Supply Chain Forecast Accuracy is usually measured with Mean Absolute Percent Error or MAPE, the average of percentage errors. But there are several other metrics to consider. Here’s an example …
Forecasts are never right but you can improve your forecast accuracy …
- You can’t improve what you don’t measure
- Aggregate – individual plus or minus swings cancel each other out
- Measure forecast accuracy at the right time fences
- Review forecasts by exception only. If the forecast deviation is within expected limits leave well enough alone
- Tolerance – decide what ‘good’ and ‘bad’ forecast means
- Triangulate – use multiple forecast methods
- Automate and remove the human factor
- Forecast less – less often, less granular
- Demand Forecast Segmentation – Forecast-ability vs. Benefit
- Reduce supply lead time, and you won’t need to forecast (as much)
Dear Larry, Andrew and Guanair; How are you doing ? I would like to share with you that We Won the Whirlpool award Sustainability-2013 , among 536 suppliers and 55 projects, With our project about plastic returnable box saving industrial waste (avoid 75000 cardboard boxes /2500 wood pallets /50 km of plastic wrap /40 ton of contaminated waste in the last 5 years) and inventory reduction from 20 to 8 days of course. We are doing our part for a sustainable world. Thank you very much for the support.
Merry Christmas and happy new year!
Luiz Antonio Basso
Associated Spring, Barnes Group
Campinas, Sao Paulo, Brazil
Flow Consulting clients have often requested us to put together a Green Belt LeanSigma class, and we now have enough demand to guarantee the minimum number of students. So we are partnering with one of our key clients, Gulf States Toyota, to run the class this fall.
- Provide a fundamental understanding of LeanSigma Principles, DMAIC Project Management, Problem Solving, and the Theory of Constraints
- Learn LeanSigma techniques to reduce waste and process variation and increase speed and efficiency, expand capacity, and improve profitability
- Receive mentoring time on LeanSigma projects required for the coursework
- Course is taught by experts with real world experience
- Students gain access to the latest LeanSigma thinking through real-world case studies and highly interactive simulations
- Participants gain exposure to diverse industries through lively discussion and student interaction
Course Structure & Registration:
- Date: 9 Days – October 1-3, November 12-14, December 17-19
- Location: Gulf States Toyota, 20412 East Hardy Street, Houston TX
- Cost: $4995
Key Goal – If the proper project is brought to the class, the training will be ‘free’ as the results of the LeanSigma project should more than pay for the class.
Contact Ebony Norflis at Flow Consulting
We know better than to try to force fit strategies yet time and again we find businesses planning and managing their processes with a ‘one size fits all’ approach. Hospital services, insurance underwriting, pharmaceutical production, consumer electronics assembly, integrated supply chain services all struggle with finding the right balance of specialization and common process. At the extremes there is the chaos of infinite flexibility where every order, customer, patient is an exception to the other extreme where the ‘cats and dogs’ are blocking traffic on the highway.
Simply sorting process demand (orders, service requests, projects, patients) by volume or frequency isn’t good enough.
Adding the dimension of demand variation is an improvement, but optimizing inventory, process performance, supply chain doesn’t end with a 2×2 graph of Volume vs. Coefficient of Variation. But before we tackle total cost, margin analysis, supply chain policies let’s go over the fundamentals.
Gather up representative data for orders, arrivals, requests – whatever you have that represents demand on your process, business, factory, enterprise. In HR this might be recruiting requests, in product development an inventory of active projects, in a factory grab customer orders. [ABC Analysis: how to]
Sort the product or services from high to low, left to right. The highest volume activities want to be streamlined, maybe even have dedicated processes and resources. Call these A items the ‘highway’. The lowest volume of orders or frequency of service items, the C items may need to be handled on a case-by-case basis. If you don’t then you’ll have a ‘mixed model’ process where the A’s and C’s get in each others way.
Products and service have a lifecycle; usually new products volume increases and the initial demand is spotty but stabilizes as the product matures. Then new products compete and demand declines and becomes irregular for the first generation.
