The ultimate guide to safety stock

When it comes to inventory optimization, for many people the first and sometimes only focus is safety stock. This is not unnatural, but actually one of the first things to understand about safety stock is that it is not the only show in town. Neglecting other types of stock, such as cycle stock, pipeline stock, anticipation stock, congestion stock, and so on, is a mistake.

But our topic for this article is safety stock. Let us start with a definition:
Safety stock is inventory kept to buffer against uncertainty on both the demand and supply side.
The value of safety stock is in allowing supply to continue uninterrupted most of the time. The downside of safety stock is that it can lead to excesses and eventually obsolescence.

So far so simple, but there are different ways of defining the amount of safety stock that you need. We will look at some of the most common ones in this article. We will also consider how often you should revise safety stock levels, where in your supply chain you should keep it and what alternatives to safety stock exist.

To start, though, we will differentiate safety stock from contingency stock and discuss how different types of inventory require different safety stock models.

Safety stock and contingency stock

First of all, let us differentiate between safety stock, which is designed to buffer against typical uncertainty, and what we shall call contingency stock, which is designed to buffer against unusual events. This is particularly relevant at the time of writing, since the impact of the Covid pandemic has been much more substantial than the regular uncertainty you get from the occasional supplier shipment being delayed or the usual fluctuation in customer demand.

Do you need contingency stock?

Contingency stock is appropriate in two scenarios.

The first is where your inventory is absolutely essential, i.e. related to the preservation or protection of life. Hospitals and the military are examples of organisations that routinely hold contingency stock: they need to be prepared to withstand substantial disruption on the supply side as well as major swings on the demand side.

The second is where you are trying to assure business continuity in the face of a specific supply-side risk. For instance, if there is only one possible source of supply for a key input material to your products and you have concerns about your supplier’s reliability, you might choose to increase your inventories of that material in case there is a disruption.

The right-sizing of contingency stock is a type of risk management and closely related to business continuity planning. You have to assess the level of protection you want to have, for instance, the ability to continue without any supply for 6 months, or the ability to handle a sudden doubling in demand (eg for flu vaccines), and so on.

You can never protect yourself against all possible scenarios (what happens if supply is disrupted for 12 months? For 24 months? What happens if demand triples or quadruples?) and you also have to weigh the risk of running short against the waste involved if you never need your contingency stock. This is why contingency stock is only appropriate in very specific circumstances, in a lot of which a public or governmental organisation will directly or indirectly be picking up the bill.

One size does not fit all

For the rest of this article, let us focus on safety stock itself. When working out how much safety stock to hold, there are 3 main considerations to factor in before you even start calculating. The first is the nature of your demand, the second is the service level you are targeting, and the third is the planning methodology you use.
  1. The nature of demand. Demand can be uniform or variable, flat or nonstationary (i.e. growing or declining), regular or sporadic, predictable or unpredictable, and in terms of probability distributions it can demonstrate normal distribution or other types of distribution such as gamma, poisson, or negative binomial. Noting that whatever it is at the moment, it can change at any time. There is a lot of complexity here. The key point to understand in a first instance is that tools and models designed to define ideal safety stock levels are highly unlikely to factor all of these dimensions in and so will work better for some items than others
  2. Service level. Service level is often poorly understood in this context, but is a very important input to a lot of safety stock models. It represents the likelihood that stock is available when required. Mathematically, only infinite stock will deliver a fill rate of 100%, but organisations can target a specific fill rate (frequently in the 95-99% range) and build safety stock to try to assure that level. Read our article on supply chain service levels for a deeper dive
  3. Planning methodology. There is an important difference between deterministic and replenishment methods. In a replenishment model, you define the required safety stock to deliver a given service level for expected average demand and variability. With deterministic models you are planning on the basis of given demand. As such, the concept of safety stock is in theory redundant, although in practice deterministic methods (such as MRP) still have to deal with some level of variability. A common cause of overstocks is when organisations size safety stock as if they were using a replenishment model when in practice they are planning deterministically, often to a forecast
The three considerations above are important factors in segmenting your inventories. Before trying to optimize your safety stocks it is important to understand what different types of inventory you are dealing with. Trying to use the same model for calculating safety stock across all items is highly likely to lead to sub-optimal outcomes.

