Integrated Supply Chain Operations Research at IBM Corporation: Integrated Supply Chain Perspective Dr. Brian Thomas Eck Director of Strategy, IT & Business Transformation, IBM International Holdings, Inc. -- Singapore Branch Integrated Supply Chain (ISC) IBM Research, Delhi India | April 2006 © 2006 IBM Corporation Integrated Supply Chain Today’s Discussion Introduction: Supply Chain Management & IBM’s Integrated Supply Chain Enablers of Successful OR Application: Demand and Support for OR Embedding in Operations Differentiated Roles Examples of OR at IBM: Simulation / Inventory Optimization Example Available to Sell: Resource Allocation e-Auctions Analysis Summary Questions 2 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Today’s Discussion Introduction: Supply Chain Management & IBM’s Integrated Supply Chain Enablers of Successful OR Application: Demand and Support for OR Embedding in Operations Differentiated Roles Examples of OR at IBM: Simulation / Inventory Optimization Example Available to Sell: Resource Allocation e-Auctions Analysis Summary Questions 3 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain What is Supply Chain Management? Customer Partners Supply Chain Suppliers Contract Mfg. Channels 4 Flows: Product, Information, Work, Cash 3PL’s Retail Direct Supply Chain Managers Exchanges Design Plan Source Make Deliver B2B Sell Resellers Return What is ISC in IBM? 19,000 employees in 61 countries (200+ with Ph.D.s) Responsible for USD$ 40 Billion of IBM cost and expense Shipping 1 Billion kilograms of product annually Part of the larger Integrated Operations team Transformation credited with savings for IBM in excess of USD $20 Billion over first three years (2002-2004) 4 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Today’s Discussion Introduction: Supply Chain Management & IBM’s Integrated Supply Chain Enablers of Successful OR Application: Demand and Support for OR Embedding in Operations Differentiated Roles Examples of OR at IBM: Simulation / Inventory Optimization Example Available to Sell: Resource Allocation e-Auctions Analysis Summary Questions 5 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Enabling OR Application within Industry Demand for OR Application Competitive Pressures Maturity in Organizational Improvement Awareness of Methods, Skill Base of Employees Business Improvement Process and Structure Support for OR Application Virtual Community Application Domain Support Center of Excellence Support Effectiveness of OR Application Business insights and OR expertise Embedding in Business Processes 6 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain OR Community and Examples Informal Network Center Center of of Excellence Excellence (IBM (IBM Research) Research) Block Scheduling for Classrooms and Instructors •Improve utilization and decrease costs •Penalty function, MIP (using OSL, C++) •Used through 7 cycles (over 4+ years) •Model size: 62,010 columns 90,002 rows (273,392 nonzeros) •Well-accepted, will spread to Europe MD Network Design Integrated Supply Chain Advanced Planning Systems Supply/Demand Process Network Optimization Business Units PSG 7 TG SG Operations Research at IBM •Logic packaging vendor offered alternate locations •Spreadsheet model, "What's Best" MIP •$650K savings identified •Extensions to full logic network and other products SSD Sourcing •Manufacturing Strategy group •Assigning flows from multiple manufacturing locations to multiple customer sites •LP and MIP © 2006 IBM Corporation Integrated Supply Chain Embedding tools within processes for decision support Investment Matrix, FEAT •Corporate-Wide Supply/Demand Process •Interlock: Supply Support Decision •Unbiased Forecast •Alternative Perspectives •Supply Support Decision •Risk (lost sales versus inventory) •Maximize Expected PTI Design for Logistics •Enable Designers At Decision Time •Consider Total Product Cost •Heuristics and Model •Inventory