Forecast “Push”, Customer “Pull”, and Hybrid Models

Forecast “push,” customer “pull,” and hybrid models Module 11.1 Bruce Pan, LFM ‘06 Nicholas Svensson, SDM ‘03 Toni Albers, LFM ’00 - Honeywell Presentation for: ESD.60 – Lean/Six Sigma Systems MIT Leaders for Manufacturing Program (LFM) Summer 2004 These materials were developed as part of MIT's ESD.60 course on "Lean/Six Sigma Systems." In some cases, the materials were produced by the lead instructor, Joel Cutcher-Gershenfeld, and in some cases by student teams working with LFM alumni/ae. Where the materials were developed by student teams, additional inputs from the faculty and from the technical instructor, Chris Musso, are reflected in some of the text or in an appendix 1 Overview ¾ Session Design (20-30 min.) ¾ Learning Objectives ¾ Define the attributes of forecast driven “push” manufacturing system. ¾ Explanation of the control phases of a “push” system. ¾ Define the attributes of customer driven “pull” manufacturing system. ¾ Comparison of “push” vs “pull”. ¾ Examine the benefits of a hybrid push-pull system (Honeywell Avionics) ¾ Part I: Introduction and Learning Objectives (1-2 min.) ¾ Part II: Key Concept or Principle Defined and Explained (3-5 min.) ¾ Part III: Exercise or Activity Based on Field Data that Illustrates the Concept or Principle (7-10 min.) ¾ Part IV: Common “Disconnects,” Relevant Measures of Success, and Potential Action Assignment(s) to Apply Lessons Learned (7-10 min.) ¾ Part V: Evaluation and Concluding Comments (2-3 min.) 07/07/04 -- 2 © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT Part I: Introduction Part II: Concepts Part III: Application Part IV: Disconnects Part V: Conclusion Key point – discussion of push system and its inherent attributes and a discussion of a pull system and its inherent attributes. Taken as extremes each has its own strength and weaknesses. We will attempt to show that taking the most appropriate pieces from each is the best strategy. Appropriate as defined by someone who fully understands the business model, its constraints as well as the principles of MRP and Lean. 2 Attributes of a “push” System ¾ Manufacturing activities are planned based on a market forecast rather than actual customer demand. ¾ Implicitly this means: ¾ There is an emphasis placed on a central planning function. ¾ Service levels are assured by increasing or decreasing finished goods inventory levels. ¾ The system optimizes “efficiency” rather than “effectiveness” by level loading the factory. ¾ Material flows through the factory in batches following a prescribed routing sheet attached to the work order. ¾ There is a heavy reliance on heuristics to compensate for the inherent complexity of the optimization problems encountered. ¾ “planning horizons” and “fences” are used to adjust the production plan on a weekly or monthly basis based on the forecast. ¾ Distinction between dependant and independent demand. Adapted from MIT LFM thesis by Sean Hilbert 07/07/04 -- 3 © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT Part I: Introduction Part II: Concepts Part III: Application Part IV: Disconnects Part V: Conclusion Highlight some of the major implicit features – time fences, forecast driven, complicated heuristics and computer delegated decision making. Works very well in a stable environment – key is the discussion of how the system behaves in a changing environment. i.e. how do you deal with change and what are some of the system constraints inherent with a push system. 3 Attributes of a “push” System cont.. Forecast Factory Purchase Orders on Suppliers MRP/ERP Work Orders Work Center A Supplier A Supplier B Supplier C Work Center B Work Center C Work Center D Router Raw Material Stores Purchase Orders On Factory Finished Goods Warehouse Physical work flow Information flow Customer A Customer B Customer C Ending Inventory = Beginning Inventory + Planned Deliveries - Net Requirements *Adapted from Nahmias – Production and Operations Analysis © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 4 Extreme case of push system with centralized decision making and little to no communication between the various stakeholders. This is an extreme example intended to highlight the unique differences between push and pull. 4 Simplified Control Phases of a “push” System Demand Forecast Resource Planning Aggregate Production Planning Rough-cut Capacity Planning Master Production Scheduling Bills of Material Inventory Status Wagner – Whitin Silver-Meal EOQ… Material Requirements Planning Job Pool Capacity Requirements Planning Job Release Routing Data Job Dispatching © Wallace J. Hopp, Mark L. Spearman, 1996, 2000 © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 5 Standard MRP cycle showing capacity requirements planning. It may be worth mentioning that due to the complexity of the mathematics heuristics are often used. The classic lot size heuristics are Wagner-Whitin (1958) and Silver-Meal (~1970). The system tries to optimize for utilization not necessarily effectiveness. 