Você está na página 1de 4

MaNagiNg FoRecaStiNg aND iNveNtoRy with LittLe iNFoRMatioN: a caSe StuDy

by Robert J. Stawicki

Discusses how to make a decision about purchasing and inventory with little information often a large percentage of sales come from few products products with excess inventory dont require forecasts for immediate ordering.
t times, a decision has to be made in business even when we dont have all the information. This paper discusses one of those cases here. This is a situation where forecasts need improvement, but there is no time to work on them. The company has to make purchasing decisions to satisfy customer demand; otherwise, it runs the risk of being delisted.

this company) relied upon contract manufacturers, many of whom were offshore. Since all the competitors in the industry also placed orders at the same time, the contract manufacturers required four to six month lead times, which is considerably longer than the endcustomers lead time. These manufacturing lead times also included shipping times from the offshore manufacturing locations. Since the manufacturing lead times were greater than the customerorder lead times, orders had to be placed based on forecasts rather than actual orders. Domestic manufacturers were available with shorter lead times, but they had higher manufacturing and setup costs. There also were local job shops with no

setup costs, but their cost per unit was extremely high. Due to their higher costs, both of these domestic sources were only used for fulfillment of the last minute orders that exceeded the available supply. Product life spans were approximately seven years but the shapes of the life cycle curves varied significantly for different products. Some product lines had large sales in the first two years after the introduction of a new or revised product, with only small replacement volumes in the remaining years. The sales volumes for other products grew and shrank slowly over their life cycle.

the coMPaNy eNviRoNMeNt


The company had little information technology infrastructure. Not only that, it did not have a formal statistical forecasting package. Instead, all forecasting was done on spreadsheets or on PC-based database software. The manpower allocated to forecasting totaled six full-time equivalents who developed forecasts manually by analyzing the sales histories of current and past history of each product. Sales history included mostly orders of kits; each kit included different SKUs, making it difficult to get a count of each SKU ordered. Further, sales and marketing personnel provided some input, though at a high level of product/market hierarchy and in anecdotal format such as Customer A is going to buy $50,000 worth of a product or Customer B is going to buy from product line XYZ. Such input had to be manually disaggregated to obtain forecasts for individual SKUs.

the buSiNeSS eNviRoNMeNt


The company, the case in point, had hundreds of customers but most of the demand came from a few large accounts. Greater than 90% of the annual demand for all products from all customers was ordered within a narrow two-month window with approximately two months of lead time. All orders had to be completely filled or the company risked being permanently removed from the approved vendor list. Some small orders continued to trickle in during the rest of the year and those orders also had to be fulfilled quickly from inventory, again for the same reason. Because all of the annual demand occurred within a relatively short time, most suppliers in the industry (including

RobeRt J. Stawicki Dr. Robert Stawicki is an independent consultant who specializes in Demand Planning, Supply chain Network optimization, and Production Sched uling. over the past 30 years he has implemented supply chain solutions for numerous Fortune 500 corporations.

34

THE JOURNAL OF BUSINESS FORECASTING, SPRING 2010

The company preferred to place only one order to a manufacturer for a SKU per year. Its logic was that since customers order only once a year (except for some small replacement sales within a year), it should only order once a year. But our study determined another reason for doing that. Setup charges of suppliers were a large portion of the total order cost; literally thousands of times greater than per unit cost (see Table 1). For an example, let us look at SKU 5: For a production run of 5,859 pieces with per unit costs of $4.67, the setup charge was $29,555. But the setup charges had no visibility within the company. Although suppliers provided quotes with setup and per unit costs, the company had no IT system to store setup charges; furthermore, no one was responsible for tracking that. The supplier quotes were converted into per unit costs for a specific order quantity, and the converted per unit cost was stored in the purchase order system. For example, the production run of 5,859 pieces in our example was entered into the purchase order system with a unit cost of $9.71. While the goal was to have only one production run per product per year, more than 40% of the products were ordered more than once, and 10% of them were ordered three or four times.

within that time span meant that the firm required about half a month to just acquire the sales data for the past two years, and then convert and sort the information by SKU. This was problematic because company staff had little time to devote to data gathering and conversion while in the midst of preparing for the annual buying season. Second, it required some form of statistical life cycle forecasting, which meant that the planners had to manually determine where each SKU was in its life cycle and link it to an analog or a complementary product at a similar point in its life cycle phase. Only the individual planners knew which SKU was linked to which SKU. However, at this point of the year, they did not have the time to perform these tasks. Sales and forecast data were only available for two years. Since this study focused on providing one overall forecast for the year for each SKU, aggregating the sales for each SKU for each year resulted in only two data points for each SKU. Because of these obstacles, the objective of the project was changed to find a tool that can provide guidance on order quantities for individual SKUs, utilizing the limited information available at the time.