Targeting in an Assemble-To-Order Environment Simulation to Model S390 Supply Chain •Express Targets as DOS by Commodity •Weekly Review of Actuals versus Targets 8 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Most OR Practice Successes in IBM Leveraged Multiple Roles Very deep Academia IBM Research ISC Technical Leaders Depth in OR Thinking ISC Practitioners & Executives Shallow Little to None General (broadly familiar) Deep and Broad Literacy in IBM’s business 9 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain S390 Inventory Analysis 70% of business transacted on IBM servers Cyclic demand production challenges Inventory management: High $ parts by commodity Ireland . Montpellier Fabricated Parts North America 46.5% Volume 95% European Suppliers (Less MCM) Poughkeepsie Japan CDCs 53.5% Volume Europe Middle East Africa 95% NA suppliers CDCs Asia Pacific Fujisawa Brazil 20 CDCs Sumare Latin America 10 Operations Research at IBM Key Strategy: Fab/Fulfillment Simulation modeling to explore behavior of BTP/CTO supply chain © 2006 IBM Corporation Integrated Supply Chain S390 Simulation Project: Inventory Study Fabrication Fulfillment Center MCMs Feature1 power . . . BOX (MTM) memory FeatureK When managing a measurement, we need to know where we expect it to be... 11 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Objectives 1. Determine Days of Supply (DOS) levels/targets for high dollar parts, for the "as is" CMOS supply chain. 2. Assess how improvements to feature ratio forecasting accuracy would impact CMOS inventory turns. 3. Establish the impact on required CMOS inventory of fab/fulfillment versus consumptive pull replenishment. 12 Operations Research at IBM Steps •Confirm objectives •Build model •Gather data •Cleanse data •Validate model •Test hypotheses •Draw conclusions •Analytical •Business implications •Present, convince, implement © 2006 IBM Corporation Integrated Supply Chain Replenishment Lead Times: Fulfillment Center BOMs: EMLS Extract Fed by SAP EMLS Extract Found incorrect (empty system LTs) Debby Carelli Denny Slocum Identifying Feature P/Ns Larry Fox Fab BOMs: Pull vs Non-pull In Practice Nick Kulick (pwr) Mike O'Dowd (DASD) Sue Cozalino Ron Shields Danielle Fields Dave Pearson (IE) Debby Carelli Don Gunvalsen Jeff Benedict Testing Yields/Usage -Gisela Hetherington (MCMs) Dave Pearson (general) Mae Ling Chen (non MCM Logic) Kai Wong (Memory) Winston Ralph/Mark Coq (power) Transportation Lead Times: Jeff Schmitt 13 Operations Research at IBM Jim Curatolo Brian Kuhn Wendy Sell Roger Tsai/Pete Weber Identifying FC P/Ns Testing Lead Times -- SAP Forecasts Box/MES: Don Gunvalsen Jeff Benedict Monthly Forecasts Larry Fox's spreadsheets SCE files (20-day process) Monthly Actuals COATS data extracts (custom SQL) Serviceability: Bethesda DB CAD=CRAD for CRAD within 3 weeks (80%) 100% otherwise (custom SQL) © 2006 IBM Corporation Integrated Supply Chain Validation against historical actuals, builds confidence in the model CMOS AVG High Dollar Inventory 60 50 PWR_SUPP PWR_MECH MEMORY LOGIC 40 30 20 10 Validation of Simulation Model 0 14 LOGIC LOGIC MEMORY MEMORY PWR_MECH PWR_MECH PWR_SUPP PWR_SUPP 97 97 % % 96 96 % % 86 86 % % 94 94 % % OVERALL OVERALL 95 95 % % Operations Research at IBM Average Inventory of High Dollar IMPACT Parts 11/16/98 11/09/98 11/02/98 May to October Average 06/23/98 06/02/98 05/18/98 05/04/98 04/20/98 Date CMOS: May through October 1998 $70 $60 $50 Millions Average Inventory Millions $ 70 PWR_SUPP PWR_MECH MEMORY LOGIC $40 $30 $20 $10 $0 Actuals Simulation © 2006 IBM Corporation Integrated Supply Chain Multiple Replications, Demand Mix and Variation To Test Effect of Fab/Fulfillment (BTP/CTO) LOGIC DOS for pDOS2QA Three Replications 50 Days of Supply 45 40 35 30 25 20 15 15 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Patterns emerged for each commodity LOGIC DOS PWR_MECH DOS 50 60 45 50 40 40 DOS DOS 35 30 30 25 20 