Silver bullets to tackle the intermittent demand problem

May 18th, 2012 by Dr. Chockalingam

Intermittent demand or sporadic demand is a fact of life.  In many cases, demand occurs in one lump followed by periods of no demand.

What drives intermittency in demand patterns?

  • Fixed Ordering Costs
  • Transportation Costs
  • Purchase Quantity discounts
  • Shoe Leather Costs….

In any case, it is important to come up with strategies to plan for intermittent demand.  We typically advise clients to follow a set of strategies in sequence to address the intermittent demand problem.  Rushing to find a magical statistical model to forecast intermittent demand is not the best first method.  The software industry has created a huge business trying to advocate solutions to solve sporadic demand, pounding the tables on their method being the best in the industry……….

SKU segmentation is an important first step before tackling the problem of intermittent demand.   This helps you to prioritize only those SKUs where intermittent demand needs to be actively dealt with from a planning point of view.

Not all non-zero demand pattern need to be considered intermittent.

  • Some could be seasonal and
  • Reasonably predictable with standard models.

If you have a good software package, expert selection will determine if a particular demand pattern is truly sporadic demand.   Typically the timing of the non-zero demand should be unpredictable.  More often than not, you should have more zeroes than positive demand points.

A typical scenario would look something like this:

What are the steps that effectively solve this challenge?

There are some business strategies you could adopt before looking at algorithmic options.  Although Segmentation is quantitative and requires some datamining, Segmentation allows you to prioritize and highlight the real problems.

Disciplined demand planning for intermittent demand should look at integrated demand-supply strategy and Sales and Marketing intelligence first.

First Step:  Come up with a holistic demand-supply strategy to address items with sporadic demand!

  • Work with your supply chain and perform a Cost benefit analysis of carrying more safety stock to handle the anticipated intermittent demand
  • In general, sporadically ordered MTS items could be also low value.  Carrying more inventory may not break your bank.
  • Correlate the deployment strategy to the Lead time and review of aggregation of demand will enable a better supply and safety stock strategy.  If your lead times are longer, then aggregation of the demand buckets may result in lower noise and hence may result in lower inventories.

Second Step:  Get Sales and Marketing intelligence on the timing of the demand – they may be privy to customer ordering patterns.  They may also be able to give you the magnitude of the demand.

  • Check if demand gets splintered because of product migrations such as bonus packs that are made available only during certain periods.
  • A better forecast can be delivered by combining the split skus into one family.
  • Model that family and route the forecasts to the available packs  – Aggregating across products
  • Ask the question if there are promotions that are driving this intermittent demand.  If we offer price promotions, customers may stock up on the item resulting in no orders for the following few months.

If product migrations cause the intermittent demand problem, then we should apply the principle of demand chaining – aggregate the related SKUs together to create a forecast model.  Then use proportional forecasting methodologies to split the forecast down to the lower level SKUs.  Once you aggregate to a certain level, you may see a smooth pattern and the intermittency may disappear.

The last resort will be to use the Statistical Models:

1.  Croston’s Models – Simply this averages the timing difference and the quantities to arrive at a constant forecast.  What is important is to look at the confidence limits to plan for inventory.

2.  Discrete Distribution models – You can use the Poisson model and the negative binomial models in certain cases.

In our Demand Planning Workshops, we will discuss this in detail.  Perhaps see you all in California in June!

http://demandplanning.net/demandplanning_tutorialCA.htm.

Tags: , , , , , , , , , , ,
Posted in Forecast Modeling, Forecasting Process, Inventory and Operations Planning, statistical modeling, supply chain collaboration, Uncategorized | No Comments »

The use of the NORMSINV function in Safety stock calculations

March 28th, 2012 by Dr. Chockalingam

There was a question on Linked in about the use of the NORMSINV function in Excel.   Initially the poster mentioned the NORMSDIST function.

But for safety stock purposes, you already know the required service level percentage.   You need to derive the Service Level Multiple or the Z-factor.

So NORMSINV does the job.

So the Z-factor = NORMSINV ( Service level percentage).

For example, NORMSINV(98%) = 2.05 etc.