5 Attributes of a “pull” System ¾ Manufacturing plan is based on actual customer demand “pull”. ¾ Implicitly this means: ¾ Control of manufacturing execution is at the working level. ¾ Service levels are assured by increasing or decreasing kanban levels between workstations (WIP). ¾ The system is optimized for “effectiveness” which is achieved through continuously improving “efficiency”. ¾ Material flows through the factory based on visual queues triggered by customer “pull” from the final kanban. ¾ The system requires a very hands-on management style. ¾ Culmination of all lean principles. Kanban, Andon, Kaizen, 5S, SMED… © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 6 Highlight some of the major implicit features – work is pulled through the supply chain based on visual queues at the working level. Pre arranged kanban levels are determined to deal with uncertainty and disruption. System focuses on effectiveness using continuously improving efficiency as a way of reaching its optimum state. No work is done unless it is needed by the customer. Forecasts are only used to size the kanbans but not for daily fluctuations in demand. A lean pull system utilizes all lean principles preferably throughout the supply chain. 6 Enabling Lean Principles Required for a “pull” System Supplier kanban level agreements Customer kanban level agreements. Manufacturer Forecast Supplier A kanban Supplier B kanban Suppliers Work Center A 5S Supplier C kanban Customer Work Center B Kaizen Work Center C Jidoka SMED Work Center D y= DL + w a Customer B kanban TPM Andon Six Sigma Continuous Flow Heijunka Takt Standardized Work Customer A kanban 7 Wastes Customer C kanban y = number of kanbans D = expected demand/time L = lead time (processing+waiting+conveyance) w = buffer stock level a = container capacity Physical work flow Information flow © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 7 Communication is at the working level. Forecast is used to form consensus amongst stakeholders (customer, supplier and manufacturer) about the capacity of the system and the levels of kanban to maintain. Note kanbans are owned by the supplier in each case. 7 Push vs. Pull Push Strength ¾ general approach ¾ MRP/ERP software available ¾ Better reaction to forecast changes by anticipating demand pattern. ¾ Advocates say it works. Pull Strength ¾ ¾ ¾ ¾ ¾ ¾ Focus on removing waste. Root cause corrective action. Minimizes WIP. Hands on management. Use of visual queues. Less expensive to implement Push Weakness ¾ ¾ ¾ ¾ ¾ ¾ ¾ Capacity planning Data integrity and training Forecast uncertainty System nervousness Masks underlying problems. Authority delegated to computer. More expensive to implement Pull Weakness ¾ Pushes inventory onto suppliers. ¾ Longer reaction time to changes in demand. ¾ Multi-sourcing more difficult. ¾ Requires higher supplier reliability and agility. ¾ Ignores future demand patterns *Adapted from Nahmias – Production and Operations Analysis © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 8 8 Hybrid Model “The issue is not to make a choice between MRP and JIT, but to make the best use of both techniques.” – Karmarker (1989) © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 9 Self explanatory. 