page contents

If we had the ability to accurately predict future demand and supply, we would have no need for any safety inventory, or buffer stock.  In reality our customers do not read our forecast.  They order the quantity they need for each item, and dictate when they require delivery.

Here I discuss the concepts of safety stock and safety time.  Safety stock is a fixed number, and safety time is a variable number based on forward looking seasonal trending demand.  Safety buffer is not an isolated and independent standalone feature, it is an integral process within an overall Demand Planning solution.  Demand Planning includes forecasting, replenishment, and distribution planning for retail, wholesale, distribution, and manufacturing companies.  All fundamental elements of effectively managing the supply chain.

Please bear in mind that in this volatile world we operate in, frequent, at least daily replanning is likely to deliver greater available inventory performance, with higher on-time delivery service levels to customers, as apposed to relying on the most accurate forecast and safety stock buffer we can calculate.  Daily we know what demand orders we receive and shipments made by item and quantity, in addition to supply receipts of raw materials, components, and packaging.  Knowing our on hand inventory available balances, the software application determines if we are running at the planned demand rate, or over or under projected demand.  Key to all this processing is keeping supply and demand in balance.  If within North America you source products from distant countries like China, you must contend with long cumulative lead times, assuming product is moved by sea, and then rail, or truck, but not air freighted.  This makes planning, scheduling, and vendor collaboration significantly more critical than operating in a Just-in-Time (JIT) process close to your suppliers.  Understanding long term demand is vital to your business, and critical to customer on-time service delivery goals.

In a business that is promotion driven, it is essential to identify event “spikes” and to normalize history.  Unless promotional activity is an ongoing everyday way of running the business (everyday low prices?), we must separate out regular demand from promotional activity by using multiple demand streams, one for regular, and as many as required for each different (BOGO—buy one, get one x% off) promotion.  When looking back in history and seeing these occasional spikes when promotional activity occurred, unless we can accurately remove these spikes, it will make the job of the forecasting engine less reliable.  With an unreliable forecast how do we plan, and what is the point of measuring actual sales to an inaccurate forecast?  I will not get into detail at this juncture, suffice to say that without addressing this issue, when processing data to determine forecast to actual deviations, the differences calculated may be over stated if we did not remove this promotional detail prior to evaluating history or generating a forecast.  Future promotional activity should be added to forward looking “regular” demand with separate demand streams for each promotional event, identifying quantity, timing, and expected lift.  Where cannibalization may take place, promoting 1 gallon milk that will impact the sales of ½ gallon milk, requires appropriate adjustments to be made to affected items.

The planning solution must be able to assign safety buffers depending on the ABC inventory classification, or Pareto 80/20 principle.  The critical few products versus the additional many.  The high volume, high value, “A” items would carry a lower safety stock or time value than a low value, low volume, “C” item.  Then too, you may wish to associate this classification with XYZ analysis, where “X” items are high frequency demand, and “Z” infrequently demanded product.

In using a best-of-breed niche Supply Chain Management (SCM) solution that adds value to your Enterprise Resources Planning (ERP) system, when interfacing, it is preferable to download data in its original daily detail.  This provides the user with the greatest flexibility to be able to work in days, weeks, months, quarters, and years.  An added benefit is how you assemble your planning calendar and supporting data if you use 4-4-5 weeks per quarter, or some other variation (4-5-4, 5-4-4, 13 X 4).  Then too this allows maximum flexibility to manage according to your, or your customer’s financial year start date.  Depending on your industry, it may not make sense to forecast in days, but you have the flexibility within your Demand Planning solution to aggregate to weeks or months.  If required to prorate back to days, you use valid historical daily information.  Subject to the seasonality of your business, and the requirement to be flexible to customer requirements, the user community can help determine the best approach to plan looking forward.  What you really do not want to do is to aggregate data in your ERP system, and then within your Demand Planning solution divide the data by 30.2 (or something) to get data into days from months.  In performing this calculation you lose the trending information, end up smoothing the daily data, resulting in invalid daily results that is now used to build an incorrect planning and scheduling strategy.  This process too will result in abnormally large requirement for safety inventory.  Think a moment about beer sales.  These take place mostly on a Friday or Saturday, and deliveries are planned ahead of peak sales.  If this data was aggregated to a week, you immediately lose visibility of this seasonal sale within the week.