5. Then calculate some simple descriptive statistics and interpret
Since 1.45 is within +/- 1.89, we can say that the data is marginally normal.
Process Model Selection
- Margin Contribution & Cost to Serve may be more insightful than just looking at demand quantities or sales dollars.
- Products and services may have different demand, inventory, customer replenishment, supplier replenishment, sourcing and total landed cost, and other policies.
- Managing these policies in a dynamic business can be a challenge; so look toward optimization automation to stay ahead having the right strategies for the right products instead of one size fits none.
Why can’t Sales give us a decent forecast? Paraphrasing George E.P. Box, “All forecasts are wrong, some are useful.” If forecasts are wrong, then why forecast? Supply chain lead time is often much longer than the customer is willing to wait, and so the business carries inventory somewhere, and inventory is a waste. The Lean approach to crush lead time and make-to-order isn’t always possible and so we may need to build on speculation, or at the very least buy raw materials. Even after making customer consumption visible to all, reducing lead time, creating flow, using pull systems, increasing velocity and reducing variation, and establishing a multi-echelon collaborative and process disciplined environment you may still need to ‘predict’ future demand volume, product mix, and/or seasonality. So whether you want to or not, or have a formal forecasting method or not, somewhere in your purchasing and planning world someone is anticipating what the customers may want and is placing supply orders or adjusting inventory levels or rearranging workloads. Lean Forecasting Principles:
- Measure forecast accuracy, seek root cause for inaccuracies, and strive for continuous improvement of the forecasting process
- Aggregate (family) forecasts are always more accurate than individual SKU’s – plan families, build SKU’s on pull
- The ‘best’ forecast is a customer order, the next best is to replenish what the customer just consumed
- Forecast accuracy usually degrades the further out – try to match the forecast period to the supply lead time
- Relentlessly seek to crush supply lead time and variation (mix and volume), and eliminate the need for near term forecasts
- Drive waste out of the forecasting process
A collection of current thinking on Lean Forecasting:
Lean and Forecasting from Michael Balle’s Gemba Coach Column
Building a Lean Fulfillment Stream by Martichenko and von Grabe .
Over time, inventory goes up and down. As customer order shipments are made inventory transactions subtract the ship quantity from the quantity on hand. When production is made or goods received into the warehouse the quantity on hand goes up. This up/down cycle can be thought of as a ‘saw tooth’. Except it’s never as neat and tidy as this.
Cycle stock can be calculated as ‘Demand during Lead Time’, or if we make or buy product once a period I should have at least enough to last until the next order arrives. One way to do this is to first determine the replenishment time (RT) or the time between orders. For high volume products we may make the product every month, so the RT would be 30 days or 4 weeks or 1 month.
Next we need to know the consumption, sales forecast, or historical shipments and then calculate the average demand during the replenishment period.
AD = average demand, RT = replenishment time, or campaign interval, or period between receipts.
In this example we receive 400 every period. Our stock on hand goes from 400 down to zero just as the next resupply arrives. The average inventory isn’t 400 or 0, it’s 200. When there are many products being received and issued every day some are near max, some are near zero, but the average for all products is ½ (max – min).
But this is still too simple. Consumption (issue from stock, or shipments) is never exactly smooth, sometime higher or lower than planned. Sometime resupply is earlier or later than planned or in different quantity than planned. So we have two types of supply & demand variation to think about. We need safety stock and safety lead time to protect our customers from variation in supply and demand.
In this example we have 100 in safety stock. If everything were perfect inventory would cycle between 500 and 100. We keep 100 just in case production is late or customer demand is higher than expected.
Demand can come in faster or in greater quantity than we had planned (red dashed line). If we can’t get the product resupplied sooner we have a period of stock out or back order.
With safety stock we can try to prevent back orders or stock outs.
We dip in to safety stock when resupply is late.
Putting this all together we get …
Demand Safety Stock can be calculated, but first we need to look at some statistics. Here is graph of weekly shipments for CARILATUDE 350MG for the period of 1/1/2012 to 12/31/2012.
