cpms daniel h. wagner prize competition project portfolio ... · ©intel corp cpms daniel h. wagner...
TRANSCRIPT
1 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Siddhartha Sampath and Karl Kempf, Intel Corporation Prof. Esma Gel, Fulton Schools of Engineering, ASU Prof. John Fowler, Carey School of Business, ASU
Project Portfolio Planning at Intel Corporation
CPMS Daniel H. Wagner Prize Competition
Agenda 1.Company Background 2.The Business Problem 3.Novel Mathematical Solutions 4.The Decision Process 5.Growing Business Impact
2 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Founded July 18th, 1968
6
3 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Predictable SILICON TRACK RECORD Executing to Moore’s Law Since
1965 Enabling new devices with higher functionality and complexity while controlling power, cost and size.
4 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Desktops Laptops Embedded Smartphones Tablets Entertainment
Since 15 November 1971 when Intel introduced the world’s first
microprocessor, we have developed a variety of products that are sold into an
ever-increasing number of markets. 4004 with ~2.3x103
transistors
Servers
with ~5.0x109
transistors
with ~4.7x107
transistors
5 CPMS Daniel H. Wagner Prize Competition © Intel Corp
The Business Problem - There are products the company wants to launch over time. - Each product require multiple substantial engineering projects. - The products all bring a benefit to the company. - The projects are interrelated in a variety of ways. - The projects all cost something.
6 CPMS Daniel H. Wagner Prize Competition © Intel Corp
- There are products the company wants to launch over time. - Each product require multiple substantial engineering projects. - The products all bring a benefit to the company. - The projects are interrelated in a variety of ways. - The projects all cost something.
- There is not enough budget to fund all the projects! - How should the budget be allocated to maximize benefit? - How can various qualitative & quantitative criteria be included?
7 CPMS Daniel H. Wagner Prize Competition © Intel Corp
The Business Problem
- There are products the company wants to launch over time. - Each product require multiple substantial engineering projects. - The products all bring a benefit to the company. - The projects are interrelated in a variety of ways. - The projects all cost something.
- There is not enough budget to fund all the projects! - How should the budget be allocated to maximize benefit? - How can various qualitative & quantitative criteria be included?
- What portfolio of engineering projects should be funded?
8 CPMS Daniel H. Wagner Prize Competition © Intel Corp
The Business Problem
At a scale relevant to INTEL Corporation … - the problem is too complex to solve well
using only business intuition … - there are more business details than can be
analytically modelled …
9 CPMS Daniel H. Wagner Prize Competition © Intel Corp
The Business Problem
At a scale relevant to INTEL Corporation … - the problem is too complex to solve well
with business intuition … - there are more business details than can be
analytically modelled …
The solution process must include: - analytics informing intuition - intuition informing analytics
10 CPMS Daniel H. Wagner Prize Competition © Intel Corp
The Business Problem
THE ANALYTICS
Our Mathematical Solution Components
11 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Mapping: To visualize projects, products, and relations
Enumeration: To efficiently enumerate decision alternatives
Valuation: To quantify business decisions and impacts
Optimization: To generate a feasible set of alternatives
What-Ifs: To enable the intuition to inform the analytics
1) MAPPING:
12 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Essential to: Help analysts visualize the business and capture project and product input in a systematic fashion ready for analysis.
Difficult Because: The relevant information and knowledge is distributed across the organization, and there are a variety of complex interactions between projects and products.
Our Contribution : Education and application of the principals of Decision Analysis as well as a review board to increase input data quality.
13 CPMS Daniel H. Wagner Prize Competition © Intel Corp
projects impacting markets
projects impacting platforms
projects impacting
silicon
Product-1 Product-3
Project-a
Project-b Project-d Project-c
Project-y Project-z Project-x
Market-1 Market-2 Market-3
Product-2
Prospect 1,1 2,2 2,3 3,3
MAPPING: Entities, Interactions and Values
required optional Influence
AND, ORs, SEQ
NPV
TAM SAM
Eng. Cost
VOL ASP
Mfg. Cost
DATA
14 CPMS Daniel H. Wagner Prize Competition © Intel Corp
projects impacting markets
projects impacting platforms
projects impacting
silicon
Product-1 Product-3
Project-a
Project-b Project-d Project-c
Project-y Project-z Project-x
Market-1 Market-2 Market-3
Product-2
Prospect 1,1 2,2 2,3 3,3
MAPPING: Entities, Interactions and Values
required optional Influence
AND, ORs, SEQ
NPV
TAM SAM
Eng. Cost
VOL ASP
Mfg. Cost
DATA
NPV
15 CPMS Daniel H. Wagner Prize Competition © Intel Corp
projects impacting markets
projects impacting platforms
projects impacting
silicon
Product-1 Product-3
Project-a
Project-b Project-d Project-c
Project-y Project-z Project-x
Market-1 Market-2 Market-3
Product-2
Prospect 1,1 2,2 2,3 3,3
TAM SAM
Eng. Cost
VOL ASP
Mfg. Cost
DATA
MAPPING: Entities, Interactions and Values
required optional Influence
AND, ORs, SEQ
2) Enumeration:
16 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Essential to: Identify and generate the list of all feasible decision alternatives.
Difficult Because: The large number of decisions and interrelationships between various quantitative metrics.
Our Contribution: Enumeration algorithm that automates generation of decision units.
17 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Enumerating Decisions: The NPV (and other attributes) of each prospect is a function of different variables that depend on the various projects that impact it as well as the other products that are entering a given market.
