INVENTORY MANAGEMENT
by Brock Davison

Inventory management traditionally has been the ultimate balancing act in retailing: carry too little and you risk losing sales as a result of stock-outs; carry too much and you tie up a precious resource - cash - unnecessarily.
Typically, the retailer wrestles with two fundamental questions: how important is inventory management?, and how do I figure out what just the right amount is?

Jim Collins, in his book Good to Great, suggests answering the first question this way: “Imagine walking back into the warehouse and instead of seeing boxes of cereal and crates of apples [or in your case boxes of wine and cans of beer], you see stacks of dollar bills - hundreds of thousands and millions of freshly minted, crisp, and crinkly dollar bills just sitting there on pallets, piled high to the ceiling. That’s exactly how you should think of inventory.” If the amount of money referenced seems far-fetched, consider that an LRS operating on the scale of a BCLDB store will typically carry over $500,000 worth of inventory, and several are pushing the million dollar threshold.
The answer to the second questions is “it depends”, and the discussion from this point on will focus on those determining factors.

(Note that in the material that follows there are references to a variety of calculations used in the process of effective inventory management. The retailer shouldn’t need to perform these calculations since the point-of-sale system should provide this capability. It is strongly recommended that you look for these capabilities when evaluating POS applications.)

Retail Model
An important first step to take is to classify the scope and scale of your retailing effort. Specifically, are you intending to operate (and compete) as a full service outlet offering a broad array of items in all categories, or as a convenience service focused on core categories and brands? The reason this distinction is critical is that it affects the amount of flexibility you have in meeting service level targets. A full service retailer can afford to run out of secondary items on occasion because there are likely to be a reasonable number of substitutes for any given product. Items in a convenience operation are by definition primary or core type products; the notions of “convenience” and “stock-outs” are contradictory so the margin for error is much smaller.

Figure 1 illustrates this principle (it is an example taken from an actual BCLDB store). Each symbol on the chart represents an Australian wine SKU and its contribution to overall Australian sales in the store. You can see the first SKU contributes 10% of the total sales, the first 10 SKUs account for 35% of total sales, the first 20 SKUs account for 50% of total sales, etc. It should be evident that if a full service retailer were to stock out of one of the items in the yellow shaded area, there are plenty of alternatives in the blue area. On the other hand, a convenience retailer is likely to only carry a portion of the items in the green area, which dramatically reduces the odds of being able to provide a direct substitute for a stocked-out item.

SKU Prioritization and Safety Stock
What should also be clear from Figure 1 is that each SKU varies in importance based on its contribution. The items in the green area are considerably more productive than the other two segments, meaning the consequences of stock-outs are more significant as well. Stocking-out of any of the top 20 items could have serious impact on store revenue in a full-service environment whereas running out of any of the last 20 would have virtually no impact at all. For this reason many retailers “stratify” their assortments based on contribution and assign safety stock targets accordingly. Continuing with the example above, the retailer might classify the green shaded area as Strata A, the yellow as Strata B, and the blue as Strata C. The safety stock target for Strata A might be set at three weeks of average demand, Strata B at two weeks, and Strata C at one week (note these numbers are intended to demonstrate the principle of stratification - there are additional factors to consider as well). By recognizing the low risk attached to a stock-out in Strata C, the retailer can save a considerable amount of money in inventory costs. Conversely, the higher investment in Strata A serves to ensure all potential sales are realized.

Variability of Demand and Safety Stock
For many items the past is a very useful indicator of what will happen in the future, and for that reason past history is generally used to predict future demand. The history usually covers several demand/replenishment cycles in an effort to identify each item’s trend. The trend gives the retailer the ability to estimate the likely demand in a given period without having to do an indepth analysis of each item. It is important to realize, however, that two items that have apparently similar demand levels over time can behave quite differently within specific periods. Figure 2 shows two items with an average of 15 unit sales per week over an 8-week period. The demand within a given week is quite different, though; one item varies little from one week to the next while the other fluctuates significantly.

If the retailer were to take any 3-week period for Brand X to predict future demand the result would be very accurate because a typical week is very similar to any given 3-week average. The same would not be true of Brand Y: using, for example, weeks 5, 6, and 7 as the average the retailer would not come close to predicting what actually occurred in week 8. Without taking into account the demand variability for Brand Y stock-outs could easily occur in weeks 2 and 7.

