Inventory Dynamics

Advanced Weighting

An Introduction

The philosophy behind an advanced weighting model is simple. If you can more accurately predict when an item will sell, you can also more accurately predict when you need to stock the item. This takes the guesswork out of ordering in that you don’t have to carry excess amounts of stock ‘just in case’ someone might buy it. Using this tool, you can reduce tens of thousands of dollars in excess overstock that you simply don’t need. This is based on solid mathematical and statistical reasoning, not just the guesswork of a $9 per hour sales clerk that you pay to do your order now. Although the methods used are complex, the implementation procedure is straightforward, albeit slightly time consuming. However, if I were to ask you if, with a few hours of work, you could save upwards of a hundred thousand dollars, would you say no? I didn’t think so. To begin, I’ll give a brief overview of how the model works, so that you may have a deeper understanding of how this will save you money.

First, you have your data. You’ve been running a computer system that has been tracking all of your sales history for years, but what do you do with all of that data? You can’t look through it manually. Try looking at just five thousand SKU’s, a mere fraction of what is in your store. It would take you a full day just to look at them all, and what have you gained anyway; it’s still just a bunch of numbers. That’s where math and computers can help. Every item’s history carries with it specific features that can be extracted. For example, here’s a graph of two years data for two top items for a store.

The blue one, Item 1, is clearly predictable; it does the same thing every year. The green one, Item 2, is all over the place; you really don’t know when you are going to sell a lot of it. This new model anticipates these items. It knows when to stock up on Item 1 and when you don’t need as much. It also anticipates Item 2 in that just because you sold a lot last year at this time doesn’t necessarily mean that you’ll sell a lot this year at the same time. Also, the opposite of this holds true. Just because an item was performing poorly last year doesn’t mean that it will this year as well. This is where many other methods fail in that they look at all items the same. A method that looks only at recent history will not be able to predict when to stock up on an item and when you don’t need many whereas a method that looks only at past history will poorly anticipate the less predictable items, such as Item 2, and you will end up out of stock when you need the item most and overstocked when you don’t.

The solution to this problem is to group the items according to how predictable they are. The answer is a Fourier transform. It’s a mathematical technique that separates the predictable items from the unpredictable ones. The graph below is a plot of every item from a typical store. Each blue dot represents the predictability of that item. The higher they are on the graph, the more they behave the same every year, whereas the lower they are on the graph, the more unpredictable they become.

While the specifics of this graph are not important, what you can see are a lot of items that pull to the top, a lot more in the middle, and a few at the bottom. These are the groups that are used to separate the items. I’ll call them Group 1 (predictable), Group 2(middle of the road), and Group 3 (unpredictable) to make it easy. An item like Item 1 (predictable) in the last example would be in Group 1, whereas an unpredictable item such as Item 2 would be in group 3. You don’t have to worry about any of this though; the new model handles all of this automatically.

The next step you may already be familiar with. This is velocity ranking, which, put simply, is how many of an item you sell in a year. It’s easy; if you sell more than 75 of it in a year, it’s an A item, between 75 and 25, it’s a B item, between 25 and 8, it’s a C item, and between 8 and 1 of an item, it’s a D item. This allows you to adjust the stock levels for an item based on how fast it sells. In this model, we only consider A, B, and C items; the fast movers. The D items are actually fairly predictable in that they are unpredictable all of the time, but at the same time, you don’t need to carry many of them. We’ll talk more about D items later.

So now we have nine groups; each Group paired with each Velocity. For example, group A1 would be items that you sell more than 75 of in a year and are very predictable, whereas group C3 would be slower moving, unpredictable items. From this point on, we are going to look at each of these groups as a separate set of items. For each of these sets, we will determine what the optimal weights are and how many weeks of stock should be kept. Again, all of this is done automatically, so if you’re not a big math buff, don’t worry; there’s no quiz at the end.

Each set of items gets its own weight curve. This can be thought of as ‘how much are we going to look at recent history’ and ‘how much are we going to look at last year’s history’. This is simpler than it may sound. Every item has the ability to look at the last 12 months of its sales history. This is simply the number of items it has sold in each of those twelve months. If you are only looking at recent history, the weights will be higher for the recent months and lower for the other months. The example below will help clarify this.

This is an example of a flat weight curve. Yes, the graph is round, but what is important is that each of the pie sections is the same. This means that each of the twelve months of data we have available is treated equally. The blue sections represent the most recent data from the item, whereas the red sections represent last year’s data. If we were going to look more at last year’s performance to gauge what we should do this year, the weight curve would look more like this.

This weight curve shows a lot more red, indicating that the group being evaluated can be easily predicted. This is typically the kind of weight you will see for predictable (Group 1) items. On the contrary, items that are unpredictable (like Item 2 in the first example) cannot look to past sales history for information as to how well they will do right now. These items can only look into the immediate past. Their weight curves will look more like this.

This item only looks at recent history to gauge what it should do in the future. This is the case for unpredictable items where past sales history is no indication of what you might do tomorrow.

By now you might be wondering how these weight curves are generated. Imagine for a minute a round bowl sitting on a table. Every point in the bowl represents a different weight curve. The points around the edge of the bowl are higher than others; these produce excess overstock. The point in the middle of the bowl is lowest to the ground; it produces the least overstock. The goal of weight optimization is to find this lowest point. This lowest point is the weight curve that ensures that you will remain in stock of everything you need while at the same time producing the least amount of overstock. Every other curve will produce more overstock or more outs. In the real world, however, the bowl has more curves, so finding the bottom isn’t as simple as in a round bowl. On top of that, the bowl has twelve dimensions, so I can’t even draw a picture of it for you. But, rest assured, there are mathematical tools that allow us to find the bottom of this strange bowl, and hence the optimal weight curve.