We need to consider the hierarchical level we plan at for safety inventory.  If we have the same item that we supply to numerous customers, then we plan in aggregate at the item level.  If we have a number of distribution centers (DC) or warehouses, then plan at the item by DC level.  These concepts of planning safety stock and safety time are more important if we are dealing with Make-to-Stock (MTS) product.  Make-to-Order (MTO), Assemble-to-Order (ATO), Engineer-to-Order (ETO), and Purchase-to-Order (PTO) may require safety inventory at a lower level in the Bill of Material (BOM) hierarchy.  We will not get into this level of discussion here.

We may be tracking different demand streams within our system, both orders received and shipments.  Generally we are more interested in tracking orders received against the orders forecasted.  Forecast predicts demand, and as the orders are received we consume the forecast.  I do not plan to get into esoteric discussions about orders consuming future orders, or what occurs when the orders do not materialize and how we handle past forecast order shortages.  That is a discussion for another time.  Realistically some client companies only track shipment data to be able to plan with, as they did not keep both demand streams including predictive orders.  In those situations we have little option but to work with shipment data until we can build a history of forecasted and actual orders.

There may be other challenges to address.  In striving to generate an accurate forecast we require a minimum of 3 years history.  That requirement is to help determine trend and seasonality.  With only a single year of history it is totally impossible to tell if we have seasonality in our data.  With 2 years history, we may not be too sure if the seasonal trends are a remarkable coincidence.  With 3 years history we have a lock on seasonality, and a more accurate view of trend.  Please do not remind me that Easter is celebrated on different dates each year, I know, but there are techniques to manage those situations as well, as with the Chinese New Year.  Starting with the most accurate forecast possible, minimizes the need for safety inventory.  Any excess inventory is generally regarded as an expensive financial and overhead waste.  The ultimate goal is to strive for very high levels of on time delivery to customers, with the lowest level of inventory possible to satisfy demand, and result in greater productivity and profitability.

I would be remiss if I did not talk about fashion merchandise, especially those products that have a one season shelf life.  Recognizing that businesses do not radically change their type of merchandize from year-to-year, there are techniques within the applications to link last years’ unique lines with this years’ fresh and new offering, and continuing the look-back process to prior years.  Here we are using a supersession technique whereby we link the prior year orders or shipments to help forecast the new season.  It could get more complicated by addressing a brand new line that is likely to sell like a different but similar line we sell.  This gets into an understanding of using a “curve” that reflects how the like item is sold by percentage contribution over time, reflecting the short term trend, and applying the curve to the new item together with a speculation of what we believe should sell in the coming season, providing us with the 100% factor that is apportioned by day or week.  This technique is also valuable if you are rolling out a new line to retail stores and need to plan for a ramp up in sales prior to reaching a plateau of regular sales.  To reemphasis the key point, attaining a usable 3 or more year history of orders or shipments may not always be straight forward, but is not an overwhelming challenge for our starting point to generate a valid forecast.

The need to apply supersession concepts is valuable when you manage engineering changes to product as you enhance function, improve aesthetics, increase safety, or make any other engineering or marketing improvements.  Each change will call for a new item number, and this is a consistent strategy when you have a change in form, fit, or function.  The change process must be managed professionally through an ECN (Engineering Change Notice) committee.  The new item does not have associated history, so here to we link new items to the replaced products.

There are at least two approaches to developing safety stock requirements that are based on information about the variability of the forecast.  The simplest approach is to determine an average demand over a period, calculate the actual to forecast differences, calculate a standard deviation for that period, calculate the lead time in days to replenish the item, apply a safety factor with the desired level of customer service, and determine a buffer stock.  No, I did not provide the mathematical equation.

