Your Privacy Matters

Parts of this website use cookies and content from third-party providers. More details can be read in our privacy policy, which also describes how you can amend your preferences at any time.

 Skip to main content

Days Inventory Outstanding (DIO) is an interesting metric. At nVentic, we often use it as a conversation starter. It is a first outside-in look at how a company is doing in terms of inventory management. The fewer days’ inventory you have, the quicker your cash conversion cycle will be. As such, DIO is a useful component of a balanced supply chain scorecard. But what exactly is DIO and what does it really tell you about a company’s inventory levels?

What is DIO?

Days Inventory Outstanding is the value of inventory held divided by an average day’s cost of sales.

Let’s break that down.

Company A’s balance sheet shows an inventory value of €1 billion. The income statement shows cost of sales of €3.65 billion. This means that an average day’s cost of sales is €10 million (€3.65 billion / 365 days). Which in turn means that the €1 billion inventory is equivalent to 100 days’ cost of sales (€1 billion / €10 million).

If you like formulas, DIO = Inventory value / (Annual cost of sales / 365)

DIO of 100 implies that you have inventory to cover you for 100 days’ average activity. However, this is misleading and we will return to what it is really telling you (and what it’s not) below.

The big advantage of DIO is that it is a very simple metric to calculate based on publicly available information. Only privately held firms that do not report their figures, and a relatively small number of firms that do not break down their costs in a way which isolates cost of sales, prevent anyone from calculating DIO for any company. For more detail, and variants on the formula, see the technical notes at the end of this article.

What does DIO tell you?

DIO is a useful comparator, although this only makes sense when comparing companies in the same or very similar industries. For instance, if you manufacture Scotch Whisky and you allow your product to mature for an average of 10 years before you sell it, there is little point in comparing your DIO with that of a business which buys and sells cut flowers.

If you compare the DIO of two companies in the same industry and with comparable products, then – all other things being equal – the one with a lower DIO is likely to be more efficient. That “all other things being equal” is an important point. Differences in DIO can be driven by either structural or performance differences.

The structure of your supply chain is a major driver of DIO. If you have highly variable demand, make to stock and have very long supply lead times, you will need much more inventory than if you have stable demand, make to order and have short lead times. When you source components from the other side of the world while your competitors source locally, you will need more inventory than them.

If there are no major structural differences at play, then comparing DIO between companies is indicative of efficiency. Having higher DIO than your competitors is a strong hint that you have good potential to improve your inventory levels. Although it is also worth saying that very few companies are anywhere close to optimising their inventories. Having a lower DIO than your competitors is therefore not necessarily a reason for complacency.

What does DIO not tell you?

What DIO cannot do is tell you how much inventory you should have. Rather you need to calculate optimal inventory levels bottom up, based on the properties of each individual item that you stock. The optimal level will depend on the service level you are targeting, your lead times, the variability in your supply chain and a number of other factors.

If you can calculate your optimal inventory levels bottom up, then you could in principle turn that into a DIO target, although this is not so simple, since any change in inventory is also likely to involve a change in cost of sales. But if you are genuinely in a position where you regularly calculate how much inventory you should hold and compare it with actuals, then you have a much more operationally useful KPI than DIO in place anyway. At nVentic, we have automated the bottom-up calculation of optimum inventory levels.

DIO is not a proxy for service level. Intuition might tell you that low DIO indicates a risk of shortages, but there is limited data to support this view. Shortages are not caused by not having enough inventory overall, but by not having enough of the right inventory. By definition, if you have a positive DIO, you have not completely run out of inventory. Companies that manage their inventories well have lean inventories and high service levels simultaneously.

And finally, DIO does not actually tell you how much inventory cover you have. In our example, Company A had 100 days inventory outstanding. In a simplistic sense, this implies that they have enough inventory on hand to cover 100 days’ worth of sales. But, of course, they don’t, since DIO is just an average based on value. They have more or less than 100 days item by item – in a lot of cases much more or less.

This much probably seems obvious. There is no way you could actually hold exactly 100 days’ worth of sales in inventory unless you had completely uniform and predictable demand. But how much do you think the distribution of days’ inventory typically varies from the mean? Do you have something akin to a bell curve in mind? Let’s have a look at some actual examples of different types and sizes.

Bar chart showing finished goods warehouse DIO 53.
Bar chart showing manufacturing plant DIO 40.
Bar chart showing manufacturing plant DIO 25.

As you can see from the examples, which are typical for manufacturing industries, only a small percentage of items (less than a quarter) is within the two quartiles either side of the mean. The average DIO is therefore rather a poor indicator of how big an inventory buffer you actually have.

However, this pattern is not necessarily as bad as it might at first appear.

Each item that you stock has its own optimal level that can vary greatly in terms of the average number of days it should have on hand. For items that are quick to acquire or produce and have stable and predictable demand, you might only need to keep a day or two on hand at any one time. Whereas for items with high variability and long lead times, you might need to keep much more on stock. If you are trying to keep the same amount of stock on hand for all items – and this is what happens when you use “cover” targets – then you are actually working against inventory optimization.