M1
MKTG P1 P2
ENG1 ENG2
ENG3
Portfolio Unit
1st Year Spending Remaining Spending
ENG1 12 12
ENG2 15 22
MKTG 11 16
ENG3 9 11
Decision Unit Portfolio Unit & Scenario
NPV/Discounted Cash Flow
C1 +Neighborhood = C1* 166
C1* - C2 75
C1* - MKTG 125
C1* -C2, MKTG 42
C1* -ENG3 55
C1* -C2, ENG3 59
C1* -MKTG, ENG3 23
C1* -C2, MKTG, ENG3 30
18 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Enumerating Decisions: The NPV (and other attributes) of each prospect is a function of different variables that depend on the various projects that impact it as well as the other products that are entering a given market.
M1
MKTG P1 P2
ENG1 ENG2
ENG3
Portfolio Unit
1st Year Spending Remaining Spending
ENG1 12 12
ENG2 15 22
MKTG 11 16
ENG3 9 11
19 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Decision Unit (Portfolio Unit & Scenario)
NPV/Discounted Cash Flow
C1 +Neighborhood = C1* 166
C1* - C2 75
C1* - MKTG 125
C1* -C2, MKTG 42
C1* -ENG3 55
C1* -C2, ENG3 59
C1* -MKTG, ENG3 23
C1* -C2, MKTG, ENG3 30
C2 +Neighborhood = C2* 23
C2* -C1 19
C2* -ENG3 20
C2* -C1, ENG3 5
Enumerating Decisions: The NPV (and other attributes) of each prospect is a function of different variables that depend on the various projects that impact it as well as the other products that are entering a given market.
M1
MKTG P1 P2
ENG1 ENG2
ENG3
Portfolio Unit
1st Year Spending Remaining Spending
ENG1 12 12
ENG2 15 22
MKTG 11 16
ENG3 9 11
Generating All Possible Decision Units
Input: A portfolio unit, its neighborhood and relationships
20 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Generating All Possible Decision Units
Input: A portfolio unit, its neighborhood and relationships
Output: All possible decision units of a given portfolio unit.
21 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Generating All Possible Decision Units
Input: A portfolio unit, its neighborhood and relationships
Call Solver
Output: All possible decision units of a given portfolio unit.
Does solution x’ exist?
22 CPMS Daniel H. Wagner Prize Competition © Intel Corp
M1: Maximize:∑ 𝑥𝑥𝑗𝑗𝑚𝑚𝑗𝑗=1 (1)
Subject to: 𝑥𝑥𝑗𝑗 = 1, ∀𝑗𝑗 ∈ 𝐼𝐼𝑖𝑖 (2) 𝑥𝑥𝑗𝑗 = 0, ∀𝑗𝑗 ∈ 𝐸𝐸𝑖𝑖 (3) ∑ 𝑥𝑥𝑗𝑗𝑗𝑗 𝜖𝜖𝑀𝑀𝑖𝑖
= 1, (4) ∑ 𝑥𝑥𝑗𝑗 ≤ 1𝑗𝑗𝜖𝜖𝑄𝑄𝑖𝑖 , (5) Let x’ be the solution to M1
∑ 𝑥𝑥𝑖𝑖 ≤ ∑ 𝑥𝑥𝑖𝑖′𝑚𝑚𝑖𝑖=1𝑖𝑖 𝜖𝜖{𝑖𝑖:𝑥𝑥𝑖𝑖
′=1} − 1 𝑥𝑥𝑥 is the previous scenario
Set Up Model
No: Stop
3) Valuation:
23 CPMS Daniel H. Wagner Prize Competition © Intel Corp
• Essential to: Value decision units in preparation for portfolio optimization.
• Difficult Because: Large number of correlated variables and scenarios that need to be valued.
• Our Contribution : Combination of user estimates and curve-fitting forecasts all values for the enumerated model.
24 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Input: Set of Decision Units from previous stage
Valuating a Decision Unit
25 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Input: Set of Decision Units from previous stage
Output: List of metrics associated with each decision unit (NPV, Revenue by Year, Spending by Year …)
Valuating a Decision Unit
Generating Values for a Decision Unit
26 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Data
(ASP
, VO
L., M
FG$,
EN
G$)
90th %ile realization
at time t
10th %ile realization at time t
Forecast Historical
Time
Data = ASP, Volume, Manufacturing Cost, Engineering Cost
27 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Capture expert estimates for simulation variables
Capture historical realization of simulation
variables
Historical curve fitting: non-linear regression
and time series analysis using ARIMA models
Custom Monte Carlo Simulation valuating
decision unit
Input: Set of Decision Units from previous stage
Output: List of metrics associated with each decision unit (NPV, Revenue by Year, Spending by Year …)
Project Planning Records
Forecast Variables
Valuating a Decision Unit
4) OPTIMIZATION:
28 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Essential to: Generating an ordered set of non-dominated portfolios.
Difficult Because: Large number of possible combinations and complex relationships between decision units.
Our Contribution : Optimization algorithm quickly generates efficient frontier of non-dominated portfolios.
29 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Step 1: We generate the first portfolio including all projects without any spending or headcount constraints.
Port
folio
NPV
Portfolio Spending
Algorithm to Generate Non-Dominated Solutions
30 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Step 1: We generate the first portfolio including all projects without any spending or headcount constraints. Step2: To ensure that the next portfolio is a non-dominated portfolio, we add three cuts.