Different methods for calculating safety stock

To define what level of safety stocks you need, a number of methods are possible:
  • Safety stock based on cover. This is where you define safety stock in terms of estimated average demand for a given period. For instance, your policy might be to try to maintain 2 weeks’ safety stock of all items. The only benefit of this method is its simplicity to understand and, no doubt for this reason, it is very commonly found. However, it is almost certain to lead to shortages and/or overstocks since predictability, demand variability and supply lead times are highly unlikely to be the same from item to item
  • Safety stock based on maximum expected demand over the lead time. This is where you consider the maximum demand over the lead time and use it to define your safety stock. Typically, historical data is used to ascertain what the maximum demand ever experienced over the lead time is. Proponents of this approach like the fact that it is non-parametric (i.e. it does not seek to apply a specific statistical distribution), but this is also its weakness, since working out what proportion of the maximum to use requires deep statistical understanding, unless you just use the maximum itself (i.e. safety stock = Maximum lead time demand – Average lead time demand) but then you are in theory delivering 100% availability which will be excessive in most situations, especially if you have outliers in your historical data
  • Safety stock based on service level. This is where you define the desired service level and then apply a statistical distribution to define the safety stock. A benefit of this method is that, unlike cover, various service level metrics, such as fill rate, can reasonably be set as standard across items. For instance, targeting 99% fill rate for all A items. The service level acts as a proxy for optimizing based on cost, but avoids a lot of the work calculating the actual costs, which can be very labour-intensive or imprecise to do. It also allows you to set higher service levels than a simple cost optimization might deliver, which is a strategic choice favoured in many industries. The main disadvantage of this approach is that many professionals are uncomfortable dealing with the statistics involved. However, most inventory optimization technology uses this approach, so in our view it is well worth making the effort to explore it further
This is a high-level overview of the main methods. We will now look a little deeper at the third method, defining safety stock based on service level. While not perfect, we find optimization based on service level most commonly to be the best option, especially when applied at scale. This is also no doubt the reason why it is the approach used by almost all inventory optimization software. We will also look at some of the major exceptions where it may not work well.

Defining safety stock using service level

To define optimal safety stock using service level, you can use one of a number of equations which use well-established mathematics. Let us break down one of the most commonly found equations to understand how it works and what some of its important sensitivities are.
Safety stock equation

SS = Safety stock
Z = A measure of how many standard deviations above or below the mean a value is
σ = (Sigma) a standard deviation from the mean
Vol = Average volume of demand in units
LT = Lead time (time between ordering and receiving an item)

What the equation does is define the amount of safety stock you need relative to three things:

  1. The lead time. This is intuitive. If you can get something the next day you will need less safety stock than if it will take 3 months to arrive
  2. The absolute variance in demand. So if demand fluctuates between 100 and 200 units you will need more safety stock than if it fluctuates between 100 and 120 units
  3. The target service level you want to achieve. This is done by considering all of the likely values based on the observed distribution of demand and then defining what percentage of those you want to cover
The example given above only allows for variability on the demand side. If you also want to factor in supply-side variability, you can use a modified version of the equation:
Safety stock with lead time equation
This isn’t a maths lecture. The thing to retain here is that this version is also factoring in how much the lead time itself varies.

So in summary, if demand, variation in demand, lead time, variation in lead time or your desired service level increase, you will need more safety stock. If any of them decrease you will need less. So much you probably instinctively know. What the equation does is help you to quantify by how much. Other equations exist, this one was chosen as it is fairly commonly found and relatively easy to understand.

Exceptions and special cases

However, there are some important sensitivities to this equation. Firstly, it uses cycle service levels. (See our article on supply chain service levels on this point, especially the section at the end on calculating safety stocks using service level.) Secondly, it assumes normally distributed demand. This is frequently good enough, but especially where demand is very variable and/or sporadic it may not be robust. Thirdly, it assumes average demand to be flat (stationary), not increasing or decreasing. There are further sensitivities, but already you can see how using a well-known equation has a number of pitfalls which could lead to inaccuracy in your modelled safety stock levels, and that it is likely to be problematic for certain types of inventory, especially those with sporadic demand.

In short, the equation given works reasonably well in a replenishment model for items which have moderate variability. If you are managing items with very sporadic demand (eg spare parts that are only infrequently needed) then it is unlikely to work as well. More complex models exist for sporadics, but in a first instance a maximum lead time demand model might be a good safe place to start.

Similarly, care needs to be taken with demand which is clearly growing or declining. This too is a more complex area, but in short, two workarounds are either to factor in the smoothing constant from your forecast or to make a series of calculations at the projected future levels of demand.

Optimizing your safety stocks - a systematic approach

So what is the best approach to optimize your safety stocks? Well maybe a good place to start would be in measuring how much you have, compared to what you think you should have. Another way of defining safety stock is the average stock on hand when a new order arrives.
Safety stock is the average stock you have on hand when a new order arrives
Since actual demand rarely matches forecast demand, there will usually be a difference between your actual safety stock at the end of each cycle and your target safety stock. (The whole purpose of safety stock is, after all, to buffer against this variability! If in your mind safety stock should never be used then you are thinking about sediment, which is surplus, or the level of inventory never used.)

However, over time and on average the actual should be the same as target unless you have forecast bias. If you consistently have less safety stock than planned then there is a case to be made for increasing it, and if you consistently have more then you should consider decreasing it. (Or, if you never run out, a case can be made for reducing your planned safety stock!)