The other thing you will notice from the examples above is how the graphs skew to the right. This is because the overall DIO figure is weighted by value, whereas the graphs simply show the number of items. Most organisations have a long tail of “C” articles, which are low volume and low value items. It is not uncommon to have well over a year’s stock for such C items, where minimum order quantities (MOQ’s) are greater than a year’s demand. This is hard to change and generally not worth the effort to try changing. If the aim is to reduce overall DIO, it is usually much better to focus on your high value/volume items.

Conclusions

DIO has a number of weaknesses. It is not a very precise metric, being open to a certain amount of manipulation (see the technical notes below). And it doesn’t, despite what its name might suggest, really indicate how well buffered you are against shortages. The actual amount of inventory held by item is likely to vary greatly from the mean.

And yet, DIO still has a useful role to play as part of a balanced supply chain scorecard. It is a valid indicator of how efficiently your supply chain uses cash. DIO may not represent how many days’ supply you have in inventory at all accurately, but it does represent how long your cash is tied up in inventory on average. And to this extent, it is a valid financial comparator when looking at competing organisations. It is a valuable lagging indicator of your supply chain efficiency. And should act as a spur to look at your inventories much more closely bottom up.

Technical notes on DIO:

You will sometimes see variations on the DIO formula we give above.

Some Finance departments calculate using 360 days in a year instead of 365.

The number of days in a year aside, the main valid alternative to the formula we use is to calculate inventory value by averaging the value of inventory on hand at the start and the end of the period in question. In our example above, we would sum the inventory value at the start and end of the year and divide the result by 2.

There are arguments for and against using this approach. If you have been growing very fast during the year, for instance, then your average cost of sales might be significantly below the cost of sales run rate at the end of the year. You could argue that averaging inventories is better, so that you treat numerator and denominator similarly. On the other hand, if your sales have been flat over the year but you have built up a significant amount of extra inventory, averaging inventory levels between the beginning and end of the year will understate the amount of inventory you’re holding at the end.

You will also sometimes see alternatives to the cost of goods sold. These alternatives typically fall into two camps:

  1. Using sales instead of the cost of sales
  2. Using a modification of cost of sales, for instance only using the cost of materials and overheads, or excluding inventory impairment charges from the cost of sales

The first of these has the one advantage that it is quick and simple – and not all organisations report cost of sales – but strictly speaking it is a dubious approach, since inventory on the balance sheet is valued at cost (or at resale value only if the value is less than cost – which you would hope to be the case only exceptionally!) By using cost of sales, you make sure you are aligning the denominator and the numerator in your DIO calculator.

The second alternative, using a modification of cost of sales, may be done for a variety of reasons. It is fair to say that cost of sales calculations leave a fair amount to any organisation’s discretion. Generally Accepted Accounting Principles (GAAP) have relatively little to say about cost of sales. The accounting methods used (FIFO, LIFO, average cost method) often have a material impact on the cost of sales recorded. Plus, inventory valuation and the speed of writing off obsolete inventory can be managed to influence short term figures. Controlling for all such variables is not possible. It is best to accept that DIO is an approximate metric and work with reported figures.

A common alternative KPI to DIO is inventory turns, or inventory turnover. This is essentially measuring the same thing, just expressed differently.

Inventory turns = cost of sales / inventory value.

Going back to our example of Company A above, the cost of sales of €3.65 billion and inventory of €1 billion indicate an inventory turn of 3.65. So DIO of 100 is expressing the same thing as an inventory turn of 3.65. Noting that those who favour inventory turns as a measure tend to prefer using the average of start and end inventory since the focus is, as the name suggests, more on the movement of the inventory – the turns – rather than the amount of inventory – the days.

Service level is one of the most important but least understood supply chain concepts. Supply chain service levels play two important roles in inventory management:

  1. in measuring how well you’re doing
  2. as a key input into supply chain planning decisions, such as setting safety stock levels

The problem is that the metrics used for the first are usually not the metrics used for the second. And although they might sound similar, they are often measuring something very different.

What impact does this have on your supply chain and what should you do about it?

Service level metrics

Here are some of the most common supply chain service level metrics:

Different service level metrics suit different circumstances. For instance, OTIF might be the favoured measure for customer service, but as OTIF is dependent on transportation as well as stock availability, it is not ideal for measuring inventory performance. Or, to take another example, ready rate is a good measure for spare parts.

Choice of metrics is often driven by what is easy to measure and what is familiar. While these are valid considerations, you may be missing opportunities to add value by using other metrics. Cover is in very common use, despite having a very limited and questionable value for service levels, whereas back orders and out of stock time are less commonly tracked, despite being very valuable metrics in inventory optimization.

Whatever metrics you use, it is essential to define them precisely and uniformly across your organisation.

Whenever you talk about, say, back orders, then everyone should measure and understand them in the same way. (Do back orders include items that are on order but not yet late, or only overdue items? Are you measuring back orders at the end of a cycle, at the beginning of a cycle or at regular time intervals not aligned with the cycle? And so on.)

The most appropriate metrics for your organisation will depend on a number of factors. However, it is always good to review what you use. New metrics might provide additional insights or drive further improvement.

If you only use service levels to measure the performance of processes then it is enough to use those service measures carefully and accurately.