Algorithm to Generate Non-Dominated Solutions Po
rtfo
lio N
PV
Portfolio Spending
31 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Step 1: We generate the first portfolio including all projects without any spending or headcount constraints. Step2: To ensure that the next portfolio is a non-dominated portfolio, we add three cuts. Step 3: We continue adding cuts and repeating Step 2, until there are no more feasible portfolios.
Port
folio
NPV
Portfolio Spending
Algorithm to Generate Non-Dominated Solutions
Algorithm to Generate Non-Dominated Solutions
Input: All decision units and associated attributes
32 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Algorithm to Generate Non-Dominated Solutions
Input: All decision units and associated attributes
Output: Ordered Set of non-dominated portfolios.
33 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Algorithm to Generate Non-Dominated Solutions
Input: All decision units and associated attributes
Output: Ordered Set of non-dominated portfolios.
Setup Model
34 CPMS Daniel H. Wagner Prize Competition © Intel Corp
M2: Maximize: ∑ ∑ 𝑝𝑝𝑗𝑗(𝑖𝑖)𝑥𝑥𝑗𝑗(𝑖𝑖)𝐷𝐷(𝑖𝑖)𝑗𝑗=1
𝑁𝑁𝑖𝑖=1 (9)
𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬 𝐬𝐬𝐭𝐭: Spending Constraints 𝐵𝐵min ≤ ∑ ∑ 𝑠𝑠𝑗𝑗 𝑖𝑖 𝑥𝑥𝑗𝑗 𝑖𝑖 ≤ 𝐵𝐵𝑚𝑚𝑚𝑚𝑥𝑥 𝐷𝐷(𝑖𝑖)
𝑗𝑗=1𝑁𝑁𝑖𝑖=1 (10)
One Valuation/Unit Constraint ∑ 𝑥𝑥𝑗𝑗 𝑖𝑖 ≤ 𝑦𝑦𝑖𝑖 𝐷𝐷(𝑖𝑖)𝑗𝑗=1 ∀ 𝑖𝑖 = 1, … ,𝑁𝑁 (11)
Inclusion Constraint ∑ 𝐶𝐶𝑗𝑗 𝑖𝑖 ,𝑘𝑘𝑁𝑁𝑘𝑘=1 𝑥𝑥𝑗𝑗 𝑖𝑖 − ∑ 𝐶𝐶𝑗𝑗 𝑖𝑖 ,𝑘𝑘
𝑁𝑁𝑘𝑘=1 𝑦𝑦𝑘𝑘 ≤ 0 ∀ 𝑗𝑗 𝑖𝑖 = 1, … ,𝐷𝐷 𝑖𝑖 (12)
Exclusion Constraint ∑ 𝐸𝐸𝑗𝑗 𝑖𝑖 ,𝑘𝑘𝑁𝑁𝑘𝑘=1 𝑥𝑥𝑗𝑗 𝑖𝑖 + ∑ 𝐸𝐸𝑗𝑗 𝑖𝑖 ,𝑘𝑘
𝑁𝑁𝑘𝑘=1 𝑦𝑦𝑘𝑘 ≤ ∑ 𝐸𝐸𝑗𝑗 𝑖𝑖 ,𝑘𝑘
𝑁𝑁𝑘𝑘=1 ∀ 𝑗𝑗 𝑖𝑖 = 1, … ,𝐷𝐷 𝑖𝑖 (13)
Let y’ be the solution to M2
Algorithm to Generate Non-Dominated Solutions
Input: All decision units and associated attributes
Output: Ordered Set of non-dominated portfolios.
Setup Model
Calculate convex hull of generated portfolios
35 CPMS Daniel H. Wagner Prize Competition © Intel Corp
M2: Maximize: ∑ ∑ 𝑝𝑝𝑗𝑗(𝑖𝑖)𝑥𝑥𝑗𝑗(𝑖𝑖)𝐷𝐷(𝑖𝑖)𝑗𝑗=1
𝑁𝑁𝑖𝑖=1 (9)
𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬 𝐬𝐬𝐭𝐭: Spending Constraints 𝐵𝐵min ≤ ∑ ∑ 𝑠𝑠𝑗𝑗 𝑖𝑖 𝑥𝑥𝑗𝑗 𝑖𝑖 ≤ 𝐵𝐵𝑚𝑚𝑚𝑚𝑥𝑥 𝐷𝐷(𝑖𝑖)
𝑗𝑗=1𝑁𝑁𝑖𝑖=1 (10)
One Valuation/Unit Constraint ∑ 𝑥𝑥𝑗𝑗 𝑖𝑖 ≤ 𝑦𝑦𝑖𝑖 𝐷𝐷(𝑖𝑖)𝑗𝑗=1 ∀ 𝑖𝑖 = 1, … ,𝑁𝑁 (11)
Inclusion Constraint ∑ 𝐶𝐶𝑗𝑗 𝑖𝑖 ,𝑘𝑘𝑁𝑁𝑘𝑘=1 𝑥𝑥𝑗𝑗 𝑖𝑖 − ∑ 𝐶𝐶𝑗𝑗 𝑖𝑖 ,𝑘𝑘
𝑁𝑁𝑘𝑘=1 𝑦𝑦𝑘𝑘 ≤ 0 ∀ 𝑗𝑗 𝑖𝑖 = 1, … ,𝐷𝐷 𝑖𝑖 (12)
Exclusion Constraint ∑ 𝐸𝐸𝑗𝑗 𝑖𝑖 ,𝑘𝑘𝑁𝑁𝑘𝑘=1 𝑥𝑥𝑗𝑗 𝑖𝑖 + ∑ 𝐸𝐸𝑗𝑗 𝑖𝑖 ,𝑘𝑘
𝑁𝑁𝑘𝑘=1 𝑦𝑦𝑘𝑘 ≤ ∑ 𝐸𝐸𝑗𝑗 𝑖𝑖 ,𝑘𝑘
𝑁𝑁𝑘𝑘=1 ∀ 𝑗𝑗 𝑖𝑖 = 1, … ,𝐷𝐷 𝑖𝑖 (13)
Let y’ be the solution to M2
Yes
Add Cut
Does y’ exist?
