts are highly inaccurate
Black Voodoo
© 2005 Marshall L. Fisher
Measuring the cost of over and under supply
Orders
Production
Lost sales
Full price = $200
4000
4
2000
2000
2000
Excess
1996
Markdown price = 130 Variable cost = 150
Lost sales cost $50 x 2000 =
$100,000
Excess cost $20 x 1996 = $39,920
Initial forecasts are highly inaccurate …
but improve
dramatically with just
a little sales data
© 2005 Marshall L. Fisher
Early write
• Bring 25 (out of 800) largest retailers to Aspen in February.
Accounts for 20% of sales.
• Put them up at the Ritz Carlton
• They ski with Klaus Obermeyer, an industry icon
• They get an advance preview of the line
• They order early
Lead time reduction
Asia
Fabric
Producer
Fabric
Dyer
undyed greige
goods
Cut/Sew
Factory
Denver
Warehouse
Retailer
Consumer
Sport Obermeyer
•
Fabric dyer lead time of several months was a problem for Obermeyer
•
Dyer has long lead time on greige goods and needed to keep their capacity utilized year round
but can change colors overnight
•
Obermeyer can predict total annual sales and sales of basic colors, but can’t predict fashion
colors
Solution
•
Offer dyer one year commitment on greige goods and capacity
•
Dye basic colors early in year and fashion colors late in
season on few days notice
© 2005 Marshall L. Fisher
Revised planning calendar
© 2005 Marshall L. Fisher
Sample buying committee projections
Which product is more predictable?
Std. Dev.
Carolyn
Laura
Tom
Kenny
Wally
Wendy
Average
Pandora
Parka
1200
1150
1250
1300
1100
1200
1200
65
Entice
Shell
1500
700
1200
300
2075
1425
1200
572
© 2005 Marshall L. Fisher
The committee process allowed us to forecast
forecast accuracy
1400
High
Error
1200
Average
Error =
155 units
1000
Average
Error =
252 units
800
Forecast
Error
600
400
200
Low
Error
0
0
110
High Agreement
220
330
Low Agreement
Standard Deviation of the Individual
Forecasts of a Six Person Committee
© 2005 Marshall L. Fisher
Obermeyer Committee Forecasts
STYLE COLOR
#
#
Wholesale
Price
LK
CO
SS
CB
JD
WS
WRO TT
GW
AM
COMM COMM
AVE
STD DEV
AS
6220
77.5
64
45
78
9
TOTAL:
6221
479
650
50
180
1359
700
200
300
300
1500
500
500
200
200
1400
350
600
350
500
1800
700
200
300
400
1600
700
200
750
450
2100
930
1450
450
1529
500
350
400
475
200
560
100
475
100
390
800 1900
600 414 300
800 614 300
300 895 300
290 876 400
1990 2799 1300
600
340
840
340
600
510
360
510
2400 1700
481
500
373
361
1715
125
233
253
217
581
500
600
400
400
400
500
300
100
385
400
400
320
1600 1500 1605
275 465
575 542
400 310
150 233
1400 1550
200
300
175
175
850
420
300
516
375
432
600
72
225
1440 1500
421
441
335
288
1485
152
128
138
137
249
700
200
500
600
2000
400 1100
350
400
400
250
500
200
350
500
300
150
1800 2000 1100
550 771 400
375 135 100
725 539 500
150 275 300
1800 1720 1300
400
600
672
300
400
800
200
300
1672 2000
597
303
526
323
1749
232
168
189
150
324
800
600
600
2100
400
300
700
300
740
1300 1000 1310
500
300
150
1400 2100 1225
400
400
320
300
100
385
1015 1415
1750 1291
150 275
1425 1650
150 233
400 310
760 1310
2028 1015
200
300
1420 1240
72
225
432
600
887
1244
323
1499
288
335
316
440
150
375
137
138
94.5
64
45
78
76
TOTAL:
6222
9.5
64
45
9
49
TOTAL:
Color forecasts
9
45
49
64
76
78
© 2005 Marshall L. Fisher
900
700
300
900
175
175
Historical Distribution of Forecast Errors
Follows the Normal Bell-Shaped Curve
© 2005 Marshall L. Fisher
The Normal Distribution Accurately Models
Demand Uncertainty at Obermeyer
Pandora Parka
15%
33%
33%
2%
2%
740
15%
970
1200
1430
1660
Probability Distribution for Sales of Pandora
Mean = 1200 Standard Deviation = 230
© 2005 Marshall L. Fisher
Cost of Under and Over Production
Pandora Parka
Wholesale Price
$ 200
Profit Margin
100
30
25
_____
$ 45
Markdown Price
$ 120
Loss
100
35
_____
($15)
Less
Supplier charges
Sales Commission
Inventory carrying/Delivery
Less
Supplier charges
Inventory carrying/delivery
© 2005 Marshall L. Fisher
Cost of Under Production
Cost of Over Production
Production Decision
if we can only buy
once
Probability = .25
Pandora Parka
1200 1430
Probabilistic Break Even Analysis
Produce to the point where
Probability we sell x Gain if we sell = Probability we don’t sell x Loss if we don’t sell
.25 x 45 = .75 x 15
Gain if we sell
© 2005 Marshall L. Fisher
Loss if we don’t sell
Accurate Response Optimization Model
•
Demand Distributions
•
Cost of Stockouts & Excess Inventory
•
Production Capacity and Minimum Constraints
Minimize Expected Cost
of Stockouts
& Excess Inventory
•
Risk Adjusted Production Commitments by Style/Color
•
Can be used as Simulation Model to measure the impact of better information or increased
supply chain flexibility
© 2005 Marshall L. Fisher
Desk top tool run by user
Factories in
China
DC in Denver
Order 50%
in April
Order 50%
in November
Forecasts
Product
Sketches
Forecast
Committee
800 Ski
Retailers
Retailers
order in Feb
& April
Results of a parallel test show profit increase equal to
1.8% of sales
Optimization Model
Decisions
Total Production (Units)
124,805
Demand
Over-Production (Units)
Under-Production (Units)
Over-Production Cost (% of Sales)
Under-Production Cost (% of Sales)
Total Cost (% of Sales)
© 2005 Marshall L. Fisher
121,432
103,831
22,036
25,094
792
1.30%
0.18%
1.48%
103,,831
7,493
1.74%
1.56%
3.30%
Legacy Process
Decisions
Retailers Loved the New Program!
© 2005 Marshall L. Fisher
Obermeyer process at World, a major Japanese retailer
© 2005 Marshall L. Fisher
Right Buy: Commercial Implementation
of the Obermeyer concept
Elements of the Obermeyer process
Which of these would be useful in your company?
• Early orders are highly predictive
• Early write -> bring 25 largest retailers to Aspen to order
early
• Cut lead times on expensive, long lead time component –
dyed fabric
• Use committee forecast process to forecast forecast errors
• Risk based production sequencing
– Replace point forecast by probability distribution
– Make predictable volume early.
– Set production volumes based on likely forecast
accuracy and cost of over and under production.
How to think about supply chain improvement
Product
Availability
How does product availability
drive revenue?
Inventory
Accurate
Forecasts
Track & improve the accuracy of
forecasts that drive decisions e.g.
parts lead time demand
© 2005 Marshall L. Fisher
Responsive
Supply Chain
Optimize cost of
lost margin,
carrying and
obsolescence
Create a framework for inventory
efficiency e.g. common parts, short
lead times, efficient small lot
production