20 15 10 10 1 3 5 7 9 11 13 1 2 3 4 5 6 7 8 9 10 Week of Quarter Week of Quarter MEMORY DOS PWR_SUPP DOS 11 12 13 11 12 13 45 60 40 50 35 DOS DOS 40 30 30 25 20 20 15 10 10 1 3 5 7 Week of Quarter 16 Operations Research at IBM 9 11 13 1 2 3 4 5 6 7 8 9 10 Week of Quarter © 2006 IBM Corporation Integrated Supply Chain LOGIC DOS 50 45 40 DOS 35 30 25 20 15 10 1 5 9 13 Week of Quarter 17 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain SQC charts are applied to the residuals to detect when to act 18 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Additional Observations Inventory Cost of Fab/Fulfillment $70 No No Capacity Capacity Constraints Constraints Quantifies Quantifies cost cost of of strategy/cost strategy/cost of of skew skew 29% 29% more more expensive expensive overall overall 74% 74% savings savings possible possible for for PWR_MECH PWR_MECH Lead Time Reduction: Consumptive Pull $70 PWR_SUPP $60 MEMORY LOGIC Inventory Millions $50 13% 11% 6% 6% 6% Logic MEM Mech Supp 30% $30 $40 $30 $20 $10 $0 As Is versus Consumptive Pull PWR_SUPP MEMORY LOGIC PWR_MECH Sensitivity Sensitivity Analysis Analysis Using Using consumptive consumptive pull pull model model (max (max savings) savings) Using Using fab/fulfillment fab/fulfillment model model (much (much less less sensitive) sensitive) PWR_MECH $40 $50 Inventory Millions Model Model run run with with consumptive consumptive pull, pull, optimized optimized reorder reorder points points $60 $20 $10 $0 Base 19 All MCM Operations Research at IBM Actuals © 2006 IBM Corporation Integrated Supply Chain Today’s Discussion Introduction: Supply Chain Management & IBM’s Integrated Supply Chain Enablers of Successful OR Application: Demand and Support for OR Embedding in Operations Differentiated Roles Examples of OR at IBM: Simulation / Inventory Optimization Example Available to Sell: Resource Allocation e-Auctions Analysis Summary Questions 20 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Available-to-Sell (AtS) •Determining how excess parts inventory can be positioned with marketing / sales as finished goods (saleable) product, to condition demand and consume the excess •Optimization aspect appears as a straightforward Linear Programming application •Production Planning LP tool already developed in IBM Research (WIT/SCE) Enterprise implosion problem: 380K resources, 185K operations, 84K demands, 800K flows, 52 periods (and this doesn't include capacity) LP formulation: 57 M variables, 24 M constraints, 118 M nonzeros 21 Sales: "What do we have in excess?" Planning Items (MTMs, Upgrades, MES loose piece) Features MFI/FFBM Manufacturing: "We have excess parts inventory." Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Data Issues Dominate Industrial Problem Solving ETIS relates planning items to p/n via history (ratios) 4. Inadequate history causes artificial 'zero' ETIS ratios 5. Which parts are called out by which features is order-dependent 2. EC causing expired effectivity dates (bill present but no demand on parts) 1.B) Card bills missing (outsourced) 1.A) Excess at a component level unknown to Manufacturing 3. Bills missing entirely for parts in excess 6.“Penny parts” 6000 of these 7.C-source SG ATS/ 03 (consigned) rev7/22/01 22 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Feature Translation: creating pseudo bills of material and appending to existing structures fc 1234 for model ABC Parts unique to ABC fc 1234 for all models New bill structures to be added... ...connect to existing bill structures 23 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Device Code to Bill Structure Example EXAMPLE (d/c 0014) 9406;170;Q01;0014;00075G2720;***;(9401.R1)ESP -DOCS 9406;500;Q01;0014;00017G0071;***;(9406.R1)PRE-GA PUBS 9406;510;Q01;0014;00017G0071;***;(9406.R1)PRE-GA PUBS 9406;530;Q01;0014;00017G0071;***;(9406.R1)PRE-GA PUBS 9406;50S;Q01;0014;00017G0071;***;(9406.R1)PRE-GA PUBS 9406;53S;Q01;0014;00017G0071;***;(9406.R1)PRE-GA PUBS 9406;***;Q01;0014;00046G0063;***;(MILL.R1)PRE-GA PUBS 9406;***;Q01;0014;00017G0071;***;(CONH.R1)PRE-GA PUBS 0014_53S 0014_50S 0014_530 0014_170 0014_510 This is expressed as follows in the SCE format: 0014_500 On model 170, 0014 requires only 1 per of 75G2720 On models 500, 510, 530, 50S, and 53S, only 1 per of part 17G0071 is required. 