So the safety stock = NORMSINV (Service Level%) * RMSE * SQRT(Lead Time).

You need to calculate the Root Mean Squared Error from the Forecast at the item level and use the Lead Time based on the supply information.

Happy Forecasting!

Tags: , , ,
Posted in Inventory and Operations Planning, Uncategorized | No Comments »

Is Statistical Modeling an After thought?

March 21st, 2012 by Dr. Chockalingam

I just had a conversation with a Fortune 500 executive recently.  He mentioned his company is spending tens of millions of dollars currently upgrading from SAP APO 4.0 to SAP SCM 7.0 Demand Planning.

Come to think of it, what are the big differences between the 4.0 or 5.0 vs. 7.0?  There are some marginal improvements that the tech shop may admire but anything for the planning community?!

Then we also hear that the planners have not been using the Statistical modeling feature in APO.  Will upgrading to 7.0 persuade the planners to use the Stat Models more?  Not just more, just even barely?  Then I hear a pause and the IT consultant says that Stat models are not a priority given the budget constraints they have.

So more millions before and no stat models.  Now five years later, we have a shiny new upgrade and again the Stats are not a priority.

I have been preaching Usability for the past few years.

Put together fine tools  – But help the users in making the transition to the tool – give them better understanding – Make the new tool more usable!

Give them the reports they need.  Provide them an exception based workflow!

APO has good statistical models.  They will help you move the peanut forward but only if they are understood and leveraged.

We just re-launched the marketing campaign for our Usability Consulting.  Model tuning and model matching to product profiles are important elements of the Usability training. 

Once implemented the Usability project will harmonize the use of models across planners from various geographies for the same business/product family.  There will be streamlined work flow.

We help you answer the following questions:

  1. Am I using a Pareto Approach in my APO planning process?
  2. How can I leverage APO DP to improve our forecast accuracy?
  3. Why does APO mostly give me flat forecasts? How do I fix this?
  4. What are Alpha, Beta, Gamma, Sigma and Theta? How do I leverage these parameters?
  5. What is the correct level to model so as to improve the overall accuracy at the SKU level?
  6. What are weighting profiles? How does it affect my final forecast?
  7. How can I control time trend using trend dampening profiles?
  8. Are there products and customers that are better left to APO’s automated modeling strategy?
  9. Which models to choose for what family of SKUs?
  10. What are custom modeling profiles?
  11. How is APO helping us simplify and improve the promotional planning process?
  12. How do I create Multiple Linear Regression Models in APO?
  13. Are we using the system defined error metrics in APO? Why are they different from the classic MAPE calculations?
  14. How do you conduct phase-in/phase-out of products?
  15. When should I not use the Croston’s Model?
  16. Why am I getting 9,000+ alerts every morning?

Perhaps the stress test of your new implementation or upgrade will be to ask the team if they can answer the above questions.

It seems like it is an easy thing to roll out new technology, but an uphill task to make it usable.  Did I hear the name Steve Jobs being mentioned somewhere………….

Follow us to learn more on the different service offerings at http://www.demandplanning.net.

Happy Forecasting!

Mark

Tags: , , , , , ,
Posted in Forecast Modeling, Forecasting Process, SAP APO, statistical modeling | No Comments »

Slow Moving, Short History and Intermittently demanded products – How to forecast?

March 1st, 2012 by Dr. Chockalingam

Every time we come across a slow moving item, we have to spend quite a bit of time analyzing the different factors to come up with a forecasting strategy that will handle the challenge.

In one of our recent projects, slow moving items accounted for more than 25% of the SKU population.   Short history for many items limited us from using many of the statistical models, simply estimates would not be robust given the limited history length.  Now this is not unique to slow moving items.

We, at Demand Planning LLC, preach forecasting by exception.  Here we will describe our approach that leverages the exception management methodology to come up with a forecasting strategy for the slow moving items.

Let us first define a slow moving item.  Using some business rules that align with the business model, we created a rule to codify an item as a slow moving if it has not been consumed in the last four months.  This can change depending on the type of industry you operate in.  Also make sure you have sufficient flexibility when it comes to life time buys and critical parts required for infrequent break-downs.  We have also recommended to clients that they look at it from months of inventory on hand based on projected usage to define slow moving.

