Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Design Architecture and Strategy Carliss Y. Baldwin Harvard Business School Information, Technology and Management Seminar March 7, 2006 Slide 1 © Carliss Y. Baldwin 2006 Unmanageable Designs—What They Are and their Financial Consequences Carliss Y. Baldwin Harvard Business School Information, Technology and Management Seminar March 7, 2006 Slide 2 © Carliss Y. Baldwin 2006 Three Points to begin Large, complex, evolving designs – Are a fact of modern life – Need design architectures— » “Description of the entities in a system and their relationships” » Way of assigning work (Parnas) Designs create value – Value operates like a force in the economy – We fight to create it and to keep it—using strategy Design Architecture and Strategy – How can you create and capture value in a large, complex evolving set of designs? – Subject of this talk Slide 3 © Carliss Y. Baldwin 2006 In the economy, value acts like a force Value = money or the promise of money Consider the computer industry… Slide 4 © Carliss Y. Baldwin 2006 The changing structure of the computer industry Andy Grove described a vertical-to-horizontal transition in the computer industry: 1980-“Vertical Silos” 1995-“Modular Cluster” Slide 5 © Carliss Y. Baldwin 2006 Andy’s Movie Stack View in 1980 Services Systems Integration Top 10 Public Companies in US Computer Industry Area reflects Market Value in Constant US $ Slide 6 Applications Layer Middleware Layer Operating Systems Hardware IBM S P E R R Y D CVC U H E N P C S Y XRC S AMP Components TI Intel © Carliss Y. Baldwin 2006 Andy’s Movie Stack View in 1995 IBM S P E R R Y D CVC U H E N P C S Y XRC S AMP TI Services Systems Integration First Data EDS Oracle Intel Top 10 Public Companies in US Computer Industry Area reflects Market Value in Constant US $ Slide 7 I Applications LayerB MSFT Middleware LayerM Operating Systems Hardware: Printers Hardware: Servers Hardware: Routers Components CA HP IBM Cisco Intel Micron © Carliss Y. Baldwin 2006 Andy’s Movie—the Sequel Stack View in 2002 IBM S P E R R Y D CVC U H E N P C S Y XRC S AMP TI Intel Top 10 Public Companies in US Computer Industry Area reflects Market Value in Constant US $ Slide 8 Services Services First First Data Data ADP EDS Systems Integration Systems Integration I CA Applications LayerB MSFT Applications LayerIBM Middleware LayerM Operating Systems Middleware LayerPrinters Hardware: Hardware: Servers Operating SystemsRouters Hardware: Components Hardware: Printers HP Hardware: PCs Dell Hardware: Servers IBM Hardware: Routers Cisco Components Intel Oracle Oracle MSFT HP IBM Cisco Intel Micron TI © Carliss Y. Baldwin 2006 Turbulence in the Stack Departures from Top 10: Xerox (~ bankrupt) DEC (bought) Sperry (bought) Unisys (marginal) AMP (bought) Computervision (LBO) Sic Transit Gloria Mundi Slide 9 Arrivals to Top 10: Microsoft Cisco Oracle Dell ADP First Data … Sic Transit © Carliss Y. Baldwin 2006 Contrast to the Auto Industry 2003 Honda FORD Dana Tenneco Eaton Other Powertrain Aftermarket AUGAT (electronics) Interior TRW Top 10 Public Companies in US Auto Industry Area reflects Market Value in Constant US $ Slide 10 Toyota Nissan Honda Daimler Chrysler GM Ford VW Others Toyota General Motors Chrysler 1982 Magna Johnson Controls Eaton Other Multiple Electrical/Electronics Other Interior Other Misc. Other Chassis/Powertrain Value stayed in one layer! © Carliss Y. Baldwin 2006 Two patterns “Manageable” designs = auto industry “Unmanageable” designs = computer industry What makes computer designs so unmanageable? This was the question Kim Clark and I set out to answer in 1987. Slide 11 © Carliss Y. Baldwin 2006 After studying the history of computer designs and correlating their changes with value changes We concluded that modularity was part of the answer… Slide 12 © Carliss Y. Baldwin 2006 IBM System/360 IBM did not understand the option value it had created Did not increase its inhouse product R&D Result: Many engineers left – to join “plug-compatible