Here are some descriptive statistics:
Total shipped = 1,066,772
Weekly Average = 20,449
Max Week = 61,776
Min Week = 6,360
Weekly Standard Deviation = 8,883
Coefficient of Variation = Std Dev/Average = 0.43
How much inventory should we keep for this product? Do we want to have enough to cover the two big spikes or is it OK to go on back order for a week or two? Before we can answer these questions we need to know a few things.
- How often do we receive this product in to stock? If we make the product once a month, then we need to have enough in stock to last for a month plus any uptick in demand and some more to cover if production is late on the next run.
- What service level does management want to achieve? You would need infinite inventory to cover all possible customer demand and supply issues.
We don’t have enough money or warehouse space to hold all the stock we might want. So we have to accept some level of stock out or back order. Where we carry more stock than we need we have consumed raw materials, production capacity, machine time, labor that could have been used to make other products. When we don’t have enough stock on hand we are often forced to change the schedule and make product earlier than planned, and this means other products get pushed out. The trick is to set just the right level of target inventory with not too much safety stock.
To set Safety Stock we can consider three types of additional inventory based on the source of variation:
Demand Safety – customer demand can come in greater volume than expected and deplete inventory before the next planned replenishment (raw material or production run).
Supply Leadtime – production can be late due to machine breakdown, labor availability, scheduling changes, lost batch, or get hung up for a variety of reasons
Murphy – unpredictable circumstances that management might want to hedge.
There is also seasonality and planned stock builds, but we aren’t addressing those here.
Demand Safety is based on service level and demand standard deviation.
The safety factor is based on the service level management is trying to achieve.
|Service Level %||Safety Factor|
When we calculate, for each product, the weekly shipment historical margin contribution average, standard deviations, and Cv and then graph the average vs Cv for each product we get the following Demand Segmentation graph …
Here we see that there are many product with low volume that are very unpredictable (Cv > 1.0) unfortunately we may have to keep stock for these products. Because the history shows that shipments are erratic we may have to carry a lot of inventory to be able to cover the unpredictable demand. Fortunately most of the high volume products are predictable, but some are on the edge of being noisy and so we may have to carry a large amount of safety stock for these too.
We would like to announce a new entrant into the lean blogosphere, it’s called Lean MathTM (leanmath.com).
We know what you’re thinking, “Lean Math?!” Now, that’s a subject that evokes passion in the heart of every lean practitioner…right?
But, the truth is effective lean transformations require some level of math, whether it’s the often deceptively simple takt time calculation, sizing kanban, calculating process capability, or anything in between. It’s hard to get away from math. There is no such thing as math-free lean and certainly not math-free six sigma!
Lean MathTM is not intended to be some purely academic study and it does not pretend to be part of the heart and soul of lean principles. Rather, it’s a tool and a construct for thinking. Here we want to integrate lean math theories and examples with experimentation and application.
Some background. Within the next year, the Society of Manufacturing Engineers will be publishing a book, tentatively entitled, Lean Math. Mark Hamel, author of the Shingo Award-winning book, Kaizen Event Fieldbook and founder of the Gemba TalesTM blog, and Michael O’Connor, Ph.D. (a.k.a. Dr. Mike) are co-authoring this work. They are also getting a ton (!) of help from Larry Loucka, friend, colleague, and fellow-blogger at Lean Sigma Supply Chain.
No surprise, we’re the three folks who are launching this blog. The formal launch date is February 14th – because we LOVE math! Ok, love may be a bit strong. We really LIKE math.
Here are some of our first blog posts:
- Cycle Time
- Square Root Law
- Min/Max Cut Theorem
- Coefficient of Variation
Ultimately, we hope that you will join our fledgling Lean MathTM community and that it lives up to our blog tag line, “Figuring to improve.”
Mark Hamel, Larry Loucka, Michael O’Connor, Ph.D.
