http://archive.ite.journal.informs.org/Vol5No2/Miyaoka/
http://wordpress.shetgar.com/blog/?p=59
http://www.scribd.com/doc/51139499/Littlefield-Simulation-2-Report
Here is what we did:
Pre-Game Activities: The team met the Tuesday before class to examine the data and discuss strategies. It was apparent that both Stations 1 and 3 were operating at full capacity, frequently hitting 100% utilization. Station 3 seemed more strained since it had higher queues (Mean=506, STD=498) than Station 1(Mean=187, STD=175).
Since the average job lead time exceeded 2 days during days 43 through 46, inclusive, we thought it would be unprofitable to attempt to move to the $1,000 contracts. We discussed the options of altering the lot sizes, but decided that the extra setup time would only create more bottlenecks downstream.
Stage 1: As a result of our analysis, the team’s initial actions included:
1. Leave the contracts at $750.
2. Change the reorder point to 3000 (possibly risking running out of stock).
3. Change the reorder quantity to 3600 kits.
4. Purchase a second machine for Station 3 as soon as our cash balance reached $137,000 ($100K + 37K).
This strategy proved successful and after the second machine for Station 3 was purchased on Day 56 and the queue cleared, we were able to switch to the $1,000 contracts. We occasionally lost a few dollars for being a little late, but we always made more than we would have under the $750 contracts.
Stage 2: The next goal was to save enough cash to purchase a machine for Station 1 so that we could switch to the $1,250 contracts. During the cash building stage, we made the inventory order quantity as high as we could afford, which was 6,900 kits at a purchase price of $70,000. When the 6,900 kits were delivered, we switched the order quantity back to 3,600 so that we could purchase a Station 1 machine as soon as our cash balance reached $127,000 ($90K + 37K).
After 21 factory days, we were able to purchase the fourth machine for Station 1 and immediately moved to the $1,250 contracts.
The average lead time declined to under a half a day during factory days 69 through 76. There was a substantial decline in arriving orders during the same time period. The team noticed the drop in lead time and regrets not having moved to the $1,250 contracts sooner. We lost $22,750 of potential revenue for not moving on the information sooner. On the other hand, orders are random and an early move could have backfired on us.
Stage 3: During our preliminary meeting, the team discussed the possibility of purchasing a fifth machine for Station 1. We decided to wait and see if the loss of potential earnings was sufficient to justify a $90 K purchase. We knew that if we were going to buy a fifth machine we should do it as soon as possible to maximize the return on investment. We calculated the loss of potential revenue as ($1,250 – actual average revenues * jobs completed). Our initial estimates showed a potential revenue loss of $266 per day, but within a few factory days the rate of potential loss rose to $419 per day.
There is another consideration in the decision to purchase a fifth machine for Station 1. The title of the Littlefield Technologies game 2 is Customer Responsiveness. The title implies that we should be concerned about the consistency with which we deliver on our service level agreements (SLAs). The potential loss of $419 per day barely covers the $90,000 machine purchase; however we were missing our SLAs 13 out of 15 days and the percent of potential revenues lost due to missing SLAs was 3%. We decided to purchase the fifth machine on Day 94 primarily to improve our customer responsiveness.
This strategy did not perform as well as we had hoped. While our potential revenues lost declined to 1%, we were still missing our SLAs six out of seven days.
Stage 4. During Stage 4, we explored job splitting as a solution to the SLA problem. First, we split jobs into two batch of 30 kits each. This strategy worked so well that we wondered why we hadn’t explored job splitting during Stage 2 or 3. We met our SLAs 12 out of 16 days and our percent of potential revenues lost declined to 0.4%. We calculated the setup times as a proportion of a machine to be 0.007, 0.003, and 0.002 for S1, S2, and S3, respectively.
2S1 + P = 0.194458 => S1 + P = 0.187256 => S1 = 0.007202
2S2 + P = 0.082479 => S2 + P = 0.079424 => S2 = 0.003055
2S3 + P = 0.064835 => S3 + P = 0.062434 => S3 = 0.002401
Where the right hand side is calculated as Sum(%Utilization * #Machines)/#Jobs Completed
We thought that if setup time was so insignificant, maybe the other job splits would be equally good or better. Accordingly, we tried the 3-way job split for eight days, but we were not impressed with the results. On one of the days, our average revenues dropped below $1,200, which we hadn’t seen since purchasing the fifth machine for Station 1.
We thought that maybe it was because of the mismatch between machines and splits. So we tried the 5-way split thinking each job would be split equally among the five machines. This turned out to be a HUGE mistake! After only one factory day it was apparent the 5-way split was a bad thing and we switched back to the 2-way split. Even so, it took an additional four days for the system to recover from the backlogs and we lost $46,693 in potential revenues. (Morale of the story – 2 way splits are great as soon as the queue clears with the purchase of machines. Forget the other splits.)
A one-way ANOVA demonstrated that the differences between the job splits were statistically significant at the alpha=.01 level. Group 1 was no splits. Group 2 was a two-way split. Group 3 was a three-way split. Group 5 was a five-way split. Data for all groups were collected after all machine purchases explaining the small number of observations for Group 1.
We chose to stay with the 2-way split not only because it had the highest average revenues, but also because the 2-way split had the lowest variance. With the 2-way split we were meeting our service level agreements more consistently resulting in higher customer satisfaction and higher profits per job.
Stage 5. With our factory humming, our attention turned to inventory purchases. We calculated the reorder quantity using the equation:
Q* = SQRT(2DS/H) = SQRT(2 * 12 * 365 * 1,000 / 66.31) = 363 batches
Where D = annual demand = 12 * 365
S = fixed cost per order = $1,000, and
H = the handling costs = $60 x (1 + .10/365)365 = 66.31
The calculated reorder quantity was surprisingly close to the value obtained from running our numbers through the Inventory example from Chapter 7 of our text (363 vs. 382).
The text also mentioned that small variations in reorder quantity do not matter much and so people usually round to a convenient number. Thus, we set our re-order quantity to 400.
Stage 6. Previously we had been stockpiling inventory by purchasing more as soon as money was available to purchase, but we realized that we may be missing out on nontrivial interest payments. So we re-set the reorder point to 3600, which provides a four day inventory plus a safety net.
Stage 7. The Exit Strategy – We do not have control of the factory during the last 100 days of its life. We know from the instructions for the game that the demand is expected to stay consistent although orders are random. We do not feel it is wise to leave a large reorder quantity while the factory is out of our control because we might have a sudden increase in jobs during the last few days that sparks a $241,000 inventory purchase, most of which will go to waste. So before we lose control, we will buy (100 * 11.8 * 60) kits and then set the reorder quantity to 60 (or 3,600 kits). We hope this exit strategy works.
The exit strategy did work although if we had purchased another 1,200 kits in Stage 7, we could have set the reorder quantity to 0 and reorder point to 0. This would have saved use another $24,000.