An objective way to do this will be to analyze the last POS or usage date.  You may also want to use the last shipment date (although there may be inventory at retail or the shop floor in the case of service shops).

Next we imposed an intersection on the amount of available history.  We sub-categorized all slow moving items based on history (long history and short history), volume (High volume and low volume) and intermittency (Intermittent and Non-intermittent).

Segmentation of SKUs by volume, historical length, intermittency

Click Here to enlarge image

Among these slow moving items, We forecasted most of the High volume-Long history items using the Forecast PRO TRAC software package.  Analyzing the R-squared and MAPE, we were satisfied that the forecasted best fit model looked decent.  We made some minor model tweaks but the expert engine did a great job at first cut without much extra effort.  AS you would have conjectured, the expert selection was good enough for High volume-Long history items.  These items accounted for only 1.5% of the total item numbers.  If you notice in the graph below, 18% of the slow moving items were also characterized by short history and of very low volume.

Click here to enlarge image

Given the predominance of the low volume items, we find models suggesting consistently one approach if they are intermittent versus non-intermittent.

  • How should one forecast low volume slow moving items?
  • What are the best practices if we notice intermittency in slow moving products?
  • How should we leverage the software tool to get the best out of for such items?

We came up with different forecasting strategies for different segments.  The segmentation approach also threw a lot of light for the business on their SKU complexity.  After a detailed business review with manufacturing and marketing, the business decided to shut off production and deployment for some of the SKUs.

The beauty of using exponential smoothing models using standard software packages is that these models learn with history and the passage of time.  When there is no historical data for the last few months, the software best fit itself suggests a zero or a near-zero forecast.  This is the first sign that should make you ask if this item is obsolete.  You may also want to create a class called Non-moving which is in between Slow Moving and obsolete.

More difficult than forecasting a slow moving item is to forecast and produce an obsolete item.  Although technology and high powered software are important, it is more important to use an exception management approach and gather appropriate business intelligence to come up with a forecast strategy.

Join us to know more about what rigorous approach we took for the rest of the slow moving items and get a fresh perceptive on inventory management of slow moving items.  We will discuss what kind of coordination is expected between operation management folks while ordering new batch of slow moving items.    This will be a new section added to our May 2012 workshop.

For more information about the content please visit: http://demandplanning.net/demandplanning_tutorialMA.htm

Contact us at info@demandplanning.net  if forecasting slow moving items is a pain point of your demand planning process.

Kudos to Rohan Asardohkar for excellent analysis and painstaking compilation of a large database of items!!

Tags: , , , , , , , , ,
Posted in Forecast Modeling, Forecasting Process, forecasting software, forecastpro, statistical modeling | No Comments »

Automatic Outlier detection – Blessing or Curse?

February 3rd, 2012 by Dr. Chockalingam

One of the puzzled questions that Demand Planners ask in our training workshops is why their software produces a flat forecast 90% of the time.  An expensive software that took an army and a couple of years to implement typically suggested a constant model or moving average model.  This resulted in a flat forecast.

Although the naked eye can graphically see (if graphs are made available to the user) a nice seasonal pattern, the expert selection in the software produced a constant forecast to eternity.  There are many tricks underlying this final result – some of them known and some of them hidden.

One of the culprits is the outlier detection process.  The software can intelligently detect outliers for a given setting and outlier detection method.  Typically you use a K-factor to develop tolerance bands around the ex-post factor to identify outliers.  What are K-factors and how to leverage the K-factor settings to produce good forecast models?

We observed in a variety of cases, people use low k-factors that then throw out all seasonal peaks and troughs.  A low k-factor is super vigilant.  It  does not allow any pattern to escape through to the modeling engine.  All the engine sees is just a cluster of a few data points that are closely scattered around the ex-post forecast or just a historical mean.

See the picture below.

A k-factor of 1 will eliminate all patterns seen in the demand profile.  It just keeps a fraction of the original data set that all point to the historical mean as a violently accurate forecast.  This has nothing to do with the power of the statistical engine available to the software.