Got asked what would happen to inventory when the number of stocking locations change. I thought for a minute and remembered a quick rule of thumb … The Inventory Square Root Law states that …
Average inventory increases proportionally to the square root of the number of locations in which inventory is held.
X2 = (X1) * √ (n2/n1)
n1 = number of existing facilities
n2 = number of future facilities
X1 = existing inventory
X2 = future inventory
Here’s an example:
Current inventory is 4000 units, 2 facilities grow to 8. Using the square root law the future inventory = (4000) * √ (8/2) = 8000 units.
* The Square Root Law was mathematically proven by D.H. Maister in his 1975, International Journal of Physical Distribution article entitled “Centralization of Inventories and The Square Root Law.”
Determining an appropriate production model starts with Demand Profile and Demand Segmentation. High volume low variability items, and low volume high variability items behave very differently. How to decide if a particular product is a candidate for a one piece flow cell or a craftsmen job bench? Look to the coefficient of variation for a clue.
Type 1 – Rate-base or Just-in-time
- forecasting of the flow rate or takt time
- RCCP – rough cut capacity planning to monitor impact of mix and volume on pace maker operation
- produce to rate (or TAKT) vs discrete order or customer pull
- demand flow vs time-phased requirements planning
- maintain flow priority and timing
- no detailed Capacity Requirements Planning required
- no or minimal shop order launch or inventory transactions
- highly visual and standardized shop floor control
- “one-piece” flow, zero inventory, standard WIP – work-in-process
- seamless flow/pull of material
- Dynamic cycle time (Little’s Law)
Type 2 – Pull
- combination of discrete forecasting and/or demand rate-based forecasting
- MRP planning — pull Kanban, Heijunka visual shop floor control
- RCCP, but no detailed CRP
- flat Bills Of Materials
- more cellular manufacturing
- point-of-use vs. central stores
- inventory is strategic: standard inventory, time-based replenishment, pull based on consumption vs. push based on demand
- based on statistically balanced rate, build to level-loaded demand with calculated standard inventory buffers
Type 3 – Push or Job Shop Discrete
- discrete requirements planning (firm orders and long range forecast)
- Rough Cut Capacity Plan
- time phasing of requirements
- application of order policies: lead time, safety stock & time
- Capacity Requirements Planning
- MRP shop order launch & order maintenance (message filters and “noise management”)
- ability to aggregate disparate requirements across multiple products by work center, supplier, product
- central stores of inventory
- multi-level inventory: stores, pick, kit, move, queue
- batch processing
- demand leveling difficult and uneconomical
Hot off the press from the Lean Enterprise Institute …
Page 12 & 13 have a brief description of Coefficient of Variation and a SKU Scatter Diagram (weekly volume vs. SKU stability). 10 weeks usually isn’t sufficient for meaningful or statistically significant calculation of standard deviation. 25 data points would be better, and you might need a lot more if there’s any seasonality to deal with.
The guidelines given need to be tempered with the granularity of the data. While a coefficient of variation of less than 1.0 can be considered stable for weekly data, it would be considered very noisy when using monthly data and quite stable when using daily demand.
This small quibble aside the authors Martichenko and von Grabe do a wonderful job describing lean principles for the supply chain, or as they prefer, the fulfillment stream.
A measure of the volatility of customer orders (or any time series), aka Demand Linearity
- Calculate the standard deviation (s) of the historical demand, use appropriate time buckets (daily, weekly, monthly)
- You might need to discard abnormal demand
- Calculate the historical mean (x) (or average) or use the forecast mean
- Then calculate the coefficient of variation (Cv) Cv = s/x
- Low variability is a Cv less then 1.0, very stable demand is a Cv less than 0.5
- Determine Total Cycle Time
- Determine Queue Times between steps
- Create Step segments proportional to the task times
- Place steps, queue’s along the line segment in the order that they happen
> Place Value Adding steps above the line
> Place Non-value Adding steps below the line
- Draw in feedback loops & label Yield percentages
- Sum Activity / Non-activity times
- Sum Value / Non-value Times
- Issue: Oil producer was experiencing long and unpredictable plant maintenance turnarounds (also known as a shutdown or an outage). Downtime durations consistently exceeded plans and budgets, and this resulted in lost production and excessive maintenance costs.
