Operations and supply chain people talk about metrics all the time, but we don’t spend a lot of time on the metrics themselves. The nature of how metrics function, and how they are put together, is worth our attention: If we don’t understand how metrics work and what different kinds of metrics are intended to accomplish, we are likely to either mis-use, mis-interpret, or fail to follow up on the insights that metrics can provide.
More than that: part of the value of metrics is their ability to communicate business-critical data. If a picture is worth a thousand words, a well-structured and well-understood metric is worth ten pictures. But if the metric isn’t put together properly or if it isn’t understood the same by everyone, it is worse than useless; it can impede communication and destroy value.
For example, here are a few issues that I’ve observed over the years:
- Misusing “KPI” as a synonym for ‘metric.’
- Assuming that everyone uses the same calculation for a metric (or not understanding the precise calculation for that metric within your company or organization).
- Failing to use higher-level metrics and analytics to understand the reasons for performance measures.
- Measuring one organization’s performance in a way that actively works against the success of another (allied) organization (like measuring the logistics organization on cost per truckload while measuring the supply chain organization on inventory turns).
- Misapplying a detail-oriented metric for general use, and then not providing any explanatory detail.
Mistakes like those can lead you off in a variety of wrong directions:
- The acronym “KPI” stands for Key Performance Indicator. It is, first of all, a metric that provides information on performance – how well a business or an organization is functioning. Secondly, it’s a key performance indicator. Usually that means that it (a) should be maintained for a period of some years, so that performance can be tracked over time; and (b) it should relate in some way to the organization’s overall strategy. Demoting “KPI” to the status of a plain old metric devalues an organization’s real
- Understanding exactly how a metric is calculated is critical to both explaining it and improving it. Inventory Turns are a good example. Most people understand Inventory Turns to mean how many times inventory is ‘turned’ in a year, but the actual calculation is often based on three months’ worth of data. Additionally, the financial calculation for turns usually includes overheads, which can make the result look different than if it is based solely on consumption. A supply chain organization may focus on forward-looking inventory turns (using the next three months of forecasted demand, for instance, annualized); the financial organization will probably look backwards (e.g. the current month and the previous two months, also annualized). That can result in the same end-of-month inventory producing very different values for what is theoretically the same performance measure.
- There are different kinds of metrics (which we’ll cover in more detail below). Some provide an indicator of whether performance has improved or not; others help you understand why performance has changed; and still others can help you predict future performance or identify actions to ensure that performance goes the right direction.
- Some metrics are most useful at a detail level, and others make much more sense at an aggregated level. “Days of Supply” is a good example. It’s possible to calculate an overall days of supply metric for a facility: (Total inventory value on hand)/(Average daily demand value) will do it. That (aggregated) metric is essentially the inverse of inventory turns (without overheads). But it doesn’t tell you much – if you have 20 days of supply, does that really mean that you can produce for 20 days before you’ll need more materials? (Not if half your inventory value is obsolete or slow-moving inventory, and not if some of your materials are near stock-out conditions.) But Days of Supply is a very useful metric at a detail level: Looking at the low values will tell you which items are near stock-out or have already stocked out, and looking at the high values will identify items that are slow-moving or obsolete.
Recently I’ve had the very good fortune of working with some supply chain experts at ASCM (the Association for Supply Chain Management, formerly APICS) on the next version of the SCOR model. One of the areas of discussion has been just how good supply chain metrics are constructed and related to each other. There are quite a few models out there – and almost all of them have some value – but my intent in this article is to simply propose a useful way to look at metrics overall.
First, let’s observe that metrics can have different purposes:
- “Event Measures” – what we might think of as Activity, Level or simply descriptive metrics – monitor how much of something has occurred. They only tend to be useful when provided as a rate, and even then only in context. For example: “10,000 passengers” is almost meaningless. 10,000 passengers per day provides a measure of activity, and sounds like quite a lot – and it would be, for a small ferry operation. In the context of, say, the London Underground, however, it only represents about 0.2% of a busy day’s passengers.
Input and Output measures can be regarded as specific types of event measures. So can “environmental” measures – which may be completely out of our control, but still affect our processes. Temperature and humidity are obvious examples, but surface area for a warehouse, interest rates or even the cost of a gallon of gas, can also describe the environment that a process operates in.
- A performance metric provides an indication of how well a process or a function is operating. Inventory turns or accuracy, or customer service level are an important indicator of how well a supply chain organization is working.
- Both of those first two categories (event measures & performance metrics) are essentially informational
- In contrast, a diagnostic metric helps to explain why a performance metric is the way it is. If inventory turns are poorer than desired, understanding that (for instance) 1/3rd of inventory value is obsolete or slow-moving provides not only an explanation but also the start to a discussion of how to improve the performance metric. Many diagnostic metrics are “quality” metrics, because they provide an indication of performance in relation to a standard. (For instance, if we were using “days of supply” as a diagnostic metric, we might designate a range of 5 to 20 days as the desired range. Any part with less than 5 days of supply would be in danger of stock-out; and any part with more than 20 might be viewed as potentially slow-moving or obsolete.)