peripheral” companies San Slide 13 Jose labs —> Silicon Valley © Carliss Y. Baldwin 2006 Modularity in computers, cont. Bell and Newell, Computer Structures (1971) – General principles of modular design for hardware – Basis of PDP-11 design—another ORMDA Thompson and Ritchie, Unix and C (1971-1973) – Modular design of operating system software (contra Brooks Law) Parnas (1972) abstract data structures, info hiding – Object-oriented programming, C++, Java Mead and Conway, Intro to VLSI Systems (1980) – Principles of modular design for large-scale chips Slide 14 © Carliss Y. Baldwin 2006 Modularity in computers (cont.) IBM PC (1983) – – – – – – Slide 15 DEC PDP-11 minimalist strategy (exclude and invite) + Intel 8088 chip + DOS system software + IBM manufacturing + Lotus 1-2-3 A modular design architecture with a mass market © Carliss Y. Baldwin 2006 Measuring modularity As scientists, we can visualize and measure modularity — after the fact DSMs, Design Hierarchies – Methods are tedious, non-automated Slide 16 © Carliss Y. Baldwin 2006 Comparison of different software systems with DSM tools Mozilla just after becoming open source Coord. Cost = 30,537,703 Change Cost = 17.35% Linux of similar size Coord. Cost = 15,814,993 Change Cost = 6.65% © Alan MacCormack, Johh Rusnak and Carliss Baldwin, 2006 Conway’s Law: Different organizations deliver different architectures Mozilla just after becoming open source One Firm, Tight-knit Team, RAD methods Coord. Cost = 30,537,703 Change Cost = 17.35% Linux of similar size Distributed Open Source Development Coord. Cost = 15,814,993 Change Cost = 6.65% © Alan MacCormack, Johh Rusnak and Carliss Baldwin, 2006 Refactoring for modularity Mozilla Before Redesign Change Cost = 17.35% Mozilla After Redesign Change Cost = 2.38% © Alan MacCormack, Johh Rusnak and Carliss Baldwin, 2006 But modular ≠ unmanageable There are modular design architectures that are very manageable – Toyota Production System (TPS) The other property of a design architecture => option potential Denoted s s /Option potential varies by system and by module Slide 20 © Carliss Y. Baldwin 2006 s /Option potential systems => High s Manageable systems => Low s Unmanageable System/360 had had low s /option potential, IBM would be like Toyota If It would be a different world… Slide 21 © Carliss Y. Baldwin 2006 An Option is The right but not the obligation to take an action – Action = Use a new design – If new is better than old, use new; – Otherwise, keep the old. Designs are Options! Slide 22 © Carliss Y. Baldwin 2006 s /Option potential is like dark matter in the universe Scientists can measure its effects but we can’t measure “it” “Wizards” can perceive s /option potential – But wizards don’t talk to scientists! we lack ways to measure s /option value scientifically Thus – It is a “research frontier” Slide 23 © Carliss Y. Baldwin 2006 Evidence of s /Option value Successive, improving versions are evidence of option potential being realized over time—after the fact Global Design Rules v.1 Version 1.0 Version 1.2 Version 1.5 Version 1.8 s = Low Slide 24 Medium Zero High © Carliss Y. Baldwin 2006 s /Option potential at work— Matlab programming contest Slide 25 © Carliss Y. Baldwin 2006 Sources of s /option value Physics— – Moore’s Law (dynamics of miniaturization) applies to MOSFET circuits and systems (Mead and Conway) – Power and heat systems vs. logic systems (Dan Whitney) User innovation – Users’ discovery of their own needs – “Killer apps” Architecture – Experimenting with different relationships among entities Slide 26 © Carliss Y. Baldwin 2006 All three are important, but let’s talk about architecture “Pfister’s anomaly” Named for: Gregory Pfister— IBM senior R&D staff in Austin, TX; Wrote In Search of Clusters Slide 27 © Carliss Y. Baldwin 2006 Pfister’s Anomaly 100,000 Key Pfister Spec CPU 95 tpmC and Spec CInt 10,000 Architectural experimentation 1,000 100 Component experimentation 10 1 Dec-91 Dec-93 Dec-95 Dec-97 Dec-99 System Availability Date Slide 28 © Carliss Y. Baldwin 2006 Dec-01 More data—same result 10,000,000 1,000,000 Key Pfister TPC-C v.3 TPC-C v.5 Spec CPU 95 Spec CPU 2000 tpmC and Spec CInt 100,000 Architectural experimentation 10,000 1,000 100 Component experimentation 10 1 Dec-91 Dec-93 Dec-95 Dec-97 Dec-99 Dec-01 Dec-03 System Availability Date Slide 29 © Carliss Y. Baldwin 2006 Dec-05 The Pfister Effect—translated Recombining modules in new ways has more option value than changing the individual module designs Not what we model in Design Rules… – We say: A stable, modular architecture supports component experimentation … – which creates value. Pfister Effect: – A recombinant, modular architecture with stable interfaces supports architectural experimentation… – and creates even more value! “Recombinant Slide 30 architecting” is a process module! © Carliss Y. Baldwin 2006 Recapping the argument Designs create value – Value operates like a force in the economy – Changes the structure of industries Designs have architectures – Modularity and s /Option value are the key economic properties of a design architecture – Modularity + High s /Option value => Unmanageable » Why computers are not like autos » Why IBM is not Toyota Next— How can you capture value in a modular, highs design architecture? Slide 31 © Carliss Y. Baldwin 2006 How can you capture value? Architectural/Technological question: – Where/when does modularization stop? Business/Strategic question: – How do you make money in a modular, high-s world? Slide 32 © Carliss Y. Baldwin 2006 Where does modularization stop? Strojwas (2005) Semiconductor Industry Top 10 Firms: 1994 and 2004 Slide 33 © Carliss Y. Baldwin 2006 How do you make money in a high-s world? Old paradigm – “Plunge in” – “Get lucky” – “Watch out for Microsoft” – “Get bought by HP” Slide 34 © Carliss Y. Baldwin 2006 Our research strategy—Look for Stable patterns of behavior involving several actors operating within a consistent framework of ex ante incentives and ex post rewards ==> Equilibria of linked games with selfconfirming beliefs (Game theory) Slide 35 © Carliss Y. Baldwin 2006 How a “stable pattern” works Anticipation of $$$ (visions of IPOs) Lots of investment Lots of design searches Best designs “win” Fast design evolution => innovation Lots of real $$$ (an actual IPO) “Rational expectations equilibrium” Slide 36 © Carliss Y. Baldwin 2006 Three Stable Patterns— (another research frontier) “Blind” Competition – PCs in the early 1980s High ROIC on a Small Footprint – Sun vs. Apollo – Dell vs. Compaq (and HP and …) Lead Firm Competition – Monopoly—MSFT – Mergers & Acquisitions—Cisco, Intel … Slide 37 © Carliss Y. Baldwin 2006 “Blind” Competition All Zeros (1, 0, 0) x A B x Ventures do best Fighters do best Zeros do best ESS (1/8, 3/8, 4/8) ) 0.0 All Ventures (1,0.1 0, 0) Slide 38 0.2 0.3 0.4 0.5 0.6 0.7 0.8 1.0(1, 0, 0) All0.9 Fighters © Carliss Y. Baldwin 2006 “Blind” Competition Slide 39 © Carliss Y. Baldwin 2006 “Footprint” Competition—Apollo 1 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 2 3 O O x x x x x x x x x 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 Processor chip—CPU Outsourced—Motorola 680x0 Floating Point Accelerator Outsourced O Memory chips DRAMs, ROM Outsourced—Commodity O Storage—Disk Drives Outsourced O Storage—Tape Drive Outsourced O Printed circuit boards Outsourced—Commodity O Display Monitor Outsourced O Keyboard, Cabinet, Fans Outsourced x x x x x x x x x x x x x Inhouse Design x x x x x x x x x x x x x OS Network x x x x x x x x x x x x Key: x= transfer of material or information from column task to row task; T= transaction: sale of good by column owner to row owner; O= outsourced task blocks; D= downstream or complementary task blocks; highly interdependent task blocks with many iterations and high within-block mundane transaction costs; Apollo's footprint (tasks performed inhouse). Aegis proprietary Operating System DOMAIN proprietary Network Architecture x Hardware Design DN series = 3-4 boards incl. IO and Display controllers, Power