- Solution: Applied lean techniques including:
- SMED (task separation, conversion, and simplification to complete outage tasks outside the outage window),
- Constraint Busting (e.g., dramatically improved critical path management),
- Visual Controls (e.g., optimized Outage Command Center),
- Streamlined Collaborative Planning (e.g., earlier and greater involvement of contractors and cross functional personnel), and
- High impact metric capture and utilization (e.g., shift in focus from “Cost only” to Cost and Duration).
- Result: Greater Profits – More stable outage planning and execution process with a reduction in outage durations of 10-15% resulting in increased production valued at $25 million increase in annual profit. Reduced Costs – Decrease in outage costs valued at $2-3 million annualized. Improved Collaboration – Significantly improved collaboration among all groups, greater performance reporting transparency, and improved continuous improvement through capture and application of lessons from one outage to another.
Here’s a more modern z Score Calculator
Well its getting to be that time of year again… What did I promise to deliver this year, what do we need to do next year?
The annual operating plan sometimes is developed and displayed using the X-Matrix. Establish the results of goals for next year, take the strategies and tie them to tactics, make sure the tactics can be measured (targets) and have individuals assigned ownership of tactics and targets.
Little gets done without marching orders, i.e. a Charter. The basic document of the hoshin process is the team charter. The A3 format connects the targets (goals) to the tactics and provides another level of critical thinking about execution. The team charter is a contact between the company to provide support and resources and the team and team members to do the hard work of problem solving, applying the scientific method, and running experiments on the management operating system.
What gets measured gets better, and so we set plans and track key performance indicators.
Ready to make your hoshin for next year?
Kanban is a system that supports level production by helping maintain stable supply and efficient operations. The question of how many kanban are needed is at the core of designing and running a kanban system. If your business process performs mostly standard, repeated operations the number of kanban can be calculated as follows:
Number of kanban = (Daily Demand * (Replenishment Time + Safety Margin))/Standard Container Quantity
- Daily Demand = monthly orders / work days in the month;
(Can use historical actual orders if demand is stable, may need to use current booked orders or forecast)
- Replenishment Time = sum of all the processing, transportation, handling, and queue times from freshly empty container to full container back to empty again
- Safety Margin = either a statistical calculation to accommodate variation in demand and/or supply, or an intuition to add zero to a few extra days
(Regardless of which approach you pick, once up and running begin removing kanban one at a time until you’ve gone too far, then add one back and spend some time to figure out the cause – i.e. lower the water level and expose a few rocks.)
- Standard Container Quantity is where we often have the greatest latitude. We usually can’t change the daily demand. Speeding up the replenishment time takes time. Selecting the right size container we can do right now. Sometimes we’ll turn the equation around and instead of solving for the number of kanban we’ll pick the number of kanban and then determine the right container size, like so …
Standard Container Quantity = (Daily Demand * (Replenishment Time + Safety Margin))/Number of Kanban
Rule of thumb: try to keep the number of kanban between 2 and 10 by adjusting the size of the kanban container.
Let’s see, DMADV should work …
|Define||Form a small team and agree on the goals of the project in a written charter with the sponsor, customers, and stakeholders. The overall intent should be clear. Deliverables and milestones should be established. Cross functional turf issues should be addressed up front.|
|Measure||Quantify the customer needs as well as the goals of management. Determine “What will project success look like and how will we know?”|
|Analyze||Each of the cost categories of Price, Transportation, Customs, Inventory, Overhead, and Risks need to be mapped and taken to another level of detail. For each element we need to study the inputs, outputs, alternatives, existing reporting processes to determine the accuracy and timeliness.|
|Design||Create, test,and document the new data collection and reporting process|
|Verify||Confirm by sensitivity testing, simulation or otherwise, the performance of the total landed cost reporting design and its ability to meet the target needs|
Here are a few more lean and six sigma tools that can be applied in supply chain. Have any additions, comments, or examples to share?