In our up coming three day workshop we will discuss the perils of automatic outlier detection and have the attendees work through a hands-on exercise that will give better visibility to the entire outlier detection process.  We will explain the features under the hood of the SAP APO Demand planning module to navigate through this perilous process.

Day 3 will be all SAP APO with hands-on training on the software platform.  Attendees in September 2011 workshop were able to directly make model and parameter changes to their live forecasts in the workshop.

Visit http://demandplanning.net/workshops.htm for more details on the workshop.  Please reach out to me if you have more questions or want to discuss Outlier detection process in APO DP.

Tags: , , , , , , , , ,
Posted in demand planning training, Forecast Modeling, Model Diagnostics, SAP APO | No Comments »

Moving the ROC to forecast…….

January 6th, 2012 by Dr. Chockalingam

We hear quite often, that forecasting is a waste of time.  People often cite the weatherman and laugh at forecasts and forecasting as a productive process.

As long as you don’t leave your demand forecasting to the weather man, I believe we can do very well.  Most supply chain problems originate by ignoring the forecasting that is happening through out the organization. In a survey I remember reading a couple of years ago, on average 50% of the people in an organization were forecasting something or other.

If the forecasting process is bad, fix it! You ignore and move on at your own peril!

It is simply impossible to ignore the forecasts, because the ROC (Rest of the Company) is hard at work forecasting something or other.
Even folks in manufacturing who can badmouth forecasting may be using an average run rate of some sort to determine their inventory calculations.  This run-rate determination is actually a forecast, the average, although they think it is no forecast.

If there is a reasonably good demand planning process installed in any organization, we can establish this will easily beat out the “run-rate” or any other average hands down.

A good demand planning process does not just rely on statistical modeling.  It leverages the information from the various players in the rest of the company.  My good friend and my manager at Schering Plough coined the term ROC – Rest of the Company.

The ROC has information.

The ROC also forecasts.

ROC forecasts for different purposes.

For us demand planners focused on the supply chain and mired in the daily fires of the live order stream and its deployment, we only think of the supply chain plan as our reality.   We wonder what does the ROC do other than feeding us some useful info to make the widgets we need to demand plan and produce.

Inventory is a problem but is only one of many problems!

Organizations have a variety of challenges and constraints to solve so they can thrive and grow. Organizations need to plan for the medium to long-term and manage the business accordingly.  At least 50% of the ROC forecasts but NOT necessarily for inventory purposes.

Senior management needs to forecast an EPS for investors and need to hit it within a reasonable threshold. Companies need a long-term forecast to assess what they need in capital investment and where and how to build the facilities for expansion. Even HR needs a forecast.

Thinking every function will be forecasting for the supply chain is like the Dilbert Cartoon “Sure – I will drop everything else and will focus on your problem.”

So forecasting and planning is embedded in various functions and various forms through out the organization and is unavoidable. You cannot tell people it is a real problem so they should stop it! The key is how to leverage the forecasting responsibility and accountability already installed into a holistic process that can let you piggy back and obtain a supply chain forecast for your short-term and long-term planning.

Ignoring the corporate forecasting machine and creating an isolated forecast or an inventory deployment algorithm is a sure way to significant troubles – what we at Demand Planning LLC caution as the fragmented planning process or the lack of the often glorified “S&OP” process.

So in summary, there is no issue with the ROC or 50% of the ROC involved in forecasting.  The real problem is when supply chain decides to ignore the forecast or the forecasting process that is already etched in the ROC(k) and decide to move on in isolation.

Demand Planning LLC does use and recommend advanced algorithms for demand forecasting and leveraging customer input.  But that is only half our story.

We work with Sales, Marketing, Supply chain and Senior management to drive a holistic process to leverage demand information and build forecasting processes that are used across most of the organization.

We salute the hero who decides to use a demand forecast that is only 50% accurate rather than ignoring or side-stepping the forecast.  Accuracy of the forecast is secondary.

Start with the forecast first and make plans and contingencies for the extent of the error.   Improve on the forecast by quizzing, dialoguing, negotiating and working with the ROC.  This should be well emphasized in the basics of supply chain management.