9 Honeywell Avionics – Toni Albers LFM 00 ¾ Scenario ¾ Predominantly RF hardware and electronic black box hardware with large amounts of software. ¾ OEM customers representing 70% of sales typically order 100 systems at a time. ¾ Dealer network representing 30% of sales typically order 1-3 parts at a time. ¾ Transition from a pure MRP shop to a hybrid model. ¾ Circuit card assembly lead times of 4-6 months. ¾ Typical 6 month forecast accuracy of 45%, 3 month accuracy of 65%. ¾ Known end of quarter hockey stick effect. ¾ Mindset at plant was that “lean” was for the production floor and suppliers but not for the “middle piece”. © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 10 10 Honeywell Avionics Factory Demand Management (FDM) Current State: Future State: Schedule Variability Materials Overdrive ¾ Supply chain over-responds to imperfect forecasts ¾ RESULT: Wrong materials available at the wrong times Analytical Modeling De-Coupling + Z*σ*√LT ¾ Quarterly trends ¾ Strategic inventory needs ¾ Supply chain stability ¾ Predictably provide strategic inventory based upon prior demand & future forecasts Inventory Control Loop 2004 Timing: Q1 Predictability Forecast BUFFER MPS BUFFER Build Plan Q2 Q3 Q4 Quarterly Skew Variability Frozen within Lead Time Pilot ¾ ¾ ¾ ¾ ¾ ¾ Factory Go-Live Impact -6 MO -3 MO TODAY +3 MO 7 -9 mo. market trend ? ? ? MPS +6 MO +9 MO Migrated for 100% MRP based system to FDM – effectively decoupling the manufacturing floor from purchasing and scheduling. Manufacturing time represents only 2 weeks of the total 6 month lead time. Flat FDM forecast looks at normal level + safety stock, Hockey Stick model uses time series modeling to determine correct distribution for quarter/part + safety stock (where σ does not include hockey stick effect) Currently only building to customer orders, previously shop floor would build just to stay busy and typically in batches. Currently working to a D-1 production schedule. 30% of suppliers are currently on pull systems with 60% coming on line by the end of 2004. All part of a 5 year lean implementation plan at Honeywell. Toni Albers LFM 00 © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 11 Explanation of Honeywell’s use of a new algorithm to better determine demand for their hybrid pull process. 11 FDM Implementation Details Problem Statement: Actions Taken: Raw materials are driven by SIOP “calculable” & “forecasted” demand. SIOP end-item forecast accuracy ranges from 15% - 65%, and is highly variable over time. Master schedule is driven based off these firm and forecast requirements. Simulation on 30 part numbers Completed FDM FMEA Model and user operating procedures completed Project Definition: Identified specifications for model parameters Draft “FDM Netting” workbook completed Completed training - MS, Buyers, CPRSs, SIOP, etc. Deployed FDM workbook for Q3/Q4, 2004 Deploy FDM for Avionics Buy De-couple M/S from factory build process De-couple the SIOP forecast from the supply chain to establish a process that: • Provides increased stability to the supply chain • Reduces inventory levels • Improves material availability Base stock (type II service) model Time-series and average demand Drive the master schedule based upon historic demand and variation data, coupled with forward-looking aggregate demand trends. Partner with suppliers, manufacturing and the customer to manage a stable supply chain. Team Leads: Next Steps: Date: • Toni Albers - LFM 2000 • Miriam Park - LFM 2000 ‡ Automate FDM system feeds ‡ Integrate FDM to improved SPS system Q3 Q3 Toni Albers LFM 00 © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 12 Emphasize that this is a long term commitment to lean implementation taking many years. To fully transform a supply chain from push to pull can take 10-15 years based on Toni Albers input. Honeywell have a 5 year implementation strategy of which the FDM initiative is a part. There are also multiple challenges with mindset and overall organizational norms which need to be overcome when implementing lean in a Brownfield site. Some of these include batch building, weekly rather than daily ship targets, disconnecting Master Scheduling from Shop Orders etc.. All of these take time and the eventual implementation of many lean principles culminating in an implementation of a hybrid model. 12 The Ice Cream Parlor Exercise ¾ The scenario ™ There are two competing ice cream parlors. Both located on Newbury St. in Boston. ™The first is located at the corner of Newbury St. and Arlington St. It is named the 5 Scoops or 5S by the locals. It is a lean operation with limited shop floor space. ™The second is located at the corner of Newbury St. and Exeter St. It is named Dairystein in honor of the companies herd of milking cows. 07/07/04 -- 13 © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT Part I: Introduction Part II: Concepts Part III: Application Part IV: Disconnects Part V: Conclusion The purpose of the exercise is to examine how two organizations within the same industry can operate given two different operating systems. The 5S parlor utilizes a pull system and the Dairystein utilizes