0014_ML6 On all models other than those listed, device code 0014 requires either one unit per of 46G0063 (for models S1*,S20,60*,62*, and 720) or one unit per of 17G0071 (for models 840 and SB3).* 0014_M10 "0014_9406ML6";"0000017G0071";1 "0014_9406M10";"0000046G0063";1 "0014_9406170";"0000075G2720";1 "0014_9406500";"0000017G0071";1 "0014_9406510";"0000017G0071";1 "0014_9406530";"0000017G0071";1 "0014_940650S";"0000017G0071";1 "0014_940653S";"0000017G0071";1 75G2720 17G0071 46G0063 *using rel3.mfc (bld level) MTMODCNV.R file 24 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Additional Observations •Objective Function •Dependent on sales price (to maximize profit) but prices unavailable •Use a ‘scaling factor’ k and maximize k *(excess consumed) – sum (cost of additional purchases) k small k large Minimize add’l payment Maximize using up excess •Business process design/implementation equally key to success •Results / Timeline • Jan 2002: Identified problem, data challenges, modeling approach • April 2002: programmed prototype; simple features only • June 2002: production version including simple+1, simple+2 f/c parser • Patent filing late 2002 • In 2002, component inventory moved = USD$ 72 million • In 2003, component inventory moved = USD$ 40 million •“Hardened” and offered commercially to clients (first sale 2005) 25 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Today’s Discussion Introduction: Supply Chain Management & IBM’s Integrated Supply Chain Enablers of Successful OR Application: Demand and Support for OR Embedding in Operations Differentiated Roles Examples of OR at IBM: Simulation / Inventory Optimization Example Available to Sell: Resource Allocation e-Auctions Analysis Summary Questions 26 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain e-Auctions to Exploit price/quantity Relationships Fixed-Price versus Auctions Selling quantity quantity Q0 p0 price price Reason Not to Auction New Products quantity quantity forecasting demand (BAU) Q0 Q0 forecasting price (auction) p0 price p0 price Auctioning Complements BAU 27 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Investigation of Product Differences and Value of Auctions Some Products are a Better Fit for Auctioning Key Driver: How unique is the purchase across customers? High unit volumes DRAMs HDDs Custom Logic (ASICs) PSG Amenable to Auction PSD Low unit volumes SSD AS/400 RS6000 S390 Common function, product across many customers Unique function, product for each customer Key Question: Is it more efficient to have inventory and idle factory capacity, or to sell the product at whatever price the market will bear? 28 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Auctioning as an Additional Channel Three Key Parameters Drive the Dynamics Percentage of total revenue targeted through the auction channel Percentage of current channel demand cannibalized by auction sales Percentage auction price effectiveness These Key Inputs Are Unknown Able to be estimated from piloting Cannibalization and auction price effectiveness are outcomes % revenue targeted translates through the other parameters into resultant auction revenue Brand- (product-) specific Approach Estimate reasonable average values for these parameters, and then test sensitivity across a range of values by randomly simulating different combinations. 