However, if you want to optimize your inventories then you also need service level as an input to planning decisions such as setting re-order points or safety stock levels. Planning technology usually either assumes a service level, or more likely requires a service level as an input. It is very important to understand exactly what service level your planning tools use and how it relates to your operational KPI’s.

Most tools use metrics like fill rate or cycle service level (or variations of these) that are usually not in use as operational KPI’s. This means that you need to be able to translate between the metrics you use to measure your performance and the metrics used for planning, or there is a major risk of sub-optimal outcomes.

The relationship between service level metrics

All of the service level metrics listed above are interrelated and can be calculated from each other if all the right data is present. But their interrelationship is not simple. Confusing them can cause serious problems.

Let us look at three service level metrics to illustrate the differences: OTIF, fill rate and cycle service level.

Let’s say you want to consistently achieve an OTIF rate of 99% to your customers.

You might be able to deliver that 99% OTIF rate with a finished goods fill rate of 0%.

Let’s also say that you want to have a fill rate of 99% for the raw materials needed. You might require a cycle service level of only 30% to achieve that.

The first thing that should be apparent is that OTIF, fill rate and cycle service level are measuring different things. The relationship between them is neither linear nor intuitive. So what is going on in our example?

Fill rate sounds like OTIF, but it’s measuring something different. In our example, the production lead time is shorter than the customer promise time, and you choose to make the product to order. You have a fill rate of 0%, since you never hold it on stock but make it when needed and dispatch it as soon as it is ready. But the OTIF is 99%, since your customers almost always get it on time.

You aim to keep the raw materials on hand for whenever your customers place an order. The fill rate for the raw materials is 99%. You order large batches of raw materials at a time. With a long order cycle, you only need a cycle service level of 30% to deliver a fill rate of 99%. (The exact numbers are an example, not a rule!)

You will notice that in this example we are actually measuring different points in the supply chain: OTIF is measured at the customers’ premises, fill rate is measuring both finished goods and raw materials, cycle service level is measuring raw materials.

Let’s dig into the three metrics a little deeper.

OTIF

OTIF is a very common performance measure for deliveries to customers (as well as deliveries from suppliers). While the example we gave is extreme, there are plenty of instances imaginable where OTIF is quite different from fill rate. We worked with a retailer who had problems with on shelf availability that were almost entirely down to picking and last mile logistics: fill rate at the distribution centre was consistently well above 99% but only 85% made it to the retail stores on time and in full.

OTIF is an important supply chain performance indicator, but has limited value in inventory management.

Cycle service level

Many applications use the cycle service level in many applications, especially for setting safety stock levels, and it is easy to calculate. Equations such as Z×σ(demand) ×√LT and its derivations use cycle service level.

Cycle service levels are not comparable between items with different order cycle lengths. If you order an item daily, then a cycle service level of 99% means on average you will be out of stock once every 100 days. If you order an item every 12 months, then the same 99% cycle service level would mean you are out of stock only once every 100 years!

Note also that cycle service level does not measure how great any stock out may be, just how likely you are to stock out in a cycle.

Cycle service level percentages therefore tend to be quite unintuitive for most people: deliberately targeting a percentage as low as 30% just feels wrong.

Fill rate

Fill rate is much closer to the percentages people have in mind when they think about supply chain performance. A fill rate of 99% means that you can fulfil 99% of demand directly from stock, regardless of the cycle time. It is useful for comparing performance across items and it is sensitive to the size of any stock outs.

The only real downside of fill rate is that it requires much more work mathematically than cycle service level.

A major pitfall arises if you use target fill rates in applications that require cycle service levels. The fill rate will almost always be higher than the cycle service level and sometimes substantially higher.

Avoiding confusion

It is very important to understand the differences between metrics. The difference between OTIF and fill rate is normally quite well understood. What is not always so well understood, is what service level metrics planning tools use and what service levels should be set at different points in your supply chain to deliver the desired performance without overstocks.

So what should you do?

The first step is to look carefully at your planning tools and methods and make sure you understand exactly what service levels they use. (We have found, unfortunately, that documentation is often unclear.)

The second step is to define what good looks like. This may already be clear from your existing operational KPI’s, but you may want to update the metrics you use.

The third step is to build a bridge between the first two. Let’s say you have defined success in terms of out of stock time and that your planning tools use fill rates. You can calculate one from the other, although not at all easily. The same is true of all the metrics above. They are all related, but you need to be careful in translating one to another.

It is quite easy to get bogged down in technical differences, or dispirited by them, but as with all things related to inventory optimization, we like to say that being approximately right is better than being precisely wrong. All “optimal” levels are approximations since variability is unknowable in advance. But it is essential to understand the big differences, such as that between cycle service level and fill rate. And it is therefore vital to understand what service levels your tools use.

Conclusions

Supply chain is complex enough without looking for further complications, but service levels are a prime example of something seemingly simple hiding a number of potential pitfalls. Metrics are very important in how organisations operate. They influence how you work, what you prioritise, even how you see the world.

If you limit your inventory metrics to cover, inventory turns and OTIF, however, then you are missing out on a huge potential to drive improvement. Service level, in its most generic sense, is almost always prioritised over leanness of inventories as the primary measure by which inventory performance is judged. But service level can refer to a range of actual metrics, each of which has a time and a place. To get full value from service levels, they need to be well understood and applied with care.