This doesn’t address how much safety stock is optimal, but it does at least allow you to cancel out any forecast bias, is simple to compute and may allow you to make some improvements quickly, by increasing or decreasing your target safety stocks where you observe systematic understocks or overstocks.

Another step to consider is what we call changing the givens. We saw above how lead time, demand variability and service level are all key inputs to any optimization calculation. Anything you can do to reduce lead times and demand variability, even to relax service levels in some situations, will reduce your need of safety stock. You still need to calculate how much you can reduce it, but all of these steps certainly help and in some instances are addressed in advance of any attempts to optimize safety stocks since they bring other benefits.

The next step is to segment your inventories. When implementing a new approach to safety stocks it is unadvisable to switch everything over in one go. Much better to trial with a smaller number of items and watch carefully what happens before rolling out further.

There are many ways to segment your inventory. A good starting place is an ABC analysis, whereby A items account for the top 80% of your inventory turnover (value * volume in a year), B items the next 15% and C items the remaining 5%. As a general rule, set C items aside. By definition they tend to be rarely used, but this doesn’t mean they can’t be critical when required. As such, it is often best to set a conservative safety stock for C items to avoid stock outs, which are likely to be more costly than the value of any benefit from optimizing them.

Amongst the A and B items you then want to identify items that have regular demand (i.e. demand in all or most time periods) that is not growing or declining markedly, and with moderate variability. (For the statisticians out there, variance coefficient is a good basis for carrying out an XYZ analysis on variability. Moderate variability would be X and Y items.)

A and B items with moderate variability are the best ones to start experimenting with. You can calculate optimal safety stock levels for these items and compare them to what you are using. What you are likely to find (based on our experience) is that you need to increase safety stock for a small number of items and can reduce it for a much larger number of items.

Before you rush in and change anything, however, there are a number of further steps to take. Maybe start with the items which show the biggest potential to change. Take the top 20 to 50 items by usage or potential, for instance. Then double check the input data to your calculations. Important parameters like lead time can frequently be wildly inaccurate (or even missing) in source data.

Once you’ve satisfied yourself that the data is as accurate as it can be, then try moving towards the suggested safety stocks for those items incrementally (i.e. bridge the gap between actual and target a little at a time, don’t jump straight to the calculated value). All models have a certain level of inherent inaccuracy, plus demand is never quite what you think it is going to be.

Once you have successfully tested this with your top items, gradually expand it to the rest of the stable ABXY items. Then you might want to consider more complex approaches for the rest of your inventories.

Doing this exercise may require you to try a replenishment model for the first time. Remember that deterministic planning in theory shouldn’t require safety stock. And in practise, if you’re using one of the deterministic MRP methods and planning to a forecast, then the safety stock calculations will not apply in the same way.

Safety stock and MRP

We sometimes see it said that replenishment isn’t even a planning method. Opinion can be very polarised on this topic. It really comes down to how predictable future demand is. With replenishment models, the way of thinking is “we don’t know what demand is going to be, but we expect it to carry on broadly speaking as it has been”. You then look at the variability in your demand and use that to set a re-order point which allows for enough safety stock to buffer against that variability. With deterministic models, the way of thinking is “we know that demand is going to be different from the past and we are going to plan for what we expect”.

What this means is that if you genuinely know future demand is going to be different to past demand (for instance, if you are planning for a promotion or a seasonal uplift, or if you are planning materials for a frozen production schedule) then deterministic planning should in theory be better than replenishment based on historical data. (Although of course, you can base your replenishment model on a forecast too, or temporarily add some anticipation stock to it.)

How do you choose between the two? Is your forecast consistently better than a naïve forecast? If not, you should see benefit from a replenishment model. What replenishment is good at is buffering the actual variability you have. However, because people are perhaps uncomfortable with probabilistic models, there is a strong perseverance of deterministic models, where you base your plan directly on what you expect. Variability can get lost in this model.

How often should you check and update safety stock levels?

As segmenting and calculating ideal safety stock levels is somewhat laborious to do well, the temptation is to do it as rarely as possible, and the good news is that this might not be a bad thing. One of the main issues with deterministic models that are not protected from variability is that they are “nervous”, i.e. they are constantly trying to readjust.

Part of the benefit of a replenishment model and the safety stocks they use is that variability is automatically absorbed by them. So does this mean you can set your re-order points and then head to the beach? Unfortunately not.

In reality, of course, change is a constant factor. The fact that variability was within a certain range in the data you used to define your safety stocks does not guarantee that it will remain in that range. While major events, like a delayed shipment, will no doubt come to your attention anyway, and be dealt with as part of the normal day-to-day inventory management work, a subtle but important change in demand patterns might escape you. Statistical process controls like the Western Electric rules are very useful where the volume of items managed is high.