Figure 3 shows the extent to which these two items vary from the average 8-week demand in any given week. Note that it is not important whether the variation is a negative or positive, it’s the amount of the variation that matters (many large retailers carrying tens of thousands of SKUs use sophisticated statistical models to account for demand variability).

The conclusion to draw from this is that items with highly predictable demand patterns require proportionately less safety stock than items with patterns that are more volatile. The value of knowing which items fall into which category is that the lower safety stock associated with predictable items means less money tied up in store inventory.

Figure 4 shows how these factors work together. The assumptions built into the calculations are: you need to replace what you predict you’re going to sell, you need more safety stock for A items than Bs and Cs (based on the stratification discussion ealier), and you need less safety stock for low variability items.

In Figure 4, an A item order would be generated whenever the on-hand inventory fell one case below the target inventory. In the case of the B and C items, where the inventory targets are less than one case, replenishment would be triggered by establishing minimum order levels e.g. a case or part case would be ordered if the on-hand inventory dropped below a specified minimum (such as 3-6 bottles, depending on demand variability). Again it should be noted that these numbers are for demonstration purposes only and more detailed calculations would be used in any moderately sophisticated POS system.

Demand versus Sales
A common mistake made in managing inventory is to equate sales with demand. Sales represent the quantities that customers were actually able to purchase. Demand reflects the number of sales that would have occurred if every attempt to purchase an item was successful. Using the A Strata item Figure 4, imagine that you run out of stock on Friday and that typically 50% of weekly sales occur on Saturday and Sunday. If you replenish strictly by looking at sales you would be basing future orders on the 12 units you sold rather than on the 24 units you would have sold if the stock-out had not occurred. The result is you will continue to stock-out each week until some kind of adjustment is made to factor in the lost sales.

Retailers use a number of ways to account for lost sales. Some simply assume that a certain percentage of sales is lost due to stock-outs and use a factor to boost the sales number to approximate demand. Thus sales might be increased by a factor of 10% across the board to ensure some consideration is given to lost sales. A slightly more sophisticated process would be to create factors based on demand variability since items that are unpredictable have a correspondingly higher risk of stocking-out. The best method, however, would be to have (or invest in) a system that actually measures lost sales. Using our original example above, our POS system should “know” that 50% of sales occur on Saturday and Sunday, it should “know” that the inventory level of the item reached zero on Friday, and it should “know” that 12 additional sales would have occurred during the time the inventory level was at zero. It should then be able to factor those lost sales along with the actual sales into the upcoming forecast.

Seasonality 
The final major factor that needs to be considered is seasonality. Not only does the time of year impact overall sales levels, it affects different categories to varying degrees. For example, categories like gin, beer, and refreshment beverages obviously have their peak seasons during the warmer summer months, whereas categories like liqueurs and rum do the highest percentage of their volume during the winter, especially December.

Figure 5 shows the seasonal profile of the table wine category and illustrates that the seasonal skew for reds and whites is somewhat different. There is a stronger bias toward white wines during spring and summer, and towards red wines in fall and winter. Note, however, that although red is the dominant segment in the month of December, both red and white experience their annual demand peak in that month. This is reinforced in Figure 6, which illustrates the extent to which each month varies from the average.

The value of understanding seasonality is it indicates the direction and extent of the adjustments to forecasts that are needed from one month to the next. For example, if you used November to predict December you would not have nearly enough stock, whereas using December to predict January would result in the opposite problem.

Summary
Effectively managing inventory requires an understanding of the variables that can contribute to having too much or too little inventory. Either scenario has important financial implications as too much inventory ties up capital that can be more productively used elsewhere, and too little will result in lower revenue from lost sales. The critical factors to examine are how important is the item, how easy/difficult is it to predict, how closely do “sales” reflect “demand”, and what is the seasonal profile? The extent to which you can measure each variable will determine the degree to which inventory investment is optimized. Most POS systems contain all the data needed to perform these functions. In fact any computerized process should be able to perform much more complex and sophisticated calculations than those used in the examples provided here. Retailers need to impress upon their POS providers the importance of having access to this kind of functionality. Just keep imagining those pallets stacked with crisp dollar bills.

Brock Davison is the Category Manager and Business Analyst for Western Canada at Andrew Peller Limited, the country’s largest Canadian-owned wine company.