Exclude Previous Solution Cut ∑ 𝑦𝑦𝑖𝑖 − ∑ 𝑦𝑦𝑖𝑖𝑖𝑖 𝜖𝜖{𝑦𝑦𝑖𝑖
′=0} ≤ ∑ 𝑦𝑦𝑖𝑖′𝑁𝑁𝑖𝑖=1𝑖𝑖 𝜖𝜖{𝑖𝑖:𝑦𝑦𝑖𝑖
′=1} − 1 Spending Cut ∑ ∑ 𝑠𝑠𝑗𝑗 𝑖𝑖 𝑥𝑥𝑗𝑗 𝑖𝑖 ≤ 𝐵𝐵𝑦𝑦′
𝐷𝐷(𝑖𝑖)𝑗𝑗=1
𝑁𝑁𝑖𝑖=1
5) PREPARING for WHAT-IFS:
36 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Essential to: Generating analytic solutions to respond to intuitive what-ifs.
Difficult Because: Even apparently simple what-ifs require analysis since forcing a project in or out may change valuations of other units.
Our Contribution : Efficient IP’s quickly generate alternate portfolios. A fractional binary integer program helps generate feasible portfolios. Both provide analytical responses to intuitive what-ifs.
37 CPMS Daniel H. Wagner Prize Competition © Intel Corp
- Dollar budget - Headcount budget
- Revenue - Volume
- Individual projects - Groups of projects
PREPARING for “Intuitive” WHAT-IFS
Spending Durations (1yr vs. 3yrs)
38 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Different Three Year Spending
Comparable One Year Spending
PO
RTF
OLI
O N
PV
PO
RTF
OLI
O N
PV
PORTFOLIO SPENDING NEXT 1 YEAR
PORTFOLIO SPENDING NEXT 3 YEARS
100m 300m
39 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Input:
Maximize: 𝑼𝑼𝑇𝑇𝒘𝒘+𝛼𝛼
𝑹𝑹𝑇𝑇𝒘𝒘+𝛽𝛽 (22)
Subject to: 𝐴𝐴𝒘𝒘 ≤ 𝑏𝑏, (23)
𝐖𝐖𝐖𝐖𝐬𝐬𝐖𝐖𝐬𝐬: 𝒘𝒘𝜖𝜖 0,1 𝑞𝑞 (24)
Spending Durations
Output: Set of additional portfolios
40 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Input:
Maximize: 𝑼𝑼𝑇𝑇𝒘𝒘+𝛼𝛼
𝑹𝑹𝑇𝑇𝒘𝒘+𝛽𝛽 (22)
Subject to: 𝐴𝐴𝒘𝒘 ≤ 𝑏𝑏, (23)
𝐖𝐖𝐖𝐖𝐬𝐬𝐖𝐖𝐬𝐬: 𝒘𝒘𝜖𝜖 0,1 𝑞𝑞 (24)
Spending Durations
Does y’ exist? Solve
Need more solutions?
Output: Set of additional portfolios
No
41 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Input:
Maximize: 𝑼𝑼𝑇𝑇𝒘𝒘+𝛼𝛼
𝑹𝑹𝑇𝑇𝒘𝒘+𝛽𝛽 (22)
Subject to: 𝐴𝐴𝒘𝒘 ≤ 𝑏𝑏, (23)
𝐖𝐖𝐖𝐖𝐬𝐬𝐖𝐖𝐬𝐬: 𝒘𝒘𝜖𝜖 0,1 𝑞𝑞 (24)
Under the assumption that the feasible region is non-empty and bounded 𝑥𝑥𝑗𝑗 𝑖𝑖𝑧𝑧 = 1