Would you like some help in better understanding your service levels?

Contact us today

Appendix – On the calculation of safety stocks using service level

Safety stocks represent one of the biggest opportunities to optimize inventories. Service level is an essential concept to understand if you want to optimize your safety stocks. See also our guide to safety stocks.

In the absence of a method factoring in service levels, you might set safety stocks using a rough estimate. For example, your policy might be to maintain 2 weeks’ average sales as safety stock for finished goods. The exact size of the safety stock might be grounded in some kind of consideration, such as usual lead times, maximum experienced demand, or similar. This type of approach has the advantages of simplicity and comprehensibility. But if you don’t factor variability into the equation, you will have a substantial opportunity to further optimize your inventories.

The next level of maturity is to factor variability into your calculations. Safety stock is by definition there to protect you against various types of variability. This is a positive step, but needs handling carefully.

Cycle service level is normally a good place to start, since it is mathematically simplest. Spreadsheet-led approaches often use cycle service level since the equations are reasonably well known. There are important limitations, in particular that cycle service level approaches only work accurately with normally distributed demand.

Theoretical inventory models frequently assume normal distribution, but you encounter it less frequently in reality. It is a good assumption for heuristics covering overall inventory, thanks to the central limit theorem. It is also not a bad approximation for most individual items, unless they display very variable and/or sporadic demand.

While it is in many respects inferior to fill rate, you can make a lot of progress using cycle service level, as long as you make some allowances for the normality or otherwise of distribution. We recommend an incremental approach. If your calculation suggests safety stocks 50% lower than today, don’t halve them straight away, but lower them by steps. And in a first instance leave aside items with sporadic demand.

The other hurdle you have to overcome is working out what cycle service level percentage you need. Especially for items with a long order cycle, the percentage might be much lower than you would imagine.

The next level of sophistication is to work with fill rates instead of cycle service level. This has some clear advantages, once you have overcome the additional mathematical complexity.

Probably the biggest advantage from a practical perspective is that fill rate percentages are more intuitive and comparable across items. You could, for instance, reasonably set a target fill rate of 99% across all items, whereas the equivalent cycle service level percentages are likely to vary considerably by item.

Another advantage of fill rate is that it does not depend on normally distributed demand, but can work with various distributions, including non-parametric. (Which is not to say that it always is working properly. Just because your tool works with fill rate or similar, it does not necessarily mean that it is factoring the actual distribution of demand in.)

Using fill rate will allow you to set safety stocks in a more exact and aggressive fashion. It can even work better than cycle service level for the normal distribution, especially if order quantities are low. But the additional complexity of working manually with fill rate means that we would not recommend it before getting a good understanding of cycle service level.

Some planning technology gives you the option to set a service level target and then factors it into re-order point or safety stock calculations. This is a great resource to have and prevents you having to wade through the calculations yourself. But be aware of the exact type of service level you are using, be aware of the potential delta between whatever the technology uses and your operational KPI’s. And be aware that the tools are most likely making various assumptions, such as distribution of demand, that will affect their accuracy to a greater or lesser degree for each inventory item.

Most practitioners know from experience that their tools don’t set good levels for at least some of their inventories. Developing a deeper understanding of service levels is likely to help you understand one of the major reasons why that might be the case.

Cover has a beguiling simplicity to it. Inventory cover tells you how much stock you’ve got, using a metric that makes good intuitive sense. If you have two weeks’ cover, then you know that you shouldn’t run out of inventory for two weeks.

Cover = Stock held / stock required over time period

Unfortunately, like most things in supply chain, it isn’t that simple. Cover as a metric is at best misleading and at worst positively detrimental to inventory optimization. In this white paper, we’ll look at why this is so, and what better alternatives you have.

The five (main) problems with cover

1. Uncertainty

To calculate the stock required, you can either use historical consumption/sales, or you can use your forecast. Forecasts are generally to be avoided, since then your cover figures bake in any forecast biases you have, but using history can also be problematic, especially when you know that the future will be very different from the past.

In volatile times like those created by the Coronavirus pandemic, where past demand is a worse guide to future requirements than usual, this issue is especially evident. Where you have poor forecastability, cover is particularly meaningless as a concept since you simply don’t know how long inventory might last. Good forecasters know how to differentiate between what can be forecast well and what cannot. Unfortunately, cover mixes both together.

2. Instability

This instability creates very practical headaches for those trying to calculate cover. It also makes the cover metric itself opaque, since changes in cover could be driven by the numerator or the denominator.

3. Distraction

Inventory sales cover graph, showing months on the horizontal axis and inventory on the vertical axis. Comparing inventory, consumption / sales, and cover.

Look at cover. It increases by more than 30% over 6 months, before dropping by around the same amount in the following 4 months. But the inventory itself is almost flat, deviating a maximum of 6% from its starting point. The consumption line shows us what has driven this divergence: cover increased because demand dropped, not because inventories grew.

In one sense this is fair enough. With dropping demand, you do indeed need less to cover you for longer. However, this raises the question of how useful the cover metric is. If you imagine the graph above with only the cover showing and the other two lines hidden, you would not know whether the cover represented a dip in demand and stable inventories, or stable demand and a surge in inventories.

