What is Missing to Enable Optimization of Inventory Deployment and Supply Planning? Professor Sridhar Tayur Carnegie Mellon University ANALYTICS FOR A COHERENT ORDER FULFILLMENT STRATEGY Availability management Key policy choices Promising and meeting order fulfillment lead times Set to maintain or gain market share Fixed or flexible Segmentation by product or customer (e.g. sales vs. rentals) Capacity management Stabilizing production rate to maximize efficiency or flexing capacity to meet demand Fixed or flexible capacity Willingness to subject plant to increased demand variability Demand management Managing sales/order rate variation Limiting number of allowed “standard” configurations in build-to-stock environment Active management of demand variability (e.g. promotions/incentives) Monitoring and managing forecast error Inventory management Optimal deployment of inventory to maximize availability at minimum cost Also used to insulate manufacturing from demand variability Static or dynamic inventory targets Rules of thumb vs. product/location/time specific targets Based on total chain or local viewpoint To achieve maximum availability at minimum cost: A comprehensive order fulfillment strategy must appropriately define a coordinated set of policies for these interrelated variables No one variable can be managed in isolation and changing or fixing one variable has implications for the others Lead time management Consistent with Lean principles working to reduce supply and in-process lead-times Monitoring and managing lead-time variability Active management of lead-times and lead-time variability Incentives and penalties for performance ©2002 SmartOps Corporation 2 ACADEMIC BUILDING BLOCKS: 40+ YEARS OF EVOLUTION, BREAKTHROUGHS, AND APPLICATION Late 1950s – 1960s 1970s-1980s 1990s Key contributors Clark and Scarf Arrow, Karlin Federgruen;Zipkin; Lee; Cohen; Roundy Muckstadt;Thomas;Zheng Glasserman; Tayur Key progress Fundamental issues identified setting the stage for decades of research Early inventory and stochastic* optimization models created Searching for simpler ways of computing optimal inventory policies for basic problems Stochastic optimization models developed to explicitly accommodate supply and demand variability, multiple time periods, capacitated, multi-echelon supply chains Improved computational approaches developed to address larger problems in “isolation” Successful “one-off” application to industrial-size problems Breaking of problems into manageable pieces Practitioners use rules of thumb and put pieces together heuristically * Stochastic: Involving or containing random or “uncertain” variables (e.g., uncertain demand, lead time, capacity, yield, etc.) ©2002 SmartOps Corporation 3 REAL WORLD: THERE IS SIGNIFICANT INEFFICIENCY IN OUR ECONOMY U.S. inventories Estimated inefficiency Economic opportunity $1.0 trillion 50+% $500+ billion Fundamental, persistent forces behind supply chain inefficiency: Inability to accommodate and actively manage inherent uncertainty, variability, and complexity across multi-echelon supply chains Local vs. global (“total cost”) optimization, metrics, and incentives – uncoordinated supply chain inventory and cost decisions within enterprises and across supply chains Underutilization of current data, systems, and available best practices, e.g., lack of dynamic, data driven reviews of “planner variability” ©2002 SmartOps Corporation What is Missing? Advanced, practical value chain planning and optimization to What is missing? accommodate and manage these forces 4 5 CASE STUDY #1: INVENTORY REDUCTION OPPORTUNITY $ Millions 1200 50 1150 275 875 365 510 Average inventory (2000) Actual reduction in 2001 Average inventory (2001) Planned Average reduction in inventory 2002 target (2002) Additional opportunity identified with SmartOps Suggested average inventory target (2002) Source: SmartOps Multistage Inventory Planning and Optimization Software 6 CASE STUDY #1: TYPE OF INVENTORY FOR FY2002: ONE PRODUCT LINE AT 95% SERVICE LEVEL 11 /4 11 / 20 0 /2 5/ 1 12 200 /1 1 6/ 20 01 1/ 6/ 2 1/ 002 27 /2 2/ 002 17 /2 3/ 002 10 /2 3/ 002 31 /2 4/ 002 21 /2 5/ 002 12 /2 00 6/ 2 2/ 20 0 6/ 23 2 /2 7/ 002 14 /2 00 8/ 2 4/ 2 8/ 002 25 /2 9/ 002 15 /2 10 002 /6 10 / 20 0 /2 7/ 2 20 02 $ 16,000,000 14,000,000 12,000,000 10,000,000 8,000,000 6,000,000 4,000,000 2,000,000 - Safety Safety+Prebuild Safety+Prebuild+Pipeline Safety+Prebuild+Pipeline+Cycle 2002 Weekly Sales Forecast Current Merchandise Inventory Key Takeaways The existing supply and demand variability drives the need for significant safety stock for products, particularly during the peak selling season Due to capacity constraints, there is also a need for pre-build inventory, meaning that plants will produce more inventory not because of system uncertainty, but because mean weekly plant capacity will exceed needed production in future periods 7 UNDERSTANDING MODELING APPROACHES The goal is to pick an approach that ensures