A better method is to use information within the forecast model itself to develop a statistically valid measure of the distribution of errors that are inherent in the model.  This is a complex calculation but one that is readily performed within the forecast engine and is applicable to any of the forecasting methods selected by the engine.  Here we need to assume that your forecasting engine uses the acceptable univariate time series techniques including: Arima, Auto Arima, Auto Regression, Croston, Exponential Smoothing with Trend Additive, Exponential Smoothing with Trend Multiplicative, Exponential Smoothing without Trend Additive, Exponential Smoothing without Trend Multiplicative, Auto Exponential Smoothing, Simple Moving Average, Weighted Moving Average, and Holt Winters Smoothing.  Stressing another important point for the user community is where they are knowledgeable, to be able to change the parameters associated with each of these algorithms to best suit the environment that users operate in.  Companies are not all the same.  Most have different selling seasons, different customers, different suppliers, and different management goals, that all require satisfying specific metrics of customer service levels, among others.  If the user does not have an advanced statistics degree, leave the settings alone, and use the forecasting engine as a “black box”.  Forecasting engines will provide the user for each item with the Mean Error, Mean Absolute Error, Mean Percentage Error, Mean Absolute Percentage Error (MAPE), and Mean Absolute Scale Error.  When running a tournament to decide on the best algorithm to fit a specific demand stream for each item, the lowest MAPE is the deciding factor.  The selected algorithm together with its best fit chosen variables uses current history for that item to project demand.

The requirement is to utilize the safety stock calculation inherent in the forecasting engine.  This calculation is performed on the basis of three input values:

  1. The error distribution of the forecast model for the item in question
  2. The lead time for the item in question
  3. The percentage service level desired for the item in question

Input 1 is provided by the forecasting engine for every forecast it produces.  Inputs 2 and 3 are settings established by the user and reflect the current state of affairs in the supply chain (lead time) and management policy (service level).  The output of safety stock quantity will be in units, but the user has the ability to convert safety stock unit quantity to any equivalent UOM (unit of measure), the monetary value of cost, sales, margin, or the safety time equivalent.

Every forecast will have a unique error distribution and therefore a unique safety stock for any given lead time and any given service level over that time span.  The service level specifies the percentage of the maximum possible demand that you will be able to cover.  Although safety stock increases in proportion to both lead time and service level, the safety stock increases at an alarming (exponential) rate as the desired service level coverage approaches 100%.  Typical service level targets do not exceed 95%, even in situations of most reliable forecasts.

Another useful concept in managing safety stocks is that of “Safety Time.”  Safety Time is a measure of how many forward periods (days, weeks, months) of forecasted demand a given safety stock covers.  It is generally stated as a percent of future demand.  The computation should be based on the forecast rather than the average observed demand (backward looking).  This equivalent unit of measure for safety stock provides an alternate perspective from which management can assess the adequacy of the buffer their current safety stock provides.  The greatest benefit with this forward looking safety time approach is that as we enter a selling peak, the safety buffer is increased, and conversely if we are marching into a slower period of sales, the buffer reduces.

Once we have the best possible forecast with a required safety buffer, that information is used to calculate projected shortages over time by item, or item by DC.  Planning out a year or two is not unrealistic.  This planning and scheduling netting process provides three key replenishment data streams: purchasing, manufacturing, and distribution center transfers, depending on the type of operation the client customer operates in.  This planning detail is used to communicate with suppliers, production, and logistics so that they can plan their operations effectively in lock step with corporate plans.  This collaborative visibility may be updated daily or weekly depending on the cumulative lead times and the operational need to be responsive and flexible in meeting customer demand.  Appreciate too that if the actual orders come in a higher demand rate than planned, the planned orders will move into an earlier time frame.  The corollary applies: Orders coming in at a lower demand rate will push replenishment orders into a later time frame.  In terms of inventory investment within a distribution network, it is best to have safety inventory held at the hub location with the understanding that the company has maximum flexibility to move inventory to the spoke locations.  Backhauling inventory is never a productive or cost-effective way to manage inventory.