tabLe 1 coSt StRuctuRe


Skus SKU-1 SKU-2 SKU-3 SKU-4 SKU-5 SKU-6 SKU-7 SKU-8 SKU-9 SKU-10 SKU-11 Forecast volume 349 1,381 3,092 4,584 5,859 9,147 13,165 20,848 25,370 38,562 64,671 unit cost $4.01 $5.01 $0.37 $0.46 $4.67 $4.07 $6.25 $5.12 $3.03 $4.06 $0.36 Setup cost $38,294 $41,428 $1,344 $1,139 $29,555 $42,827 $50,890 $33,202 $35,105 $35,095 $1,112

position of those SKUs short of write offs, which, of course, was not within the scope of this project. The current inventory positions of the A items were compared to the current forecast. In a number of cases, the current inventory was greater than the current forecast so there was no immediate need to improve the forecast for these items. Forecasts for these items could be reviewed at a future date in the event that actual orders exceeded the inventory available. This reduced the set of remaining SKUs to approximately 350. With that, what had seemed to be a gargantuan task (30,000 SKUs) became much more manageable. To determine how well forecasts were prepared of the remaining 350 A products, we examined their forecasts and actual sales of the previous year. We found that, on a volume basis, under-forecasted SKUs exceeded over-forecasted ones in excess of 2.5:1; whereas, on a SKU basis the over-forecasted ones were exceeded by 1.5:1. Our conclusion was that a few high volume items were significantly under forecasted, for which the company had to incur setup charges again for the second, third, or more production runs. Next, we calculated the percentage of actual sales to the forecast for each SKU. If, for example, actual is 100 and forecast

aPPRoach uSeD
To get a handle on this project, we first collected the data of actual sales, forecast sales, and current inventories. Then, we converted all the sales data into individual SKUs. In other words, we converted the data at the kit level into individual SKUs. We interviewed purchasing staff to identify setup costs of various SKUs, which was a major issue. Once data were collected and converted into SKUs, we conducted an ABC analysis of current inventories, which revealed that less than 800 SKUs constituted 80% of the sales. Additionally, a quick review of the inventory, recent sales history, and current forecast for the remaining items showed that many of them had very little recent or expected sales history. There was very little we could do to change the inventory

obJectiveS oF the PRoJect aND coNStRaiNtS


The company had us start our project approximately one month (20 working days) prior to the date it needed to start placing orders with the manufacturers. The clients goals were two-fold: One, improve forecast accuracy; and two, by using improved forecasts, reduce inventory by the end of the annual selling cycle. It became apparent in the first couple of days that the goals of this project would not be met within the time period allotted for two reasons: First, the lack of IT infrastructure to collect the required data

THE JOURNAL OF BUSINESS FORECASTING, SPRING 2010

35

Number of SKUs

Number of SKUs

randomize d data 1 2 5 1 3 7 7 7 16 6 5 1 5 4 2 1 2 1 2 1 1 6

FiguRe 1 gRaPhicaL RePReSeNtatioN oF FoRecaSt accuRacy


Product Line a
20 20 Number of SKUs Number of SKUs 16 12 8 4 0 0% 100% 200% 300% 16 12 8 4 0 0% 100% 200% 300%

Product Line b

Percentage of Sales to Forecast Product Line c


20 16 12 8 4 0 0% 100% 200% 20 16 12 8 4 0 0%

Percentage of Sales to Forecast all other Products

1 1 1 2 12 17 6 12 5 3 4 1 2 1 1

100%

200%

300%

Percentage of Sales to Forecast

Percentage of Sales to Forecast

3 1 1 4 1 15 2 11 5 11 5 2 1 2 4 1 4 3 1 13

was 90, then the actual as a percentage of forecast is 111%. If the percentage is more than 100%, it means that we under forecasted; if less, we over forecasted. After that, we broke down SKUs on the basis of product classification (that is, A, B, C, and Others) as well as on the basis of the percentage difference between the actual and the forecast. For each product line we used the increment of 10 percentage points, that is to say, how many SKUs we had in product line A where the difference between the actual and the forecast was between 5 and 15 percentage points, how many SKUs were where the difference between was 15 and 25 percentage points, and so on. Plots of all the four classifications are given in Figure 1, which shows, for example, in Product line B, there are 15 SKUs where the actual sales were somewhere between 85% and 95% of forecast sales, while only two products had actual sales between 95% and 105% of forecast sales. Tick marks on the figures represent the midpoint of each 10% range.

The first insight the plots provided was that the forecasts were not very accurate and forecast error did not seem to fall into any known distribution. Also, in most cases, products were under-forecasted because, in most cases, their percentage number was higher than 100.

RecoMMeNDatioNS
Based on our research, we made the following recommendations: Where the inventory of A items is greater than the forecast, there is no need to do anything at this point. How to improve their forecasts would be investigated later. Concentrate on improving forecasts of the high volume items that were significantly under forecasted last year. Because of time constraints to improve statistical forecasts, we recommended to change order quantities somewhat by balancing

the effect of carrying costs and setup costs. This resulted in both increases and decreases in the recommended order quantity. We estimated, by doing that, the company will save $1.5 million in setup and inventory carrying costs. This was calculated by comparing the expected cost of their existing order quantities to the expected cost of the recommended order quantities. The company must store in the purchase order system not only the cost per unit quoted by the suppliers but also the setup charges. Additionally, one member of the companys staff must be held accountable for tracking setup costs. To get a good handle on inventory, the company must keep a record of each SKU along with kits.

(The author wishes to thank Dr. Don Shobrys for his assistance and encouragement in preparing this article). (info@ibf.org)

36

THE JOURNAL OF BUSINESS FORECASTING, SPRING 2010

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Você também pode gostar