∑ ∑ 𝑅𝑅𝑗𝑗 𝑖𝑖 𝑥𝑥𝑗𝑗 𝑖𝑖𝑧𝑧𝐷𝐷 𝑖𝑖
𝑗𝑗=1𝑁𝑁𝑖𝑖=1 +𝛽𝛽
. 𝑥𝑥𝑗𝑗 𝑖𝑖 (38)
𝑦𝑦𝑖𝑖𝑧𝑧 = 1∑ ∑ 𝑅𝑅𝑗𝑗 𝑖𝑖 𝑥𝑥𝑗𝑗 𝑖𝑖
𝑧𝑧𝐷𝐷 𝑖𝑖𝑗𝑗=1
𝑁𝑁𝑖𝑖=1 +𝛽𝛽
.𝑦𝑦𝑖𝑖 (39)
𝑡𝑡 = 1∑ ∑ 𝑅𝑅𝑗𝑗(𝑖𝑖)𝑥𝑥𝑗𝑗 𝑖𝑖
𝑧𝑧𝐷𝐷(𝑖𝑖)𝑗𝑗=1
𝑁𝑁𝑖𝑖=1 +𝛽𝛽
(40)
M3: Maximize∑ ∑ 𝑈𝑈𝑗𝑗(𝑖𝑖)𝑥𝑥𝑗𝑗 𝑖𝑖𝑧𝑧𝐷𝐷(𝑖𝑖)
𝑗𝑗=1𝑁𝑁𝑖𝑖=1 + 𝛼𝛼 (25)
𝐒𝐒𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬𝐬 𝐬𝐬𝐭𝐭: (10), (11), (12), (13)
∑ ∑ 𝑅𝑅𝑗𝑗 𝑖𝑖 𝑥𝑥𝑗𝑗 𝑖𝑖𝑧𝑧𝐷𝐷 𝑖𝑖
𝑗𝑗=1𝑁𝑁𝑖𝑖=1 + 𝛽𝛽𝑡𝑡 = 1, (26)
−𝑥𝑥𝑗𝑗 𝑖𝑖𝑧𝑧 + M𝑥𝑥𝑗𝑗 𝑖𝑖 + 𝑡𝑡 ≤ M,∀𝑖𝑖𝜖𝜖 1, . . ,𝑁𝑁 , 𝑗𝑗𝜖𝜖{1, … ,𝐷𝐷 𝑖𝑖 } (27)
−𝑦𝑦𝑗𝑗𝑧𝑧 + M𝑦𝑦𝑖𝑖 + 𝑡𝑡 ≤ M,∀𝑖𝑖𝜖𝜖 1, . . ,𝑁𝑁 (28) 𝑥𝑥𝑗𝑗 𝑖𝑖𝑧𝑧 + M𝑥𝑥𝑗𝑗 𝑖𝑖 − 𝑡𝑡 ≤ M,∀𝑖𝑖𝜖𝜖 1, . . ,𝑁𝑁 , 𝑗𝑗𝜖𝜖 1, … ,𝐷𝐷 𝑖𝑖 (29) 𝑦𝑦𝑗𝑗𝑧𝑧 + M𝑦𝑦𝑖𝑖 − 𝑡𝑡 ≤ M,∀𝑖𝑖𝜖𝜖 1, . . ,𝑁𝑁 (30) 𝑥𝑥𝑗𝑗 𝑖𝑖𝑧𝑧 − M𝑥𝑥𝑗𝑗 𝑖𝑖 ≤ 0,∀𝑖𝑖𝜖𝜖 1, . . ,𝑁𝑁 , 𝑗𝑗𝜖𝜖{1, … ,𝐷𝐷 𝑖𝑖 } (31) 𝑦𝑦𝑗𝑗𝑧𝑧 − M𝑦𝑦𝑖𝑖 ≤ 0,∀𝑖𝑖𝜖𝜖 1, . . ,𝑁𝑁 (32) −M𝑥𝑥𝑗𝑗 𝑖𝑖
𝑧𝑧 + 𝑥𝑥𝑗𝑗 𝑖𝑖 ≤ 0,∀𝑖𝑖𝜖𝜖 1, . . ,𝑁𝑁 , 𝑗𝑗𝜖𝜖{1, … ,𝐷𝐷 𝑖𝑖 } (33) −M𝑦𝑦𝑗𝑗𝑧𝑧 + 𝑦𝑦𝑖𝑖 ≤ 0,∀𝑖𝑖𝜖𝜖 1, . . ,𝑁𝑁 (34) −𝑡𝑡 ≤ 0 (35)
𝟏𝟏∑ ∑ 𝑅𝑅𝑗𝑗(𝑖𝑖)
𝐷𝐷(𝑖𝑖)𝑗𝑗=1
𝑁𝑁𝑖𝑖=1
≤ 𝑡𝑡 ≤ 1min
j(i)={j(i):Rj(i)>0,𝑖𝑖𝑖𝑖 1,…,𝑁𝑁 ,𝑗𝑗𝑖𝑖 1,…,𝐷𝐷 𝑖𝑖 }(𝑅𝑅𝑗𝑗(𝑖𝑖))
(36)
𝑀𝑀 = 1min
j(i)={j(i):Rj(i)>0,𝑖𝑖𝑖𝑖 1,…,𝑁𝑁 ,𝑗𝑗𝑖𝑖 1,…,𝐷𝐷 𝑖𝑖 }(𝑅𝑅𝑗𝑗(𝑖𝑖))
(37)
Let y’ be the solution to M3
Transform
Record the solution y’ Set up the following cut: ∑ 𝑦𝑦𝑖𝑖 − ∑ 𝑦𝑦𝑖𝑖𝑖𝑖 𝜖𝜖{𝑦𝑦𝑖𝑖
′=0} ≤ ∑ 𝑦𝑦𝑖𝑖′𝑁𝑁𝑖𝑖=1𝑖𝑖 𝜖𝜖{𝑖𝑖:𝑦𝑦𝑖𝑖
′=1} − 1
Set up model
No
Spending Durations
42 CPMS Daniel H. Wagner Prize Competition © Intel Corp
-4,564 Drop Unit A
-1,985 Drop Unit C -7,269 Drop Unit D -2,885 Drop Unit E
PORTFOLIO #1 TOTAL NPV = 51,326
Drop Unit B
An incremental value report showing the
consequence of dropping a unit from a portfolio under consideration.
Incremental Value Report
2,142
43 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Drop Unit D -1,844
PORTFOLIO #2 TOTAL NPV = 44,057
PORTFOLIO #1 TOTAL NPV = 51,326
Devalued Unit B -5,425
DELTA - 7,269
A more detailed report showing the
consequences of dropping unit D
when unit B depends on the
presence of unit D.
