Improve Forecast Accuracy, Eliminate Excess Inventory, & Maximize Service Levels

In this video, Dr. Thomas Willemain, co-Founder and SVP Research, talks about improving forecast accuracy by measuring forecast error. We begin by overviewing the various types of error metrics: scale-dependent error, percentage error, relative error, and scale-free error metrics. While some error is inevitable, there are ways to reduce it, and forecast metrics are necessary aids for monitoring and improving forecast accuracy. Then we will explain the special problem of intermittent demand and divide-by-zero problems. Tom concludes by explaining how to assess forecasts of multiple items and how it often makes sense to use weighted averages, weighting items differently by volume or revenue.

Inventory Planning from the Perspective of a Physicist

In a perfect world, Just in Time (JIT) would be the appropriate solution for inventory management. If you can exactly predict what you need and where you need it and your suppliers can get what you need without delay, then you do not need to maintain much inventory locally.  But as the saying goes from famous pugilist Mike Tyson, “everyone has a plan until they get punched in the mouth.” And the latest punch in the mouth for the global supply chain was last week’s Suez Canal Blockage that held up $9.6B in trade costing an estimated $6.7M per minute[1].  Disruptions from these and similar events should be modeled and accounted for in your planning.

The assumption that you can exactly predict the future was apparent in Isaac Newton’s laws. Since the 1920’s with the introduction of quantum physics, uncertainty became fundamental to our understanding of nature. Uncertainty is built into fundamental reality.  So too should it be built into Supply and Demand Planning processes.  Yet too often, black swan events such as the Suez Canal blockage are often thought of as anomalies and as a result, discounted when planning. It is not enough to look back in hindsight and proclaim that it should have been expected. Something needs to be done about addressing the occurrence of other such events in the future and planning stocking levels accordingly.

We must move beyond the “thin tailed distribution” thinking where extreme outcomes are discounted and plan for “fat tails.”  So how do we execute a real-world JIT plan when it comes to planning inventory? To do this, the first step is to estimate the realistic lead time to obtain an item. However, estimation is difficult due to lead time uncertainty.  Using actual supplier lead times in your company database and external data, you can develop a distribution of possible future lead times and demands within those lead times. Probabilistic forecasting will allow you to account for disruptions and unusual events by not limiting your estimates to what has been observed solely on your own short-term demand and lead time data.  You’ll be able to generate possible outcomes with associated probabilities for each occurrence.

Once you have an estimate of the lead time and demand distribution, you can then specify the service level you need to have for that part. Using solutions such as Smart Inventory Optimization (SIO), you will be able confidently stock based on the targeted stock-out risk with minimal inventory carrying cost. You may also consider letting the solution prescribe optimal service level targets by assessing the costs of additional inventory vs. cost of stockout.

Many of our customers that saw demand dry up during the pandemic are now seeing demand return.  Some are seeing a significant demand surge. Other customers in critical industries like plastics, biotech, semiconductors and electronics saw demand surges starting as far back as last April. For suggestions about how to cope with these situations, please read on.

Surging demand usually creates two problems: inability to fill orders and inability to get replenishment due to supplier overload. This situation requires changes in the way you use your advanced planning software. Here are three tips to help you cope.

Tip #1: Narrow your temporal focus

In normal times (remember those?), more data implied better results. Nowadays, old data poison your calculations, since they represent conditions that no longer apply. You should base forecasts and other calculations on data from the current situation. Where to cut off past data may be obvious from a plot of the data, or you may decide to set a “reasonable” cutoff date based on a consensus of colleagues.  Smart Software has developed machine learning algorithms that automatically identify how much historical data should be optimally fed to the forecast model. Be on the lookout for these enhancements to the software that will be rolling out soon. In the meantime, conduct accuracy tests using held-out actuals using different historical start dates.  Smart’s forecast vs. actual feature will support this automatically.

Tip #2: Increase your planning tempo

When operations are stable, you can set your inventory policies and trust them to be appropriate for a long time. When times are turbulent, it is important to increase the frequency of your planning cycles to keep old policy settings from drifting too far away from optimality.  More frequent recalibration of your stocking policies and forecasts means that you’ll be quicker to catch trends that will surprise your competition and always keep you steps ahead.  With software capable of automatically selecting optimal values, all that work can be done in one shot by the software. You should review those changes and possibly tweak them, but it makes sense to let the software do the bulk of the work.