Cover is at least one degree removed from useful measures. Planners work with units, Finance measures in cash. Cover is not a good proxy for either. Discussions of cover need to dig back into the underlying metrics to understand what is going on and what needs doing.

4. Inconsequence

How much cover do you need? Here, cover as a metric falls down on two counts. Firstly, because it is an average, it gives you no indication of how likely you are to be long or short in anything. You can have high cover and high shortages simultaneously, which perhaps more than anything else indicates how misleading the very term cover is. Secondly, it is impossible to say how much cover you really need. Unlike a service level of 95%, which gives you some sense of how well you’re doing, cover of, say, 4 weeks is quite unanchored.

5. Value destruction

Because of the previous four problems, cover risks creating and perpetuating value-destroying ways of thinking about inventory:

  1. That good inventory levels are only a matter of planning to forecast
  2. That good inventory levels can be defined in terms of cover (rather than in absolute terms – units and value – of cycle stock and safety stock)
  3. That current cover levels are more or less ideal, since there is no obvious anchor to anything else – how much cover “should” you have? Using cover will tend to perpetuate the status quo

Cover is often, inappropriately, used to justify more inventories when sales rise. While a drop in sales justifies why cover has increased. It demands improved forecasting even when forecastablity is low. You know this is happening when bad forecasts are usually blamed for inventory levels and when high efforts are put into measuring forecast accuracy. Advanced companies rather look at forecastability and the value add of forecasts.

Given the problems with cover outlined above, what are some better alternatives to track and drive improvement in inventory levels?

Three better alternatives to cover

1. At a high level, use DIO

For the highest-level aggregate inventory KPI, use Days Inventory Outstanding (DIO) instead of cover. DIO suffers from some of the same weaknesses as cover – it is a ratio rather than an absolute measure and it is unanchored – but it has the big advantages that the denominator (cost of sales) is clearly defined, makes aggregation across items simple, and is a metric already recognised by Finance. No work is needed to bridge the gap to financial reports. Of course, cost of sales also varies month to month, so we recommend only using DIO over longer time periods to look at longer-term trends: perhaps only yearly; at most quarterly, and even then comparing year on year figures.

2. At an operational management level, use inventory value

DIO is not so useful operationally, where weekly planning cycles are more common. Here we recommend using absolute inventory value as the key metric. Target safety stock and cycle stock should be calculated item by item and then rolled up into an aggregate. This then allows the gap between target and actual to be calculated on a weekly basis. In this context, “target” should not be understood as the output of weekly planning cycles, but as the output of analysis carried out over longer periods: yearly is normally a good starting point, although 6-monthly or quarterly target setting should be considered by more mature organisations. This whole process should be aligned with the monthly S&OP cycle so that expected fluctuations (allowances for seasonality, new product introduction, planned production shut downs, etc.) can be built into the longer-term plan.

3. At a planning level, balance service level and cash

At a detailed planning level, the two guiding measures should become service level and cash: finding the right balance to keep service level at or above target and inventory tied up in cash at or below target. To achieve this, and constantly improve on it, inventory needs to be segmented and the various inventory levers optimised. Focusing on the absolute requirements for cycle and safety stock at an individual item level, rather than thinking about cover in terms of days across material groups, is a powerful driver of improvement. Inventory targets and planning should be done bottom-up at an item level, not top-down at a material group level.

Conclusions

Cover is still a very common supply chain metric. It has the benefit of familiarity. And it reflects an intuitive understanding that the amount of inventory you need is not unrelated to expected demand. Where cover is used it should not be abandoned overnight. Instead, it is worth keeping cover in place while the organisation gets used to the better measures proposed here, before being gently retired.

But, for the reasons outlined in this paper, the known weaknesses of cover as a metric can be positively detrimental to the way you think about inventory. To become truly world-class in inventory management you need to move away from cover.

You don’t need to explain the importance of time to a supply chain professional. Working out when you need something is as important as working out how much you need. But as a topic, time horizons in inventory optimization is an underappreciated factor.

There are various complexities related to time horizons in inventory management that few supply chain managers understand well. In this article, we will set out some of the key time-related concepts for inventory management, from simple to complex. Most organisations have room to improve in at least one of the areas below, meaning their inventories are sub-optimal.

(More advanced supply chain practitioners might like to skip Part 1.)

Part 1 – Basic time concepts in inventory management

Lead Time

Lead time is the time between placing an order and receiving it or between scheduling and producing it. This is very important in inventory management as the longer the lead time, the more safety stock you need. Long lead times also make organisations less responsive and agile.

Lead time refers to the time for materials to come from suppliers and the time it takes to produce something. Manufacturing lead time can be dependent on considerations such as capacity utilisation. With bill of material items, concepts such as cumulative lead time, exposed lead time and decoupling points are essential, but not the focus of this paper.

Cycle time

Cycle time is the time between orders. Do not confuse it with the lead time. If you order a component every two months and it takes 2 weeks to arrive, then the cycle time is two months and the lead time is two weeks. Note that the cycle time will be greater or less than (or identical to) the lead time.