-4,564 Drop Unit A
-1,985 Drop Unit C -7,269 Drop Unit D -2,885 Drop Unit E
PORTFOLIO #1 TOTAL NPV = 51,326
Drop Unit B 2,142
Incremental Value Report An incremental value report showing the
consequence of dropping a unit from a portfolio under consideration.
Input: Starting portfolio 𝑦𝑦𝑚𝑚, Desired Metric 𝑍𝑍𝑗𝑗 𝑖𝑖
44 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Incremental Value Report Algorithm
Input: Starting portfolio 𝑦𝑦𝑚𝑚, Desired Metric 𝑍𝑍𝑗𝑗 𝑖𝑖
Output: List of changes in value 𝛾𝛾𝑖𝑖𝑇𝑇for all remaining units in the portfolio when portfolio unit 𝑖𝑖 is dropped and the overall incremental value of the unit 𝜁𝜁𝑖𝑖.
45 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Incremental Value Report Algorithm
Input: Starting portfolio 𝑦𝑦𝑚𝑚, Desired Metric 𝑍𝑍𝑗𝑗 𝑖𝑖
Initialize: 𝜁𝜁𝑖𝑖 ← ∑ ∑ 𝑍𝑍𝑗𝑗(𝑖𝑖)𝑥𝑥𝑗𝑗 𝑖𝑖
𝑚𝑚𝐷𝐷(𝑖𝑖)𝑗𝑗=1
𝑁𝑁𝑖𝑖=1
For each i ∈ {𝑦𝑦𝑚𝑚 = 1}
Output: List of changes in value 𝛾𝛾𝑖𝑖𝑇𝑇for all remaining units in the portfolio when portfolio unit 𝑖𝑖 is dropped and the overall incremental value of the unit 𝜁𝜁𝑖𝑖.
End of list?
Record Solution, update 𝛾𝛾𝑇𝑇 , 𝜁𝜁𝑖𝑖
46 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Incremental Value Report Algorithm
1. 𝜁𝜁𝑖𝑖 = 𝜁𝜁𝑖𝑖 − ∑ ∑ 𝑍𝑍𝑗𝑗(𝑖𝑖)𝑥𝑥𝑗𝑗(𝑖𝑖)𝐷𝐷(𝑖𝑖)𝑗𝑗=1
𝑁𝑁𝑖𝑖=1
2. L1: for 𝒕𝒕𝜖𝜖{1,2, … ,𝑁𝑁} do 3. 𝛾𝛾𝑖𝑖𝑡𝑡 ← 0 4. L2: for 𝑢𝑢𝜖𝜖{1,2, …𝐷𝐷 𝑡𝑡 } do 5. 𝛾𝛾𝑖𝑖𝑡𝑡 = 𝛾𝛾𝑖𝑖𝑡𝑡 + ∑ 𝑍𝑍𝑗𝑗(𝑖𝑖)𝑥𝑥𝑗𝑗 𝑖𝑖
𝑚𝑚𝐷𝐷(𝑖𝑖)𝑗𝑗=1
6. 𝛾𝛾𝑖𝑖𝑡𝑡 = 𝛾𝛾𝑖𝑖𝑡𝑡 − ∑ 𝑍𝑍𝑗𝑗(𝑖𝑖)𝑥𝑥𝑗𝑗 𝑖𝑖𝐷𝐷(𝑖𝑖)𝑗𝑗=1
7. end L2:for 8. end L1:for
Solve M4: Minimize∑ 𝑦𝑦𝑖𝑖𝑖𝑖={𝑖𝑖:𝑦𝑦𝑖𝑖𝑎𝑎=1}
Subject to:
∑ 𝑥𝑥𝑗𝑗 𝑖𝑖 ≤ 𝑦𝑦𝑖𝑖 𝐷𝐷(𝑖𝑖)𝑗𝑗=1 ∀ 𝑖𝑖 = 1, … ,𝑁𝑁
∑ 𝐶𝐶𝑗𝑗 𝑖𝑖 ,𝑘𝑘𝑁𝑁𝑘𝑘=1 𝑥𝑥𝑗𝑗 𝑖𝑖 − ∑ 𝐶𝐶𝑗𝑗 𝑖𝑖 ,𝑘𝑘
𝑁𝑁𝑘𝑘=1 𝑦𝑦𝑘𝑘 ≤ 0 ∀ 𝑗𝑗 𝑖𝑖 = 1, … ,𝐷𝐷 𝑖𝑖
∑ 𝐸𝐸𝑗𝑗 𝑖𝑖 ,𝑘𝑘𝑁𝑁𝑘𝑘=1 𝑥𝑥𝑗𝑗 𝑖𝑖 + ∑ 𝐸𝐸𝑗𝑗 𝑖𝑖 ,𝑘𝑘
𝑁𝑁𝑘𝑘=1 𝑦𝑦𝑘𝑘 ≤ ∑ 𝐸𝐸𝑗𝑗 𝑖𝑖 ,𝑘𝑘
𝑁𝑁𝑘𝑘=1 ∀ 𝑗𝑗 𝑖𝑖 = 1, … ,𝐷𝐷 𝑖𝑖
𝑦𝑦𝑖𝑖 = 0
Let 𝑦𝑦′ be the solution to M4. Yes
Set up model
47 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Drop Unit M -475
+532 Add Unit W
Drop Unit R Drop Unit Q
Drop Unit P -75 Drop Unit O
-78 Drop Unit N
+119 Add Unit V +78 Add Unit U
+50 Add Unit T +41 Add Unit S
PORTFOLIO #3 TOTAL NPV + 69,429
DELTA +192
PORTFOLIO #4 TOTAL NPV + 69,621
Waterfall Report A detailed trajectory of moving from one portfolio to another.