And even setting sudden or substantial changes aside, most things are undergoing a gradual process of change. A lot will come down to your specific situation. If you have mostly mature products with limited change, once a year might be enough to check and update target safety stock levels. In a more dynamic environment, you will probably need to do it more often. If you find you need to constantly revise the levels, it is questionable whether you really need safety stock at all and some other approach, such as reserving spare capacity, might be preferable.

Where should you build safety stock?

The question also arises, where in your supply chain you should keep safety stock? This is partly a matter of business strategy (how quickly do you want to be able to fulfil customer demand) and constraints (how fast can you source materials and manufacture your products). As such, there is no “right” answer to this question. But as a generalisation for manufacturing industries, most keep some level of safety stock in finished goods since end demand is usually variable and make to order is not so prevalent. A replenishment model is frequently appropriate here.

At the other end of the supply chain, safety stock in raw materials can help protect you against delays or quality issues on the supply side. However, where there is considerable variability on the demand side and MRP has propagated it all the way through to raw materials orders, there is often a marked bullwhip effect and buying raw materials to a forecast that has negative forecast value added (FVA) leads to overstocks.

The most complex area for safety stock is WIP (understanding WIP here to include semi-finished goods) since it is heavily tied to capacity utilisation. In an ideal world, demand from the end customer would flow through and you would need no WIP except for the amount of time it takes to produce what is required. (And indeed, leading proponents of “Lean” manufacturing achieve impressively low WIP inventory levels.)

In reality, however, a lot of factories run at or close to capacity, which means there are frequent bottlenecks and waiting times for machinery to become available. Safety stock may usefully be built at relevant points in the manufacturing process such as at bottlenecks, at processes with very variable yields, or for semi-finished goods that are used for multiple end products. You will often find such safety stock in WIP referred to as decoupling stock (to distinguish it from the safety stock which is buffering against external uncertainties).

Such decoupling stock can be managed in a variety of ways, such as Kanban, CONWIP or DDMRP. It is a valid approach to define safety stocks at decoupling points based on the replenishment model, and the judicious positioning of replenishment points within an end-to-end production flow can mitigate some of the worst effects of MRP. A well-designed approach might build decoupling inventories with a replenishment model, optimizing safety stock for target service levels, and avoid safety stocks altogether for the parts of the process that are genuinely deterministic.

Are there any alternatives to safety stock?

While safety stock is perhaps the most common way to buffer variability, it is not the only alternative.

Another option is safety time. This is where you have quite a good idea of the size of demand, just not when exactly it will come. An example of this might be if you know that Production is planning to manufacture a fixed amount of a product, but the schedule hasn’t yet been fixed. Instead of getting additional raw materials, you just bring them in early.

Another option is reserving spare capacity and/or expediting products. Where you have a very limited ability to predict what demand is going to be, it might be preferable to be able to produce and despatch quickly in response to firm orders rather than building speculative safety stocks. An example for this might be for a new product launch where it is unclear how high demand will be.

A third option is waiting time. While in our modern world we have perhaps become used to having everything on demand and without delay, in certain circumstances it is possible just to make customers wait longer. This comes about through necessity in instances of major shortages and is also the option of choice in many instances of high value, low volume make-to-order products


Like many things in supply chain, a few simple safety stock principles can rapidly become extremely complex once you dig a little deeper and try to find optimum levels in the real world.

Supply chain professionals need to find solutions that are practical, proportionate and easy enough to understand to take people with them. However, safety stock is one of the most important levers you control and the benefits of getting it right, such as reducing production stoppages, late customer deliveries, or the need for expedited deliveries, are substantial.

There is perhaps a tendency to feel that having too much is better than having not enough, but having too much safety stock also has a real downside. Warehouses and production areas can get clogged up, production actually slows down because of the excess, working capital is stuck in unproductive surplus inventory and the risk of waste through obsolescence increases.

Many organisations are already taking advantage of various optimization models, either built in-house using spreadsheets or in tools from software vendors. This can deliver real benefits where it takes people away from cover targets. However, for some of the reasons explored in this article, a lot more is usually possible once you go beyond “one size fits all”.
Due to the complexity involved in optimizing safety stock, even companies already using advanced optimization technology usually have an opportunity to further improve their levels, reducing shortages and excesses simultaneously, often by 20% or more.

nVentic has automated the evaluation of inventory data to help clients identify and act on the improvement opportunity quickly and systematically. If you would like to discuss your safety stocks or other aspects of inventory optimization, please
contact us.
Would you like to receive more content like this, direct to your inbox? We publish white papers on a range of supply chain topics approximately once every one to two months. Subscribe below and we will notify you of new content. Unsubscribe at any time.

Would you like to talk to one of our experts? Contact us