a push system. As the demand uncertainty is the same for both each team is forced to make compromises on their business strategy based on the constraints placed upon them by their supply chain models. The inherent constraint in the lean case is the kanban replenishment rate. The inherent restriction in the push case is the herd size and the continuous requirement to milk the herd regardless of demand. It is hoped that by the end of the exercise the teams would have approximately equal profits and when asked would have similar ideas on how to improve their supply chains. Depending on the demand profile provided during the exercise it should be possible to demonstrate the benefits of a push system in dealing with short term forecast variations (hockey stick) versus the benefits of a lean system in dealing with a more even but still uncertain demand profile. 13 Rules of Engagement ¾ An ice cream cone consists of the following parts. ¾ Ice cream ¾ Cone … 1 Scoop Vanilla … 1 Waffle Cone ¾ An ice cream cone is made using the following process. ¾ Take order (cashier) ¾ Scoop ice cream (attendant) ¾ Deliver/Store in Freezer ¾ The profit on each cone is $2. ¾ The cost of carrying inventory is $0.50/day/cone equivalent. ¾ Mean demand for July is 100 cones/day +/- 30 and is strongly correlated to the weather. The certainty interval is 1 σ. © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 14 Have two teams of 5 participate. I person from each team should witness the other team during the simulation to ensure consistent adherence to the rules of engagement. Each team fills in column 1 of the tables on charts 16-17 with their initial kanban sizes and herd sizes respectively. These “optimum” estimates should be a best guess based on the above forecast statement and other background information provided on slide 15. The simulation starts with the initial high of the day being given and each team completes column 1. The profit is recorded and charted on the black board on an x-y chart ($profit vs. day). Once complete the second day’s high temperature is provided and the days actual demand calculated. Each team can now begin to re optimize or leave their supply chains alone. The pull group can change their kanban size, which takes 5 days. This is to reflect the amount of time required to reach consensus in the whole supply chain. Alternatively the push team can choose to buy milk externally which has a 2 day delay associated with it. 14 Score Sheet 5 Scoops Ice Cream kanban Scoop Cashier Deliver Waffle Cone kanban Day 1 2 3 4 5 6 7 8 9 10 kanban size demand inventory $ sales $ profit/loss • Max kanban replenishment rate is 4 times daily. • It takes 5 days to re size the kanbans © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 15 Note: Inventory $ = $0.5 x what is left in the kanban each day. Sales $ = (kanban-demand) x $2 15 Score Sheet Dairystein Milk Cows Cashier Make Ice Cream Scoop Deliver Freezer Waffle Cone inventory Day 1 2 3 4 5 6 7 8 9 10 herd size milk order - demand inventory $ sales $ profit/(loss) • 1 cow produces 20 cones worth of milk/day (cows must be milked daily.) • milk can be ordered at $5/gallons with a lead time of 2 days (10 cones/gallon) • inventory $ = demand - (herd size * 20 + milk order (gal) * 10)*$0.50 + previous day © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 16 Remember to add the previous day’s inventory to the current days inventory. Unlike the lean model in the push model the inventory continues to grow and is carried over from day to day until consumed. 16 Other Factors and Predictions ¾ The farmers almanac predicts a very hot and humid July with above average temperatures. ¾ Due to the expected heat wave many Bostonians are leaving for Cape Cod. ¾ The DNC is being held in Boston this year at the end of July, a large influx of visitors are expected. ¾ Tourism levels 15-20% below 2000 but up from last year. © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 17 17 The Demand Actual Demand = (Expected High – 75)x5 +100 Boston, Massachusetts 10 Day Forecast by Intellicast © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 18 Actual equation can be modified to prove a point. This specific equation is based on the forecast for Boston July 2004, the url is directed to the actual 10 day weather forecast for the Boston area in this example. 