Input: Starting portfolio 𝑦𝑦𝑚𝑚, Ending portfolio 𝑦𝑦𝑏𝑏, Desired Metric 𝑍𝑍𝑗𝑗(𝑖𝑖)
Waterfall Report Algorithm
Input: Starting portfolio 𝑦𝑦𝑚𝑚, Ending portfolio 𝑦𝑦𝑏𝑏, Desired Metric 𝑍𝑍𝑗𝑗(𝑖𝑖)
Output: Migration plan from A to B with a feasible portfolio at every step, i.e. An ordered set of Portfolios Y
49 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Waterfall Report Algorithm
Input: Starting portfolio 𝑦𝑦𝑚𝑚, Ending portfolio 𝑦𝑦𝑏𝑏, Desired Metric 𝑍𝑍𝑗𝑗(𝑖𝑖)
Determine 𝛽𝛽 the list of portfolio units such that 𝑦𝑦𝑖𝑖𝑚𝑚 = 0 and 𝑦𝑦𝑖𝑖𝑏𝑏 = 0. Determine 𝜈𝜈 the list of portfolio units such that 𝑦𝑦𝑖𝑖𝑚𝑚 = 1 and 𝑦𝑦𝑖𝑖𝑏𝑏 = 1. 𝑦𝑦𝑐𝑐 ← 𝑦𝑦𝑚𝑚 ,
Output: Migration plan from A to B with a feasible portfolio at every step, i.e. An ordered set of Portfolios Y
Does y’ exist?
End of list?
record the solution add to Y, update 𝑦𝑦𝑐𝑐with 𝑦𝑦𝑥 Update 𝛿𝛿 and 𝛼𝛼
Do nothing since no solution exists
Yes
Determine 𝛿𝛿 the list of portfolio units such that 𝑦𝑦𝑖𝑖𝑐𝑐 = 1 and 𝑦𝑦𝑖𝑖𝑏𝑏 = 0 Determine 𝛼𝛼 the list of portfolio units such that 𝑦𝑦𝑖𝑖𝑐𝑐 = 0 and 𝑦𝑦𝑖𝑖𝑏𝑏 = 1 𝑡𝑡 ← 1
50 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Waterfall Report Algorithm
for 𝑡𝑡 < 𝛿𝛿 + |𝛼𝛼| do Solve M5: Minimize: ∑ ∑ 𝑍𝑍𝑗𝑗(𝑖𝑖)𝑥𝑥𝑗𝑗(𝑖𝑖)
𝐷𝐷(𝑖𝑖)𝑗𝑗=1
𝑁𝑁𝑖𝑖=1
Subject to: (11), (12), (13) ∑ 𝑦𝑦𝑖𝑖𝑖𝑖𝜖𝜖𝛼𝛼 − ∑ 𝑦𝑦𝑖𝑖𝑐𝑐𝑖𝑖𝜖𝜖𝛼𝛼 + ∑ 𝑦𝑦𝑖𝑖𝑐𝑐𝑖𝑖𝜖𝜖𝑖𝑖 − ∑ 𝑦𝑦𝑖𝑖𝑖𝑖𝜖𝜖𝑖𝑖 = 𝑡𝑡 ∑ 𝑦𝑦𝑖𝑖𝑖𝑖𝜖𝜖𝛽𝛽 = 0 ∑ 𝑦𝑦𝑖𝑖𝑖𝑖𝜖𝜖𝑖𝑖 = 1 Let 𝑦𝑦𝑥 be the solution to M5
Yes
Solve
No
Port
folio
NPV
Portfolio Spending
Port
folio
NPV
Portfolio Spending
51 CPMS Daniel H. Wagner Prize Competition © Intel Corp
REAL TIME FORCE IN/OUT IMPACTS
-40m
-15m
Port
folio
NPV
Portfolio Spending
Port
folio
NPV
Portfolio Spending
52 CPMS Daniel H. Wagner Prize Competition © Intel Corp
REAL TIME FORCE IN/OUT IMPACTS
+4m -4m
Constraint Cuts
Constraint Cuts
Obj
ectiv
e Cu
ts
Headcount
Volume
Port
folio
NPV
Portfolio Spending
Port
folio
NPV
Portfolio Spending
53 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Reve
nue
REAL TIME FORCE IN/OUT IMPACTS
-40m
-15m
+4m -4m
THE BUSINESS PROCESS Analytics Intuition
using “Elimination by Aspects”
54 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Essential to: Allow the intuition to inform the analytics.
Difficult Because: Challenging to align qualitative and strategic objectives with quantitative objectives.
Breakthrough Math: “Elimination by Aspects” allows the combination of many objectives to narrow the generated feasible set to a few desirable portfolios. Supports decision maker in feeling satisfied with their final choice of a portfolio.
Using the Budget Aspect
55 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Check Portfolios wrt Budget Portfolio spend within
+/- 10% of budget
56 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Efficient Frontier
of non-dominated
portfolios
Budget
Strong portfolios within +/- 10% of the budget …
… checked over (multiple) year
budget.
Using the Budget Aspect
Port
folio
NPV
Portfolio Spend
57 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Check Products wrt Strategy Force selected products in / out and re-optimize
Check Portfolios wrt Budget Portfolio spend within
+/- 10% of budget
Using the Strategy Aspect
58 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Efficient Frontier
of non-dominated
portfolios
Budget
Port
folio
NPV
Portfolio Spend (next year)
Projects/Products that are in ALL of
the portfolios selected
Projects/Products that are in NONE of
the portfolios selected
Projects/Products that are in SOME of
the portfolios selected
Force OUT ??