Review Periods

The review period is how frequently you check how much stock you have. You might, for example, check your inventory levels weekly and then decide whether to place an order or not. This is called periodic review.

The main alternative to periodic review in make or buy to stock scenarios is continuous review. Here you monitor inventory levels constantly and place an order as soon as inventory drops to the reorder level.

Many equations and other applications assume continuous review. This means they assume that the instant inventories fall to the reorder level, you will place an order. If this is not the case then the impact of review periods needs factoring into lead time.

There are good practical reasons why organisations don’t use continuous review in many situations. It often makes sense to aggregate demand at least daily, if not weekly or even monthly. This aggregation allows procurement, logistics and production to optimize their own processes more easily.

However, infrequent reviews can be the cause of both excess inventories and shortages. Increasing review frequency will normally in itself reduce the need for inventory and improve responsiveness.

Planning horizon

The planning horizon is the period in the future for which you make plans. Organisations may well be working to multiple planning horizons at a time, combining, for instance, approximate plans for the coming 3 months, with very detailed plans/schedules for the coming week. Having the required materials to hand to execute the plan is a core function of inventory management. Plans are updated on a rolling basis and are subject to almost constant change. Obvious challenges arise where lead times are longer than planning horizons.

Fixed horizon

In order to avoid the inefficiency created by changes to production schedules, it is common to have a fixed horizon. During this fixed period production proceeds according to plan regardless of changes in end demand. During the fixed horizon, you should not cancel orders or change production schedules.

While beneficial for production efficiency, fixed horizons propagate end demand variability further upstream. They can cause or exacerbate bullwhips. With a two-week fixed horizon, for instance, you are constantly producing to the expectation of demand from two weeks ago. The difference between actual demand and that expectation needs mitigating in the next period.

On the other hand, fixed horizons do provide certainty in the short term. You can place orders and schedule production with a high degree of certainty for everything within the fixed horizon. In principle you shouldn’t need safety stock for a fixed horizon as you have removed the demand uncertainty from it.

Safety time

Safety time is a sister concept to safety stock. Instead of buffering uncertainty with quantity (i.e. safety stock), you buffer it with time (i.e. by producing or buying it ahead of when you need it). Safety time is usually good at dealing with uncertainty in timing (on the demand and supply sides). Safety stock is usually best at dealing with uncertainty in quantities.

Variability

Variability describes how much values vary from the mean. Demand and lead time being the main two values whose variability influences inventory management. Variability can be measured using the coefficient of variation (the standard deviation divided by the mean), or a related measure, so that it is comparable between items. Important is that as the dispersion from the mean goes up then the measure of variation goes up.

To give a simple illustration:

Intermittence

Intermittence describes how likely it is that you have demand in a given time period. Demand series with many zero values, such as for rarely-needed spare parts, are intermittent.

Part 2 – Time concepts for inventory analytics

Buckets and slices

When analysing inventory, two basic concepts relating to time are essential: the slice and the bucket.

A slice is the granularity in which transactional data is available. With modern ERP systems, every stock movement might be time stamped to the second. However, for most practical purposes it is not necessary to use slices any more granular than a day. I.e. demand can be aggregated at a daily level to create the basic “slices” of data.

A bucket defines the frequency with which it is meaningful to analyse the data. If, for instance, inbound deliveries are only possible once per week, then it makes sense to work with weekly buckets of data rather than daily buckets.

Buckets can be expressed as a number of slices. So you might have 365 slices in a year and each bucket is just one slice (a daily bucket). Or you might have 12 slices in a year and each bucket is 3 slices (a quarterly bucket). And so on.

In inventory analytics these buckets and slices are the essential building blocks of everything else. Generally speaking, organisations can benefit from increasing the granularity with which they analyse their inventories.

Your choice of bucket size also in itself influences your perception of variability. Variability decreases as your bucket size gets larger. So monthly variability is usually lower than weekly variability which is lower than daily variability. (This is why forecasts tend to be more accurate the more you aggregate them.)

One further important consideration here relates to working days. You might operate 7 days per week, in which case you might work with 365 slices in a year. But if you only operate 5 days per week then you should choose 260 slices in a year (or fewer, if you also factor in other non-working days like public holidays). Otherwise, any subsequent analysis will be inaccurate because of the impact of the zero values.

To give a simple illustration:

Demand per day:

Considering all 7 days, average daily demand = 7.14 and standard deviation = 5.04. Coefficient of variation (CV) = 0.706

With just 5 days (Monday to Friday), average daily demand = 10 and standard deviation = 1.58. Coefficient of variation (CV) = 0.158

(Note also that if you work with weekly buckets, the working day problem disappears. Regardless of the number of slices that are greater than zero, weekly demand is identical.)

Appropriateness and “one size fits all”

In the previous section we talked about the frequency “with which it makes sense” to analyse data. But how do you determine what makes sense when it comes to setting bucket size?

Consider a fresh produce business. For one half of the year, it sources fruit and vegetables from local or regional suppliers, with lead times ranging from 1 to 3 days, in order to supply retailers based on daily requirements. For the other half of the year, however, it sources the same products from the other side of the world, with lead times ranging from 4 to 6 weeks.