18 Push/Pull system disconnects ¾ Technical Factors Push • MRP/ERP complex • Data integrity critical • Sys nervousness issue • Decision-making delegated to computer Pull • Requires multi lean factors to be in place • Requires highly dependable suppliers • Serial communications ¾ Social Factors Push • Worker disengagement • Root cause difficult to ascertain • Workers not idle • Relies on a single function group for execution Pull • Job security • Heavy reliance on upstream processes and groups • Metrics alignment critical 07/07/04 -- 19 © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT Part I: Introduction Part II: Concepts Part III: Application Part IV: Disconnects Part V: Conclusion Some of the more common disconnects typically caused when the systems undergo dynamic change or have to deal with uncertainty. With either system it isn’t practical to discuss any aspect of change until stability has been reached. This is the first and most important consideration for both systems. How stability is achieved is based on the business realities of each company. 19 Concluding Comments ¾ Continuous improvement and removal of waste applies equally well to both push and pull systems. ¾ The entire supply chain needs to be considered for either system. ¾ Pull systems provide visual and real-time feedback for improvement. Push systems tend to hide inefficiencies. ¾ Pull systems require more collaboration with suppliers as they expose to inventory-carrying cost via the kanbans. 07/07/04 -- 20 © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT Part I: Introduction Part II: Concepts Part III: Application Part IV: Disconnects Part V: Conclusion Lean principles apply equally well to both push and pull systems. Getting rid of waste and improving agility and flexibility help both systems to be effective, efficient and predictable. Typically it is easier to postpone dealing with problems in a push rather than a pull system due to the extra inventory and lack of immediate feedback of urgency. This however also tends to be one of the significant Achilles heals of the push system. 20 Appendix: Instructor’s Comments and Class Discussion on 11.1 ¾ Goal of activity: show that both pull and push can achieve similar results ¾ The difference lies in how each deals with unexpected developments (prevention and reaction) ¾ True optimization comes from hybrid in most cases © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 21 21 Bibliography 1) Steven Nahmias. “Production and Operations Analysis” Fourth Edition. McGraw-Hill Irwin: New York, 2001 2) Pascal, Dennis. “Lean Production Simplified: A Plain-Language Guide to the World’s Most Powerful Production System”. Productivity Press: New York, 2002. 3) Wallace Hopp & Mark Spearman. “Factory Physics” Second Edition. McGraw-Hill Higher Education: New York, 2000. 4) James P. Womack, “The Machine That Changed The World”, Perennial; ISBN: 0060974176 Websites: www.mhhe.com www.factory-physics.com www.iee.org/oncomms/pn/manufacturing/leanvillage/pull.pdf © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT 07/07/04 -- 22 22 Appendix: Instructor’s Guide Slide Time Topic Additional Talking Points 1-2 2-3 min Introduction, overview and learning objectives • Quickly walk through the learning objectives which are to understand the principles of push and pull systems. And to evaluate a hybrid implementation at Honeywell – Toni Albers LFM00. 3-9 3-5 min Key Concepts • See speaker notes on each slide. Discuss how the push and pull systems deal with demand uncertainty and disruption. 10-18 7-10 min Exercises/Honeywell Case • The Honeywell case is an excellent example of a lean implementation in a brownfield site. The material was provided by Toni Albers LFM 00. The Ice cream parlor exercise. The intention of the exercise is to show that in the long term both systems perform relatively well. Also, the push system can react to short term changes more effectively than the pull system whereas the pull system is more effective in the long run as problems faced need to get addressed right away and are obvious. In this case the problem is either too much or too little capacity. 19 5-7 min Disconnects • See speaker notes on each slide. The points given highlight the most sensitive areas of each type of system. 1-2 min Concluding comments 20 •Understand your business and use your expertise to tailor a system which optimizes07/07/04 your-- 23 Operation. © Bruce Pan/Nicholas Svensson–ESD.60 Lean/Six Sigma Systems, LFM, MIT Part I: Introduction Part II: Concepts Part III: Application Part IV: Disconnects Part V: Conclusion 23