Force IN ??
Drill Down, Decide !!
Using the Strategy Aspect
59 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Drop Unit D -1,844
PORTFOLIO #2 TOTAL NPV = 44,057
PORTFOLIO #1 TOTAL NPV = 51,326
Devalued Unit B -5,425
DELTA - 7,269
INCREMENTAL VALUE REPORT
Drop Unit M
-475
+532 Add Unit W
Drop Unit R Drop Unit Q
Drop Unit P -75 Drop Unit O
-78 Drop Unit N
+119 Add Unit V
+78 Add Unit U
+50 Add Unit T
+41 Add Unit S
PORTFOLIO #3 TOTAL NPV + 69,429
DELTA +192
PORTFOLIO #4 TOTAL NPV + 69,621
WATERFALL REPORT
Using the Strategy Aspect PRODUCT ROADMAP What products are
in (colored) and out (grey)
of the portfolio under inspection.
60 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Check Skill Set Changes Consider the suggested skill changes over time
Check Products wrt Strategy Force selected products in / out and re-optimize
Check Portfolios wrt Budget Portfolio spend within
+/- 10% of budget
Using the Resources Aspect
61 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Using the Resources Aspect
TIME
NU
MBE
R RE
QU
IRED
SKILL SET
62 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Check Balance across Markets Consider budget allocation and return in all business segments
Check Skill Set Changes Consider the suggested skill changes over time
Check Products wrt Strategy Force selected products in / out and re-optimize
Check Portfolios wrt Budget Portfolio spend within
+/- 10% of budget
Using the Balance Aspect
63 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Branches as business segments (enterprise, workstations, high performance, cloud, etc.) Leaves as … Projects sized as dollar budget Projects sized as headcount Products sized by NPV Products sized by volume Colors as … Projects in flight, new projects, strategic projects (forced in) Investment Efficiency (prod. NPV/proj. SPEND) Year a product enters market
Using the Balance Aspect
64 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Senior Management Intuition Special knowledge from Sales & Marketing, Product Design, etc
Check Balance across Markets Consider budget allocation and return in all business segments
Check Skill Set Changes Consider the suggested skill changes over time
Check Products wrt Strategy Force selected products in / out and re-optimize
Check Portfolios wrt Budget Portfolio spend within
+/- 10% of budget
Using the Special Aspect
The “old process” = advocacy.
65 CPMS Daniel H. Wagner Prize Competition © Intel Corp
THE (SERVER) BUSINESS IMPACT
The “old process” = advocacy.
- Interviewed portfolio tool vendors and thought we needed stronger mathematics. - Interviewed companies using portfolio ideas and thought we needed higher quality data.
Launched … - Decision Analysis approach for data quality - Optimization approach for decision quality
66 CPMS Daniel H. Wagner Prize Competition © Intel Corp
THE (SERVER) BUSINESS IMPACT
The “old process” = advocacy.
- Interviewed portfolio tool vendors and thought we needed stronger mathematics. - Interviewed companies using portfolio ideas and thought we needed higher quality data.
Launched … - Decision Analysis approach for data quality - Optimization approach for decision quality
The “new process” = quality and transparency.
67 CPMS Daniel H. Wagner Prize Competition © Intel Corp
THE (SERVER) BUSINESS IMPACT
- 1st trial: partial SERVER business plan end 2011 - 2nd trial: full SERVER business plan mid 2012 - Process of choice (‘mission critical’) ever since
68 CPMS Daniel H. Wagner Prize Competition © Intel Corp
THE (SERVER) BUSINESS IMPACT
- 1st trial: partial SERVER business plan end 2011 - 2nd trial: full SERVER business plan mid 2012 - Process of choice (‘mission critical’) ever since
- Three dozen “analysts/mapping builders” for 2 weeks - Senior VP and staff (8) for 2 days - Approval with CEO/President/CFO for 2 hours
69 CPMS Daniel H. Wagner Prize Competition © Intel Corp
THE (SERVER) BUSINESS IMPACT
- 1st trial: partial SERVER business plan end 2011 - 2nd trial: full SERVER business plan mid 2012 - Process of choice (‘mission critical’) ever since
- Three dozen “analysts/mapping builders” for 2 weeks - Senior VP and staff (8) for 2 days - Approval with CEO/President/CFO for 2 hours
- Time and effort cut by more than half. - Value of portfolio up ~10% at constant budget. - Being adopted by other Intel business units.
70 CPMS Daniel H. Wagner Prize Competition © Intel Corp
THE (SERVER) BUSINESS IMPACT
The Project Portfolio Decision Support Suite is written in C# utilizing Microsoft Windows Presentation Foundation (WPF) and IBM’s Concert/CPLEX.
Running on an remote server with a quad-core Intel i7, 8 GB ram, and 100 GB disk … for problems with a few hundred projects and products … accessed through analysts’ laptops … ENUMERATION takes between 30 and 60 seconds VALUATION takes roughly 2 hrs. OPTIMIZATION takes between 30 and 60 seconds WHAT-IFs take between 5 and 30 seconds
71 CPMS Daniel H. Wagner Prize Competition © Intel Corp
IMPLEMENTATION
72 CPMS Daniel H. Wagner Prize Competition © Intel Corp
Project Portfolio Planning at Intel Corporation
CORPORATE VISION: This decade we will create
and extend computing technology to connect and enrich the lives of every person on earth.