confidence in the answer, quick hit improvements, and sustained execution Timing/dynamic frequency Low detail/granularity High detail/granularity Quarterly/monthly Annually/quarterly Weekly/daily Planner N/A Business Unit Planning and Operations Timed, regular data loading Planner & O.R. engineer Data-loader with manual start Organization Data management/update process ERP/APS detailed, dynamic data inputs Corporate/ Business Unit Strategy Data wizard and interface Manual, “metalevel” inputs, click and drag design N/A O.R. engineer Relation to existing processes Stand-alone Dynamic One-off studies Structural changes Driving execution “Dynamic value Continuous chain” improvement ©2002 SmartOps Corporation 8 WHAT IS THE OPTIMAL INVENTORY DEPLOYMENT FOR YOUR BUSINESS? To enable continuous and sustained improvement, a comprehensive approach must accommodate all forms and purposes of inventory ©2002 SmartOps Corporation 9 STOCHASTIC OPTIMIZATION IS NECESSARY Non-linear Total Cost Optimization – – – Linear and Integer Cycle stock Pre-build stock Pipeline stock Managing uncertainty Safety stock Shortfall stock APS challenges – Scheduling a factory – Packing a truck – Routing a truck Certain or near-certain “Deterministic” Linear, deterministic models are not appropriate for most critical inventory decisions in multistage, multi-product, capacitated, stochastic environments Uncertain “Stochastic” 10 A SUPPLY CHAIN MODELING PROCESS Commence data integration process Map the current value chain Select relevant variables, constraints, and objective function Entire network or subset All nodes or simplification of nodes Initial collection, cleaning, and QA of data Simplifying Understand assumptions to include underlying data or exclude variables, assumptions constraints, or nodes Ensure data makes considering quality of sense in business answer vs. speed of and supply chain answer terms Changes to “nodes” and “arcs” vs. changes to echelons and BOMs Review outputs - send to operational system/ process Post-process and summarize QA outputs Full, partial, or no automation of inputs and outputs Selection of planning granularity Select optimization algorithms Days, weeks, months Product hierarchy – sales model vs. MA # of nodes and time periods Refresh inputs Change structure of value chain Scenarios/ what-if Stationary or nonstationary model (e.g. # of forecast periods) Single or multi-echelon or hybrid Capacitated, uncapacitated Calculation/ optimization Manual, Aggregation/dis- Compare results Design, build, and Run test cases exception-based, aggregation with expectations run logical vs. actual data or automatic based on theory scenarios Units/$s/Weeks Understand export of targets Rounding and domain Test boundary processing to planning expertise conditions speed systems ©2002 SmartOps Corporation Load data and pre-process meta-data Compute metadata: lead-times, lead time variabilites, forecast disagg. etc. 11 SOFTWARE ARCHITECTURE FOR ENTERPRISE INVENTORY PLANNING ©2002 SmartOps Corporation 12 OVERCOMING PRACTICAL DIFFICULTIES Reality Possible Approach Scale Scope: Many Factors Exist Simultaneously Data: Existence, Accuracy, Ease of Availability Silos within Organizations Multiple Companies in a Supply Chain Current IT Infrastructure Existing Execution and Decision Support Tools Metrics and Measurements Motivation, Discipline and Incentives Training and Capability People: Corporate supply chain and business planners/super users as well as business unit planners Consultants: Internal and External Professors and Education Exception Driven Scalable Software Comprehensive Approach Pre-processors, Inheritors, Data Loaders Net Landed Cost View Collaborative Framework with Trust ‘Bolt-on’s to co-ordinate/synchronize Productize recent OR/MS Intellectual Property Management 101: Track Key Performance Indicators Dynamically Culture and Metrics/Bonus Structure Need to have a Grassroots Revolution Flexible platform for Multi-tier use and communication Do not rely entirely on Spreadsheet based Optimization! Appreciate Reality and Train Students to Handle Reality 13 CLOSING REMARKS Despite ERP and APS investments significant inventory inefficiencies persist Fundamental causes of supply chain inefficiency must be addressed: – Inherent uncertainty and complexity in multistage supply chains • – Uncoordinated planning decisions • – Total cost optimization by providing visibility and coordination between functional and external groups Inconsistent and/or insufficient planning practices • Stochastic optimization approach is the appropriate solution Software can provide a standardized “best planning” solution All the drivers of inventory must be measured to determine: – Optimal inventory targets for all inventory purposes • safety, cycle, shortfall, pipeline, pre-build, and merchandising stock – Total cost solution to deliver service levels – Optimal service levels given budget objectives, product margins, and portfolio of products 14