Join hands together to move the ROC to make a better demand plan!

Tags: , , , , ,
Posted in Forecast Modeling, Forecasting Process, S&OP, S&OP Best Practices, Sales, Uncategorized | No Comments »

Questions on Demand Planning Training Workshop in India

December 13th, 2011 by Dr. Chockalingam

In response to our workshop announcement, we have received several emails with number of questions.  Given that we have only experienced the tip of the iceberg when it comes to demand planning and supply chain careers in India, the opportunity set is explosive for a knowledgeable and skilled demand planning professional.

Demand Planning is one of the unique areas where the individual can either move up in Marketing, or Finance or into senior cadres in Supply Chain.

Why is this a very important career enhancing workshop to attend?

Demand Planning is a highly demanded skillset for a lucrative job in the supply chain industry in India or abroad.  This course is aimed at practitioners and newly aspiring professionals who want to enter the supply chain industry.

Attendees who complete the two day course and all individual and group exercises will be awarded a certificate of completion.  This will be the pre-cursor and required workshop to our certification program to be launched in 2012.

Get skills you can use at work

We will explain and demonstrate best practices in model selection, illustrate how to improve model quality, and teach you how to leverage the forecast measurement process.

Learn from industry experience

We will bring practical examples from our consulting experience.  We have consulted with clients in Consumer goods, Food and Beverage, Chemicals, Pharmaceuticals, Heavy Manufacturing, Aerospace, Medical Devices, Oil and Gas etc.

Network with peers

You will have the opportunity to meet, interact, and learn from other demand planning professionals with team challenges and networking exercises.  Attending this workshop will introduce you to our vast network of supply chain professionals and career opportunities in North America and Europe.   This is not a job offer, but this education is aimed at providing a skillset that is in short supply through out the world.

Add to your credentials

Upon completion, you will be awarded a certificate of completion from Demand Planning LLC, attesting to your newly-acquired skills in Demand Planning and Forecasting.  You will also get an opportunity to obtain additional resources to earn a certification in the future.  Program is in the works to be announced shortly.

Does the two day workshop in Bengaluru contains topics related to forecasting of rarely used spare parts or not?

Yes we talk about models to forecast and stock for rarely used spare parts otherwise known as intermittent demand.

While the focus is on forecasting and modeling, our approach is to marry up the business problem with the technology.

Look forward to seeing you in Bengaluru, Bahrain or Boston!!

Review the course at http://demandplanning.net/demandplanning_seminarIndia2.htm. 

We have collaborated with Knowerx India to take rupee payments if you can only register and pay in local currency.

Happy Forecasting!

Posted in Uncategorized | No Comments »

Demand Planning.Net announces the Training workshops for 2012

November 14th, 2011 by Dr. Chockalingam

Demand Planning.Net is pleased to announce the following dates for the Demand Planning and Forecasting workshops in 2012.  If you are preparing your training budgets, you may want to plan for these workshops for your new hires as well as experienced professionals that are new to forecasting or just as a refresher for those who know the theory but need a hands-on perspective and experience in forecasting.

Demand Planning and Forecasting 2-Day Hands-on Workshop at three locations:

Bengaluru, India Jan 27-28, 2012 at the Chancery Pavilion, Bengaluru India. Learn more…
Bahrain, Mar 12-13, 2012 at the Crowne Plaza Hotel for Middle East clients Learn more…
Boston, MA May 23-24, 2012 at the Four Points by Sheraton Boston/Norwood. Learn more…

This will be followed by the Fall 2-day seminar in September 2012.  There is a possibility of a one day workshop on S&OP during the Summer in Burlington, VT.

We also are offering the one day modeling and metrics workshop that focuses on SAP APO DP.

SAP APO add on 3rd Day Workshop available in Bahrain on March 14, 2012 and Boston on May 25, 2012. Learn more…

Please review http://demandplanning.net/workshops.htm for details on all of the above training workshops.