supply Purchase Components Hardware T T T T T T T T T T T T Keeps Design Control Slide 40 T T T T Component Test Kits Board stuff and Solder Test Boards Board Assembly System Assembly System Test Quality Assurance Consolidate and Ship x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x Inhouse Manufacturing x x x x x x x D Many Software Applications D D x x x x x x x x x x x x T Many OEMs T T D D D © Carliss Y. Baldwin 2006 Then Sun came along… Apollo Computer Aegis proprietary Operating System Inhouse Design OS DOMAIN proprietary Network Architecture Network Hardware Design DN series = 3-4 boards incl. IO and Display controllers, Power supply Purchase Components Hardware Component Test Kits Board stuff and Solder Test Boards Board Assembly System Assembly System Test Quality Assurance Consolidate and Ship x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x Inhouse Manufacturing x x x x x And did even less! How? Slide 41 x x x x x x x x x x x x x x x x x x x T T T T Component Test Kits Board stuff and Solder Test Boards Board Assembly System Assembly System Test Quality Assurance Consolidate and Ship Customize Unix Proprietary MMU Internal bus Single Board Layout Purchase Components Inhouse Design x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x O T x Manufacturing O T x x x x x © Carliss Y. Baldwin 2006 Then Sun came along… Apollo Computer Aegis proprietary Operating System Inhouse Design OS Hardware Design DN series = 3-4 boards incl. IO and Display controllers, Power supply Purchase Components Hardware Component Test Kits Board stuff and Solder Test Boards Board Assembly System Assembly System Test Quality Assurance Consolidate and Ship Design Architecture for high performance with a small footprint DOMAIN proprietary Network Architecture Network x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x Public Standards for outsourcing Inhouse Manufacturing x x x x x x x x x x And did even less! How? Slide 42 x x x x x x x x x x x x x x T T T T Component Test Kits Board stuff and Solder Test Boards Board Assembly System Assembly System Test Quality Assurance Consolidate and Ship Customize Unix Proprietary MMU Internal bus Single Board Layout Purchase Components Inhouse Design x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x x O T x Manufacturing O T x x x x x © Carliss Y. Baldwin 2006 Result: ROIC advantage to Sun Average over 16 Quarters: Apollo Computer Sun Microsystems 29% 24% 57% 15% 13% 31% Low is good Low is good Low is good Profitability Net Income/Sales 0% 6% High is good ROIC ROIC (excl Cash, Annualized) 2% 20% High is good Invested Capital Ratios (Annualized) Net Working Capital/ Sales (%) Ending Net PPE / Sales (%) Invested Capital/Sales (%) Sun used its ROIC advantage to drive Apollo out of the market Apollo was acquired by HP in 1989 Slide 43 © Carliss Y. Baldwin 2006 Compaq vs. Dell Dell did to Compaq what Sun did to Apollo … Dell created an equally good machine, and Used design architecture to reduce its footprint in production, logistics and distribution costs – Negative Net Working Capital – Direct sales, no dealers Result Slide 44 = Higher ROIC © Carliss Y. Baldwin 2006 Higher ROIC always wins! 1997 Invested Capital Ratios (Annualized) Net Working Capital/ Sales (%) Ending Net PPE / Sales (%) Invested Capital/Sales (%) Profitability Net Income/Sales ROIC ROIC (excl Cash, Annualized) Compaq Computer Dell Computer -2% 8% 8% -5% 3% -2% Low is good Low is good Low is good 8% 7% High is good 101% -287% !!! Dell started cutting prices; Compaq struggled, but in the end had to exit. Compaq was acquired by HP in 2002 Slide 45 © Carliss Y. Baldwin 2006 Competition and Beliefs “Blind” competitors – don’t know others exist “Footprint” competitors – Don’t expect to influence others—just compete “Lead firms” – Must influence the beliefs of their competitors – FUD — “Fear, uncertainty and doubt” – Others cannot be blind! Slide 46 © Carliss Y. Baldwin 2006 “Lead Firm” Competition Monopolist needs to deter all potential entrants with