For those of you who don’t like math, I’ll simplify this further. For every group of items, there is one unique weight curve that best predicts what those items will do. This program finds that curve. In this case, we have nine groups of items, groups A1, A2, A3, B1, B2, B3, C1, C2, and C3. Each of these groups of items has their own unique weight curve that is best suited to those items. A typical weight set might look like this.

These graphs represent the weight curves for each of the nine groups. The three to the left represent the A items, or items that sell more than 75 in a year. The three in the middle column represent the B items and the three to the right represent the C items. The top row represents the Group 1 items, or the predictable items. The middle row represents Group 2 items and the bottom row represents Group 3 items, the unpredictable ones. You can see that Group 1 items, the top row, have more red in their weight curves. This indicates that they can look more at the past for what they will do in the future, hence confirming their predictability. With these items you already know from last year when you need them and when you don’t, so it’s easy to have them in stock when you need them and not when you don’t. The items in the middle row, Group 2, have fairly flat weight curves. This is especially true for group C2, which is almost completely flat. This indicates that these groups of items are fairly stable; they will sell the same every month, month after month after month. Since this is the case, we can look at all twelve months for a better picture of how many we need to keep in stock. The bottom row, Group 3, is the unpredictable items. There is almost no red in these graphs, indicating that last year’s sales history provides no indication of what these items might do now. These are the items that are hot one year and dead the next. They are the dogs that sit on the shelf for ages, until one day you can’t keep them in stock any more. In essence, they are unpredictable. This is exactly why there is so much blue in their graphs. This weight curve only looks at the recent performance of an item to gauge what you will need now. If one of these items suddenly takes off, you will automatically stock up to meet the demand. But if an item had a fluke good year last year, you won’t accidentally stock up this year in anticipation of something that won’t come.

The key to all of this is that this is just one typical store with its own unique weights. Every store will have its own unique set of weights. There is nothing in any store that will prevent this program from adapting to your sales history and save you money. On one hand, a store that sells the same items year in and year out will have more red in their weight curves. This store can benefit by reducing stock in the off season while still being well in stock when they need to be. On the other hand, a store that is dynamically changing and reworking itself to keep up with new trends will likely have more blue in their weight curves. This store can benefit by keeping up with promising new sales trends without running out of stock while at the same time reducing overhead in dead items. Or your store can do both at the same time. The program will fit itself around your data, helping you to remain in stock when you need it without costing you a small fortune trying to carry everything at once.

The next step is to determine how much stock you want to carry. You’re probably thinking, doesn’t the program do that for me? The answer is yes and no. It does calculate when you will sell an item, but still allows you to determine how much safety stock you want to have. With any weight curve, you will be out of some things at some times. This is an inevitable and calculated risk. This program shows you, and allows you to change, how much stock you want to carry and how many outs it will produce. In the spring, you can dynamically increase the amount of stock you have, just to be extra sure that you don’t run out of anything; while in the fall, you can cut back your inventory to save money on taxes at the end of the year. It’s all in your control. The following graph shows you what you will see with your inventory.

The top graph is overstock, how much stock you are carrying that you won’t sell. This will always be there. If there were no overstock, you would order exactly what you were going to sell each week and your store would sit empty just as the truck arrives bringing new stock. While you can’t eliminate overstock entirely, you can reduce it significantly without impacting understock. Understock, the bottom graph, is when you are out of an item and someone else wants to buy it. You missed a sale, and moreover, made the customer angry. This is also inevitable. There is no plausible way to have everything all the time, this is a mathematical fact.

The solution is to find a balance between overstock and understock. In these graphs, the red bars show what the Ace way of retailing method would do for your store and the blue bars show what this advanced weighting model will do for your store. The items are broken down into three sets; the A, B, and C items from before. For each of these sets, you can determine how important it is to be in stock for those items. You can click the + and – buttons to increase or decrease your stock and see dynamically how that change would impact your stock levels. If you decrease your stock too much, you will see understock levels jump up, indicating that you will be out of stock more. If you increase your stock too much, you will see the understock levels drop, but your overstock will skyrocket at the same time. The key is to find a balance between the two that you are comfortable with. You can increase the overstock levels to where they match the Ace levels. This will result in you having the same amount of inventory, but far fewer outs. Or, you can reduce the stock level to where the understock is the same that the Ace method would produce, but you will carry significantly less overhead to achieve the same goal. Or you can pick something in between. The choice is up to you. You can tailor the program to fit your needs. The benefit is that this model will always outperform the Ace model because of the custom weight curves.

The only step left is the implementation. This process is straightforward and fairly easy. The program requires that you export your data into a spreadsheet and load it into the program, but full in-depth instructions are provided to assist you with this. After you decide what stock levels you wish to keep, you will need to export the groups and set up the weight curves, but again there are easy step-by-step instructions for doing this. Still there are those pesky D items, the slow movers that sell less than eight per year. The benefit is that these items can be set up the same regardless of your store and they will perform the same. The key is to keep two of them at all times, and you will exceed any sales expectation that the item ever had.

To summarize, increased accuracy in analyzing sales history equals more accurate sales predictions which equals monetary benefits for your business. This program can be used to reduce overhead, reduce outs, or both; the choice is yours. This advanced weighting model allows you to custom tailor your inventory to how you see fit. Top

Find out more!