 

Tags: , , , , , , ,
Posted in Demand Planning India, demand planning training, Forecast Modeling, Metrics, Model Diagnostics | No Comments »

Draper Team Aims to Improve Predictions Through SPADE 10/05/2011

November 7th, 2011 by Dr. Chockalingam

Press Release from Draper Laboratories

CAMBRIDGE, MA – Draper Laboratory is leading a team that aims to improve long-term intelligence predictions through software that weights most heavily forecasts from analysts who tend to be most accurate in particular fields like politics, world events, and economics.

The Intelligence Advanced Research Projects Activity (IARPA) is funding the effort through the Aggregative Contingent Estimation Program (ACE) with the hope of finding the most precise and timely way to crowd-source its predictions amongst its widely dispersed analysts.

Draper is leading a team named SP♠DE – the System for Prediction, Aggregation, Display, and Elicitation. The SP♠DE technical team is led by Dr. John Irvine, capability leader for information and decision support, and Dr. Sarah Miller, a senior cognitive science researcher. The team includes Drazen Prelec, a professor at the Sloan School of Management, as well as the Departments of Economics and Brain & Cognitive Sciences, at the Massachusetts Institute of Technology; Alexander Kirlik, professor and acting head of the human factors division at the University of Illinois; Dan Martin, director of MRAC, a psychology research and consulting firm; and Bill Welch, director of the Center for Intelligence Research Analysis & Training at Mercyhurst College in Erie, Penn.

While forecasts made a few days or weeks before an event are often accurate, the challenge is far greater to look long term in order to give policy makers the best opportunity for planning, said Stuart Peskoe, associate director for mission systems in Draper’s tactical systems group and SP♠DE project manager.

The SPADE team will be enhancing Prelec’s method of Bayesian analysis that scores most highly those analysts whose predictions exhibit a surprising level of mutual consistency. Miller and Kirlik have also studied how best to take advantage of the strengths of experts and algorithms while compensating for the weaknesses of both –and successfully tested their concept with fantasy baseball predictions and actual Major League Baseball results.

The team is looking for those with interest or expertise in economics, politics, culture, and global security to participate in the study via an interactive website at www.iSpade.net.

Draper Laboratory is a not-for-profit, engineering research and development organization dedicated to solving critical problems in national security, space systems, biomedical systems, and energy. Core capabilities include guidance, navigation and control; miniature low power systems; highly reliable complex systems; information and decision systems; autonomous systems; biomedical and chemical systems; and secure networks and communications.

For more information, contact www.ispade.net

Tags: , , , , , ,
Posted in forecasting software | No Comments »

Demand Planning in the Middle East – Training Workshop

November 4th, 2011 by Dr. Chockalingam

For the first time, we are bringing our hugely popular two day workshop on Demand Planning and Forecasting to the Middle East at the request of many Saudi and Emirates based Forecasters.  We have received many requests for this popular two-day course in Demand Planning and forecast Modeling.  We will also provide you with a forecasting software for a limited trial to explore the power of statistics in Forecasting.

March 12th (Mon) and 13th (Tues), 2012 – 2 Day Interactive Workshop in Manama, Bahrain at The Crowne Plaza

View and print seminar brochure (PDF)

Register Now! – Regular Price $995
Early-Bird pricing $895 until January 20, 2012

In this specialized two-day course, we will explain the modeling methodology and process behind accurate demand forecasts.

We will also illustrate how to effectively use promotional information to arrive at an event based forecast. The focus will be on demand modeling using statistical techniques, the methodology to perform model diagnostics, forecast accuracy measurement and the process to incorporate market intelligence.

If you are a new demand planner looking to enhance your knowledge of business forecasting, you cannot afford to miss this opportunity!

View more information at

http://demandplanning.net/demandplanning_tutorial_Bahrain.htm. 

I look forward to facilitating this workshop in March in Bahrain and to meeting many of the colleagues based in the Middle East and Europe.

There is also a one day workshop on Modeling and Metrics in SAP APO on Wednesday.   For SAP APO planners, it is ideal to take the three days as one combined workshop.

Please contact us if you have any questions.

 

Tags: , , , , , , , , , , , ,
Posted in demand planning training, Forecast Modeling, forecasting software, Metrics, Model Diagnostics | No Comments »

« Older Entries