threats of price war – Very fragile equilibrium – Potentially expensive to create “enough” FUD M&A Lead Firm does not try to deter all entry in the design space – Expects to buy most successful entrants ex post – More robust equilibrium – Maybe more advantageous, when you count the cost of FUD Slide 47 © Carliss Y. Baldwin 2006 How do you make money in a high-s world? A new paradigm… – Expect a cluster – Use M&A to be the “lead firm” in some slice(s) of the stack – Use design architecture to reduce your “footprint” in each slice => high ROIC – Use open source methods to clone your complementor’s products => price and innovation discipline Slide 48 © Carliss Y. Baldwin 2006 Remember Designs create value – Value operates like a force in the economy – Changes the structure of industries Designs have architectures – Modularity and s /Option value are the key economic properties of a design architecture – Modularity + High s /Option value => Unmanageable » Why computers are not like autos » Why IBM is not Toyota Capturing value in a modular, high-s design architecture requires new competitive strategies Slide 49 © Carliss Y. Baldwin 2006 Thank you! Slide 50 © Carliss Y. Baldwin 2006 Modularity in computers— IBM System/360 First modular computer design architecture (1962-1967) – Proof of concept in hardware and application software – Proof of option value in market response and product line evolution – System software was NOT modularizable » Fred Brooks, “The Mythical Man Month” » Limits of modularity Slide 51 © Carliss Y. Baldwin 2006 Strategically—IBM wanted to be the sole source of all of System/360’s Modules 1 1 2 3 4 5 6 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 SLT architecture and standard circuits Erich Bloch - August 1961 New Processor Line Architectural Ground Rules SPREAD Task Group - 12/28/61 New Processor Line control, product and programming standards Corporate Processor Control Group (CPC) - 4/1/62 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 SLT Transistors SLT Modules SLT Cards SLT Boards and Automatic Wiring Processor 1 - Endicott, New York Processor 2 - Hursley, England Processor 3 - Poughkeepsie, New York Processor 4 - Poughkeepsie, New York Processor 5 - Poughkeepsie, New York Main memories, Corporate Memory Group (1) Internal memories, CMG Read-only memories for control, CMG "Binary-addressed" Random Access Files Corporate File Group (2) Tape devices running at 5000+ char/sec Corporate Tape Group (3) Time-multiplex system for switching I/O devices DSD Technical Development Group Techniques to measure processor performance, system throughput and software efficiency, Group Staff A unified Input/output Control Structure (IOCS) System Software for Configuration I (4) System Software for Configuration II (4) System Software for Configuration III (4) FORTRAN and COBOL compilers A unified programming language 33 34 35 Announcement and Marketing Production, Testing and Integration Shipment, Delivery and Installation Slide 52 © Carliss Y. Baldwin 2006 35 By 1980, 100s of firms made S/360 “plug-compatible” components Code 3570 3571 3572 3575 3576 3577 3670 3672 3674 3678 7370 7371 7372 7373 7374 7377 Category Definition Computer and Office Equipment Electronic Computers Computer Storage Devices Computer Terminals Computer Communication Equipment Computer Peripheral Devices, n.e.c. Electronic Components and Accessories Printed Circuit Boards Semiconductors and Related Devices Electronic Connectors Computer Programming, Data Processing, and Other Services Computer Programming Services Prepackaged Software Computer Integrated Systems Design Computer Processing, Data Preparation and Processing Computer Leasing 1960 1970 1980 5 1 1 2 1 3 11 2 8 5 2 8 6 5 1 5 7 19 4 15 9 29 36 23 10 12 11 39 10 16 1 0 0 1 9 2 7 3 26 * 12 * 13 * 16 0 0 41 5 10 108 29 * 7 * 298 * Firms in these subindustries make modules of larger computer systems. Firms making modules = 34 95 Percent of total = 83% 88% Slide 53 * * * * * * * * 244 82% © Carliss Y. Baldwin 2006