In generating a forecast, and measuring actual to forecast is a total waste of time if management does not examine results.  Why is our forecast outside of some predetermined tolerance level?  What is happening in the market that has caught us by surprise with either greater or lower than expected demand?  Measuring everything at a granular level may be too onerous, so studying results at various aggregated levels may be appropriate.  Did you configure your solution applying Demand Segmentation?  Learn more at this blog or video.  Safety inventory is there to protect a company from unwelcome surprises, and if we experience them, what was the cause?  Metrics often play a role.  Management requires sales personnel to develop a forecast.  Sales might earn bonuses if they overachieve with significantly higher order quantities.  That causes chaos if we need to scramble to get additional supplies.  Buffer inventory will not protect the operation in this situation.  Customers may begin shopping the competition knowing that they can be assured of getting high quality product, in the correct quantity, and delivered on-time.  It is preferable to penalize sales for over achieving.  If it is not in the forecast, sales will not get credit or commission for the sale, and will not count toward any sales quota or target recognition.

We need to only allow forecast adjustments outside of the cumulative lead time.  If you tolerate adjustments up to the last minute, people will game the system to show 100% actual to forecast accuracy.  That takes responsibility for effective communication and collaboration out of the process.

An additional point to stress here is that we strongly recommend a demand-pull philosophy.  Operating in this sphere requires that we start with the potential customer or consumer demand and plan backwards to purchasing or production, and if applicable through distribution.  The opposite approach is following a push strategy where we purchase or produce product and shove it into the market hoping and praying that what we have available will actually sell.  This latter strategy is not the way to profitability for any business.

Since we hold people accountable for results, we need to allow users to override the system generated forecast, safety stock, or safety time.  It is helpful to record results with and without the overrides.  If the system shows that it was better able to calculate demand than with an intervention, recommend uses stop second guessing the system.

Reporting forecast accuracy is one helpful tool, more especially if we rank it in descending order of actual to forecast deviation accuracy.  The other is using Waterfall or Slant Charts.  Here we show the aggregated monthly or weekly forecast by item over a 12-month period and reflect actual orders.  In this way we could see, say, 12 months out what was forecasted, and once we reach that point in time to see the actual sales.  At a minimum we should be reviewing the performance at the cumulative lead time for each item.  With Waterfall Charts we need to see the performance of actual to forecast in units, cost, selling, and margin currency.  An additional report that is helpful to management is to aggregate all safety inventory.  This can be further divided and presented by a variety of inventory classifications to help determine if the approach management is taking is affordable and provides a cost-benefit to the required investment in additional inventory versus increased sales.

If we wish to move into a more esoteric realm, we can introduce sophisticated planning for suppliers, and use their actual performance to deliver quality product, in the correct quantity, on time.  However, it is preferable initially to track their performance based on collaborative data shared with them, and use a veiled threat that you will source from a different supplier if they cannot support your build or purchase strategy.  This may be difficult to achieve if the supplier is a division of your corporate company.  As always when we are measured, and recognize room for improvement, and get performance metrics presented on a regular on-going basis, suppliers will quickly understand that they are an extension of your business and must take their responsibility seriously.  Executive to executive communication helps to communicate the seriousness of shared responsibilities.

A final comment: If as a company you are managed using a Sales and Operations (S&OP) executive management process, it is probable that your performance will elevate your company to one of the excellent few within your industry.  Having executive level supply and demand issues addressed, resolved, and communicated on a regular, at least monthly basis, goes a very long way to ensure everyone in the company is playing from the same musical score.

Recollect the cliché that the worst level of customer service possible is to not have the available inventory when the customer requires it.  And contrast that with having too much inventory that becomes redundant inventory with all the added value that must now be scrapped or sold at a fire sale.  It is worth making the supreme effort to develop the best forecast possible, and add a little buffer stock to ensure great on-time customer service levels of quality product when required.  That supports a greater productivity and profitability strategy.

Comments are closed.