|Lean Sigma Tool||Definition||Supply Chain Application|
|Location Checksheet||A common visual quality data display in manufacturing is to take a product drawing and make a mark or place a sticky dot on the location of a defect or touch up. After a period of time you’ll often see clusters. Then we use good old Pareto and focus our team based problem solving skills on the areas of interest.||Plotting the physical location of inventory accuracy errors can often be a clue for getting to the bottom of and eliminating a significant source of wasted time. Similarly marking the location of packaging damage can help identify problems with overhang, pallet specification, strapping, and handling.|
|Sampling||Manufacturing process and quality engineers have been taking product and process samples for over 50 year as a routine part of statistical process control or designed experiments. 100% inspection is actually less accurate in quality control than is a well designed sampling plan and the use of descriptive statistics. As an aside the US Census could stand to use more sampling and less door to door canvasing.||A full physical inventory count or ‘stock take’ is also less accurate that a well designed cycle counting program. But even a cycle counting program is a waste of time if the errors discovered aren’t studied for root cause and permanent corrective action taken. Whether its an annual full inventory or a daily cycle count if all we do is adjust ‘the book’ then we aren’t doing anything to improve our future.|
|Statistical Distributions||The widely known ‘bell shaped curve’ of the normal distribution is often a good approximation of the spread we find in machining operations. Paint thickness, electrical resistance, tensile strength can vary plus or minus around a mean or average. Descriptive statistics such as mean and standard deviation help us understand and describe the behavior of the systems we are studying.||Caution Will Robinson. Playing with statistics without the proper training can be dangerous. Real example: when calculating safety stock and expected inventory we often need to consider the supplier lead time. Like any variable measurement there is always some spread, the expected 10 days could be 9 days or 15. Lead time is almost never bell shaped. Suppliers are rarely early. So which distribution to use? Find a good black belt and give’m a job.|
|Control Charts||Control Charts are how we display the behavior of a process and help process operators decide when to make an adjustment, stop the process, or start an investigation. We plot data taken from periodic samples and then follow SPC rules to determine if there has been a change in the process since the last sample.||Kanban are containers or cards used to control the replenishment, supply, production of product. The number of Kanban in circulation can be calculated based on the average consumption, replenishment time, and container size. A single card or container then has an expected lifecycle from empty to empty. By periodically sampling the time the container last passed through a ‘tollgate’ we can get an early warning on shifts in demand or replenishment time, hopefully in time to avoid a stock out.|
|5 Whys||First impressions are sometimes wrong, so when we are brainstorming or investigating a situation we’ll ask about the cause of the cause of the cause. A method for pushing our thinking beyond superficial solutions that don’t really solve the problem.||Took 20 minutes to get started picking this morning. Why? Because the printer was jammed? Why was the printer jammed? I guess the rollers were dirty. Why where the rollers dirty? … You get the idea? We keep asking Why until we get to something we can do something about like adding a weekly printer maintenance task to our TPM schedule and assigning responsibility for doing it.|
|Pull Systems||Trying to predict (forecast) what to make and when is tough to do in many industries. Toyota found great advantage in only making what was needed when needed, that is to replenish only what was consumed. Ideally a supplying operation would hand off one piece at a time to the down stream consuming operation. But when supplier and customer can’t be in close physical proximity we need some way to communicate what is needed and when. 2 Bin, kanban, FIFO flow lanes are just a few types of pull systems common in manufacturing.||Some have tried using pull thinking in distribution inventory management, only replacing stock at customer facing warehouses when product is shipped out (Toyota accessories for example). The traditional approach is to forecast the demand and then make or buy a batch large enough to cover the future demand, and hope you didn’t plan too much or too little. Pull works well in some industries and not at all in others. Most warehouses regardless of industry can use pull techniques for resupply of packaging, fresh pallets, wave picking period.|