During the first half, it makes sense to look at daily data, since you need much less inventory with such short lead times. But during the second half there is both a logistics constraint (there are a finite number of vessels sailing each week) and a greater need for safety stock due to the increased lead time. At this stage it makes more sense to consider weekly data.

The simple example above is intuitive. However, in more complex manufacturing contexts, there are usually significant differences on both the demand and supply sides. Demand may vary greatly by day but not by week, or it might vary significantly by week too. Lead times may be as low as a day or as high as six months or more and are themselves subject to variability. Logistics constraints will vary. High runners may be closely monitored and managed “day to day”. Some C articles may be provisioned for the whole year and then ignored.

This range poses a challenge for the inventory analyst. Some items might most meaningfully be considered in daily buckets, others in weekly buckets, others in monthly or even larger buckets.

There is also potentially a difference between how you manage an item today and how you should. For instance, planners may place monthly orders with suppliers simply because they don’t have time to do it more often or just because “that’s how it’s always been done”. But this monthly review cycle might be sub-optimal for the demand in question.

To avoid the complexity of working with multiple bucket sizes simultaneously, you might take a “one size fits all” approach to inventory analytics. This is often a good, pragmatic move that shouldn’t even matter too much in a first instance.

However, a differentiated approach can, in a second phase, add significant value over a one size fits all approach. If you’re working with a standard bucket size across all items regardless of their own characteristics, then how optimal each item’s inventories are will be influenced by how close their most appropriate bucket size is to the standard in use.

Part 3 – The impact of using mismatched or sub-optimal time horizons

Time is an essential factor in inventory management. Failure to take account of the time horizons in play leads to several sub-optimal outcomes. It affects both your understanding of your inventories and, accordingly, how you manage your inventories.

Here we are going to look at the impact on inventory management of using sub-optimal time horizons in three areas: variability, intermittence and forecasts.

Working with Variability

Variability is one of the most important concepts for inventory optimization, but it tends to be poorly understood, perhaps because dealing with it well requires a certain understanding of statistics.

In practical terms, to optimize inventories, variability should inform two main decisions: the setting of safety stocks and the setting of stock policies. But there are very important time-related considerations here that we often see under-appreciated, both of which come down to one key concept:

The variability that really matters is variability over the lead time.

Once you have placed an order, you have to wait at least the lead time before you receive it. The whole purpose of safety stock is to protect you against this variability. (This is still the case even if your cycle time is shorter than your lead time.)

What matters when setting safety stock levels is the deviation (how far expected values are from the mean) over the lead time.

The more important consideration when choosing appropriate inventory policies (such as whether to use replenishment or not) is variability. It is not that replenishment cannot handle variability – quite the opposite – but it is important to understand that replenishment ceases to be a good model at very high levels of variability. (Variability is of course not the only factor when deciding whether to use replenishment or not – forecastability is the most important consideration.)

But to do both of these things well, you should ideally look at them over the lead time.

To continue with the simple example we gave above, with 5 days of demand in a week:

Monday 8
Tuesday 12
Wednesday 10
Thursday 11
Friday 9

Looking at daily variability:
Average demand = 10
Variance (s2) = 2.5
Standard deviation = 1.581
CV = 0.158

However, the lead time is 3 days. So over that (3-day) lead time:
Average demand = 30
Variance (s2) = 9
Standard deviation = 3
CV = 0.1

Comparing the two, the deviation is larger over the lead time (3 vs 1.58) – logically, since there are more days’ demand – but the variability (the CV) is smaller (0.1 vs 0.158) because the values are closer to the average as a percentage of the average.

The deviation is the number you should factor into safety stock decisions, and the variability is the number you should use to help inform inventory policies and XYZ classification.

XYZ analysis is a method to classify your inventory based on variability, typically using the Coefficient of Variation (CV). Items are classified as flat (X), variable (Y) or erratic (Z).

XYZ analysis should in principle deliver different splits depending on how variable demand is relative to a pre-defined standard. Eg you could have predominantly X’s if demand shows little variability for most items, or you could have only Z’s if demand is erratic for all items. Sometimes we see organisations manipulate the thresholds between X, Y and Z to deliver a certain proportion in each. This is fair enough if the purpose is purely to divide up responsibility (although you could do that with significantly less mathematical effort!) but that is not the best approach if you want to use XYZ analysis to actually understand and manage variability.

Working with intermittence

We gave a simple illustration above of the impact of including or excluding zero demand days (i.e. weekends in the example given) on some of the key statistics such as mean and standard deviation. Including zeros will tend to reduce the mean and increase the standard deviation.

The same principle applies to intermittent demand. Intermittence relates to the time between demand points.

Demand can be variable and intermittent or just variable or just intermittent.

Here is a data sequence that has significant variability but low intermittence:

5,15,20,7,10,2,15,4,9,11,6,6,19,0,14

Here is a data sequence that has high intermittence but low variability:

0,0,0,0,0,2,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,1

And here is a data sequence that has high variability and high intermittence:

25,0,0,0,0,0,0,2,0,0,0,0,0,0,0,0,0,0,0,142,1,0,0,0,0,0

Like XYZ, you can divide demand into 3 classes (frequent, intermittent or sporadic) based on the probability of having zero demand in any given time period.

