Recently I received a call from a distributor who had a dilemma. They were running out of a particular popular product every month. The buyer’s frustration vibrated through the phone line as I spoke to her.
She explained, “We’re doing a pretty good job at estimating future demand of the product. The average difference between the forecast demand for the month and actual usage is less than 10%.”
I explored another area. “How consistent are the lead times? Could shipment delays be causing the stock-out problems?”
“No,” she sighed, “the vendor always delivers four business days after we order the product.”
“How many customers do you have for the item?”
“We have several large customers who place a large order for the item each month, and several smaller customers who pick up a few pieces when they need them.”
“When do you receive the large-customer orders?”
“Usually in the first 10 days of the month.”
The buyer had just uncovered the problem. Her replenishment system forecast demand of future usage by month. For example, in September it predicted sales of 550 pieces or about 25 pieces per business day (assuming 22 business days in the month). Actual usage was 539 pieces. Her system automatically calculated the following order point for the item:
Anticipated Lead Time Demand = 25 pieces/day x 4-day lead time = 100 pieces
Safety Stock = 50% of lead-time usage = 50 pieces
Order Point = 100 pieces + 50 pieces = 150 pieces
The problem is that the distributor does not sell 25 pieces per day throughout the month. Here is the weekly usage recorded for the September:
| Week | Business Days | Weekly Usage | Usage/Business Day |
| 1 | 4 | 324 | 81.0 |
| 2 | 5 | 132 | 26.4 |
| 3 | 5 | 40 | 8.0 |
| 4 | 5 | 33 | 6.6 |
| 5 | 1 | 10 | 10.0 |
The majority of the demand for the product (324 pieces, or 60.1% of total monthly usage) occurs in the first week of the month. The order point of 150 pieces represents less than a two-day supply in the busy first week of the month. If we reorder a product when there is a two-day supply on the shelf and it takes four days to receive a replenishment shipment, it is not surprising that the item experiences regular stock-outs.
We recommended that the company change the time period of their forecast from months to weeks. Analyzing history over the past 12 months, we found that a formula that averaged usage for the same week in each month over the past four months resulted in the least forecast error. We went back and reforecast the five weeks of September. Note that the fifth week included one business day for September and four business days for October (i.e. the start of the next high volume time period):
| Week | Weekly Forecast | Business Days | Forecast/Day |
| Week 1 (Sept) | 331 pieces | 4 | 82.8/day |
| Week 2 (Sept) | 135 pieces | 5 | 27.0/day |
| Week 3 (Sept) | 37 pieces | 5 | 7.4/day |
| Week 4 (Sept) | 47 pieces | 5 | 9.4/day |
| Week 5 (Sept) + Week 1 (Oct) | 336 pieces | 1 + 4 | 67.2/day |
Each week would have had its own order point:
| Week | Lead-Time Usage | Safety Stock | Order Point |
| Week 1 | 82.8/day x 4 days = 331 | 166 | 497 |
| Week 2 | 27.0/day x 4 days = 108 | 54 | 162 |
| Week 3 | 7.4/day x 4 days = 30 | 15 | 45 |
| Week 4 | 9.4/day x 4 days = 38 | 19 | 57 |
| Week 5 | 67.2/day x 5 days = 336 | 168 | 504 |
Each new order point becomes effective at a date equal to the first business day of the week minus the lead time. So, four business days before the start of week one, if the stock level of the product was less than 497 pieces, a replenishment order would be issued. If the stock level was equal to or greater than 497 pieces, the buyer would leave the item alone because he has plenty of stock available to see him through the first week of the month. The result: The distributor will have this critical item available when their customers want it. Also notice that the order point drops during the period of the month with less usage activity. We are not ordering the material far in advance of when it will be needed. This will help improve the inventory turnover and overall profitability of the distributor.
Weekly forecasting works well for products that experience cyclical patterns of usage throughout a month. Next month, we will examine two additional applications for weekly forecasting: new stock items and products whose sales or usage is dependent on a particular event.