We also may assign activity, performance and diagnostic metrics to specific functions or operations – so we can think of financial metrics, supply chain metrics, purchasing metrics, etc.
Finally, people often think of “analytics” as just a more sophisticated sort of metric, but this makes the line between metrics and analytics fuzzier than it needs to be. One useful distinction is that metrics usually focus on data starting at today and looking backward; they try to understand what has happened or what is, and why it is that way. Analytics tend to look forward; they try to evaluate what is likely to happen, and identify potential action(s) to enable or prevent that likely event. It can be a bit confusing, since analytics may be based on much of the same data as metrics. Analytics look for predictive patterns in the data and try to project forward from the current situation.
Analytics are frequently broken down into descriptive, predictive, and prescriptive categories. Descriptive analytics differ from descriptive metrics in that analytics look for insights on how to approach the future, where descriptive metrics simply try to describe the past and present. Predictive analytics generally use modelling, analysis, simulation and other tools to generate a quantifiable view of the future. (This may not seem like a lot, but it is really pretty important. Contrast the difference between someone telling you that they ‘forecast’ sales of $10M next month; versus a forecast of $10M with a 50% confidence, $12M with 10% confidence, and at least $8M with 80% confidence. That more quantified view is a lot more useful for planning.) Prescriptive analytics go even farther and use simulation and optimization to identify how to obtain the best possible outcome.
A graphic view of how these all exist together is shown in the figure above. This isn’t “the only right way” to look at the differences in metrics and analytics – there are many others that may be useful. But I hope it is a useful way to understand the distinctions between the different types.
How would these function in real life? Here’s a scenario that I’ve actually lived through (edited and expanded somewhat): the Controller at an electronics manufacturer announced to the senior staff that cash reserves were at a particular level. This was an “Event Measure” or an Informational Metric – meaningless without context. In this case, the context that the Controller provided was that the amount of available cash was far lower than we needed for ongoing operations.
We reviewed the three major performance metrics associated with cash flow: Days of Accounts Payable and Receivable, and Inventory Turns. As you might expect, we had a few late-paying customers but this wasn’t the root cause of our cash problem; neither were our AP Days Outstanding out of control. (We chose to pay some suppliers late to preserve what little cash we had.) The problem was that our inventory turns were far too low. Inventory was using up all of our cash. Why?
I wish I could tell you that we used a sophisticated diagnostic metric at this point. In fact, at my next company that’s exactly what we did. At that (next) company, we built an “inventory quality” metric that compared the actual amount of inventory on hand to our target inventory, by part number.  At this company, however, we did something less sophisticated and more time-consuming, but ultimately effective: we walked around the stockroom. We noticed one thing right away: lots of air-freight boxes with dust on them. Someone had worked hard to bring additional materials in, and then we had failed to use them. Using the “5 Why’s” strategy, we figured out eventually that our buyers had been given confusing direction and poor tools to keep inbound material receipts aligned with our production plan.
Going a bit farther, at another company we built a simple analytic tool that looked at projected inventory levels (by part number or material code) at a point many months in the future. The idea was that by that time, under normal circumstances the MRP process should have brought inventory into alignment with our guidelines – if not, it was something we should look at. Any part that had inventory value over a particular level (around $10,000) was worth reviewing to identify an action we could take to bring its level down over time.
Without an understanding of how diagnostic metrics could expand and explain the causes of poor performance, that first company spent a lot more time and energy than needed to find the root causes behind the cash flow problem. And without the use of analytics, the second company was unable to identify future inventory imbalances – excess and obsolete inventory continued to appear with seemingly no warning.
Understanding at least these basic concepts about metrics can help an organization not only understand its performance, but identify (diagnose) the causes of its performance – and even predict future problems in time to correct them.
 A useful, if slightly different take on analytics vs. metrics can be found at https://www.hipb2b.com/blog/metrics-analytics-and-kpis-whats-the-difference
 A relatively straightforward explanation of the different types of analytics can be found at https://www.dezyre.com/article/types-of-analytics-descriptive-predictive-prescriptive-analytics/209. (You may regard the Thomas Jefferson quote as spurious.)
The idea of an inventory quality metric is pretty simple: if you are trying to operate with, for instance, 1 to 2 weeks on hand of every “A” part, 2 to 4 weeks of Bs, and 3 to 6 weeks of Cs; then you can compare the actual inventory on hand of each buyer’s complement of parts to the midpoint of those ranges. Significant amounts over 1.5 weeks of As, 3 weeks of Bs, and 4.5 weeks of Cs will be very obvious, and can be quickly prioritized and corrected. For example, if a particular “A” part used 10,000 pieces/week, and we had 40,000 on hand, that would be 25,000 too many. If each part had a standard value of $2, that would be worth $50,000. Fix that problem and inventory will go down by at least tens of thousands of dollars.