Here again you need to use appropriate buckets. If you receive monthly orders, then demand is not intermittent just because you have ~29 days of zero demand followed by one day with demand. Instead, you need to work with monthly buckets and then you find have frequent demand. But for genuinely intermittent demand, when you don’t know when orders might come and they are not evenly spread out by week or month, there is value in separating variability from intermittence.

Because in day-to-day work planners have to deal with both variability and intermittence (how big will demand be and when will it come?) the two challenges can seem to be just one (uncertainty). But to optimize inventories it is useful to separate the two.

There are many things to be said on how best to manage inventories that are variable, intermittent or both, but the main point here is that to do it well you need to understand which of those combinations you are dealing with.

It is also worth being aware that some analyses, including software commercially on the market, do not separate variability from intermittence. This is not wrong if you only want to consider variability in one dimension instead of two, but we believe there is value in separating the two since the three sample data sets above can clearly, from a practical standpoint, usefully be managed in different ways.

To summarize these sections on variability and intermittence, your selection of time horizons has an important impact on the analysis you carry out. Especially since many of the most common inventory management models like replenishment break down beyond a certain degree of variability and/or intermittence, there is significant value in being able to segment inventories and manage them accordingly. Often, we find, the problem is not that replenishment is over-used, but that it is under-utilised because it has been found not to work equally well for all items.

Forecasts

Another common blind spot in inventory management relates to forecast value added.

Forecasts themselves are usually bucketed by time – i.e. forecasts are made per time unit, such as a week or a month.

It is sometimes said that forecasts are always wrong, but the essential thing to understand is that the point of a forecast is to add value, not to be right or wrong per se. You make a forecast to facilitate better decision making. While, as we have written elsewhere, the importance of forecasting in inventory management tends to be overestimated, one of the principal functions of a forecast is to allow you to make decisions concerning inventory.

This is where the concept of Forecast Value Added (FVA) comes in. FVA measures whether your forecast is adding or destroying value relative to a naïve forecast (i.e. that whatever happened in the last time period will happen again in the next time period). It does this by comparing actual demand with forecast demand, such that you can see if your forecast gets closer to actual demand than a naïve forecast would. (You can also use FVA to compare the value of two different forecasts.)

The issue arises when it comes to timing. Since it is normal to update forecasts periodically, for instance weekly, it seems natural to also measure forecast performance using that same time period. For example, what was the forecast for week 10 and what were the actuals in week 10?

Measuring your forecast performance from one forecasting period to the next is a valid measure of how “good” your forecast is, but it is not necessarily a valid measure of how much value it is adding if your forecast review period is not the same as your lead times.

To give a simple example of this: Imagine your forecast accuracy for the chance of rain tomorrow is higher than assuming the weather will be the same as today. Great, your forecast is adding value! But only if you have to choose today whether to wear a raincoat tomorrow. If you can decide tomorrow, then your forecast is worse than waiting to see whether it really is raining tomorrow. Or if you had to make a decision last week whether or not to have a BBQ party tomorrow, when the rain forecast for tomorrow might have been 50/50 at best, then your forecast is in reality no better than a naïve one.

In other words, to judge whether your forecasts are really adding value or not, you need to measure the forecast at the point in time when the related decisions are being made, which is often when orders are placed, which depends on the lead time.

Unfortunately, to measure the true value of a forecast in this way means measuring it separately for every item you stock given all items don’t have the same lead time. And even then, it is subject to lead time variability. You can understand why, for simplicity, many prefer a “one size fits all” approach, where you ignore lead times and just measure FVA weekly or monthly, but you can also see how this can easily give a misleading impression of how “good” or useful your forecasts really are.

Conclusions

In summary, the impact of sub-optimal time horizons on the management of inventory comes down to sub-optimal safety stock levels, to the underutilisation or rejection of replenishment models, since it is poorly understood for which items it can best be used and how, and to a lack of transparency as to how variable demand really is or how good forecasts really are.

It is not uncommon to find organisations who think their forecasts are strong and their variability well understood, but with inventories far from optimal. What makes it worse is that precisely because they think their forecasts are strong and their variability well understood, they don’t even realize how far from optimal their inventories are!

So what are some practical steps that organisations can take to better master time in their inventory management?

  1. Straightforward, pragmatic steps include reducing lead times, reducing cycle times and reducing review periods. While there is an optimum minimum for each – and at nVentic we base all our analysis on such optimization calculations – as long as you can reduce these three things without adding cost then there are clear benefits to the shortened time horizons.
  2. Another thing to pay attention to is the quality of data in your systems, especially lead times. Since so many inventory optimization calculations rely on lead time, having poor lead time data in your systems is a common Achilles heel for automation and other digitisation initiatives. Sometimes organisations are not so much held back by a lack of understanding of the importance of lead times as frustrated by their inability to reliably ascertain them.
  3. Once you have made good progress on the first two steps, it is worth delving deeper into some of the more advanced concepts discussed in this paper. With an improved understanding of time, inventories can be better classified and optimal approaches applied to different classifications. Rather than going straight to bespoke time horizons per item, a good intermediate step is to segment your inventories based on lead times, eg items under 1 week, 1-4 weeks, 1-3 months and over 3 months.

Is it time to look again at time in your inventory management?