Demand Sensing and the Art of Hitting a Curve Ball – Part 1

They say sports is not just about doing the right thing, but being able to do it very fast. Think of an outfielder in baseball who, at the crack of the bat, senses whether the ball is headed behind or in front of him and starts immediately heading in the right direction. Or a batter who senses what kind of pitch is thrown – fastball, curve, slider, changeup – very soon after it leaves the pitcher’s hand, thereby increasing their chance of accurately hitting the ball. The result can be clearly seen in performance. Right hand batters perform significantly better against left handed pitchers because they can see the ball coming off the pitcher’s hand a split second sooner than left handed batters (The same is true for left handed batters and right handed pitchers). This ability to sense what is happening as soon as possible is critical to the success of any professional athlete in almost any sport.  And it matters in supply chain too. Being able to sense demand a little sooner, a little faster, has a similar impact on forecasting and supply chain performance. As you’d expect, sensing demand is exactly what many more advanced – and higher performing — demand forecasters are doing. We’ll look at this phenomenon at three levels of skill and performance:

1. Sensing channel demand
2. Sensing demand at the customer
3. Sensing demand before it happens (we’ll explain later)

Let’s start with sensing channel demand (also called ”ship-to data“).  This is the most common type of demand sensing. It includes capabilities like detecting daily replenishment patterns and using advanced consumption logic at the SKU/Location level. Channel demand sensing analyzes partial period actual demand to perform automatic short-term forecast adjustments. It uses pattern recognition to improve short-term forecast accuracy by automatically detecting a pattern in replenishment orders. And it uses advanced analytics to sense demand signal changes compared to a detailed statistical demand pattern, evaluating the statistical significance of the change. The key is that by understanding demand at a more granular level and looking a little downstream, you see demand closer to the end customer and reduce latency. Looking at channel demand also reduces aggregation. Working with aggregated demand may seem better because it’s easier to deal with, there is simply less data, and because it’s often less noisy. But the problem is that aggregating demand also washes out not only the noise, but the signal as well. For instance, by aggregating demand across multiple channels, you can’t see what’s happening in the individual channels. You lose valuable demand cues. Sensing channel demand is a good first step – but going back to our baseball analogy, this capability may get you into the big leagues, but probably won’t make you a star player. Good, but not best.  Here’s why.  When you’re dealing with data working its way through the supply chain, the signal degrades with each echelon further back. Putting real numbers to it makes the performance degradation immediately clear. A large consumer goods manufacturer we know was achieving an overall 98% service level at their DC, a 95% service level at the retailer’s DC, but only a 92% service level at the retailer’s stores. Meanwhile inventories were overstocked by almost 50%. What was happening? Fulfillment was being driven by the vendor based on aggregated data and latency introduced with each additional echelon. The vendor was fulfilling the demand they were seeing, but it wasn’t the demand the retail stores were experiencing. So the further away the point of demand, the worse the signal, the worse the forecast, and the worse the service level performance at the point of sale. So what’s the next step? Next week we’ll see how the best forecasters are overcoming this problem.   Click below to download a copy of an eBook on Demand Sensing.