10% 5% 15% Targeted Auction Revenue Auction Revenue = targeted revenue times price effectiveness 82.5% 70% Cannibalized Revenue 95% Price Effectiveness 57.5% 25% Total Revenue 29 Operations Research at IBM 90% % Cannibalization © 2006 IBM Corporation Integrated Supply Chain e-Auctions Appear to be an Attractive Channel Although auction price effectiveness is less than 100% (70% to 95%), profit margins improve by using free capacity (leveraging fixed cost across more revenue) and from selling excess inventory. The incremental profits and revenues are fairly robust across a wide range of cannibalization and auction prices: Net change in profit 30 million $ million $ Revenue potential Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Auctioning versus "Working Off" in Supply Chain Decisions Recognize an "Excess Supply" Situation Determine Whether to Auction or Work Off Determine How to Auction After Building Product Timing After Missing Forecast Relative to Building the Box Relative to Calling the Missed Demand Factors: START HERE Before Building Product ...OR HERE Before Missing Forecast Component part leadtime k periods Price takedown Cost of inventory Pk c Cost of production Selling expense: usual channel(s) Selling expense: auctions Waiting penalty s a wk Price received through auctioning product (random variable) Pa Auction if we can get at least Pa ,so that the margin is at least what we could get by working it off through the supply chain Pa - c 31 - a m Pk - c - Operations Research at IBM - s - wk © 2006 IBM Corporation Integrated Supply Chain Recognize "Excess" -1 Decide on and/or Conduct Auctioning Declare "on-hand" and net out of demand 1 0 After Missing Forecast and Before Building Product 2 3 k k+1 Demand Plans booked (reforecast and netted on-hand) Requirements Passed to Suppliers Materials Received from Suppliers (leadtime=k periods) Auction at minimum opening bid of Pa = Pk - s + a - wk where wk = cki if I have already purchased the component inventory 32 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Dynamic Programming Approach to the Auctioning Decision Difference Equation Approach Sell through the auction, and sell through working off, with certain probabilities (< 1): Let Ci = maximum return given one unit (box) of excess supply in period i, then Ci = max (Pk+i - c - {max Pr(Pa m r){E(Pa | Pa m r) r c- - a } + Pr(Pa < r)(w1 + C i+1 ), } - s - wk+i )Pr(sell it in period i+k for Pk) + ( C i+k - wk+i ) Pr(don't sell it) and Clast = scrap value (for some well-defined period in the future) Then, solve for C0 Issues Auction (market) price distributions may be poorly understood Probabilities (of selling one item at price Pk in period i+k) unknown 33 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Today’s Discussion Introduction: Supply Chain Management & IBM’s Integrated Supply Chain Enablers of Successful OR Application: Demand and Support for OR Embedding in Operations Differentiated Roles Examples of OR at IBM: Simulation / Inventory Optimization Example Available to Sell: Resource Allocation e-Auctions Analysis Summary Questions 34 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Enabling OR Application within Industry Function embedded in S/W instantiation Very deep Academia “Wrapper” Concept: IBM Research OSL / WIT / SCE / AtS DES / BPMAT / AMT ISC Technical Leaders Depth in OR Thinking ISC Practitioners & Executives Shallow Little to None General (broadly familiar) Deep and Broad Literacy in IBM’s business 35 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain In Summary Supply chain management is a dynamic, exciting, growing application area Data management (gathering, cleansing, workarounds) is a critical success factor and often consumes most project resources Ingredients for success include: ► Readiness (maturity, awareness, skill base, burning platform) ► Support community ► Combined OR expertise with business insight – Differentiated roles helpful 36 Operations Research at IBM © 2006 IBM Corporation Integrated Supply Chain Thank you! Brian T. Eck BrianEck@sg.ibm.com or Drbteck@yahoo.com.sg 37 Operations Research at IBM © 2006 IBM Corporation